WO2005020798A2 - Multi-domain motion estimation and plethysmographic recognition using fuzzy neural-nets - Google Patents

Multi-domain motion estimation and plethysmographic recognition using fuzzy neural-nets Download PDF

Info

Publication number
WO2005020798A2
WO2005020798A2 PCT/US2004/027973 US2004027973W WO2005020798A2 WO 2005020798 A2 WO2005020798 A2 WO 2005020798A2 US 2004027973 W US2004027973 W US 2004027973W WO 2005020798 A2 WO2005020798 A2 WO 2005020798A2
Authority
WO
WIPO (PCT)
Prior art keywords
signal
plethysmographic
plethysmographic signal
domain
neural network
Prior art date
Application number
PCT/US2004/027973
Other languages
French (fr)
Other versions
WO2005020798A3 (en
Inventor
Alvin Mark Terry
Original Assignee
Datex-Ohmeda, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Datex-Ohmeda, Inc. filed Critical Datex-Ohmeda, Inc.
Publication of WO2005020798A2 publication Critical patent/WO2005020798A2/en
Publication of WO2005020798A3 publication Critical patent/WO2005020798A3/en

Links

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A61B5/7207Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal of noise induced by motion artifacts
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/145Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue
    • A61B5/1455Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue using optical sensors, e.g. spectral photometrical oximeters
    • A61B5/14551Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue using optical sensors, e.g. spectral photometrical oximeters for measuring blood gases
    • A61B5/14552Details of sensors specially adapted therefor
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7264Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7253Details of waveform analysis characterised by using transforms
    • A61B5/7257Details of waveform analysis characterised by using transforms using Fourier transforms
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7253Details of waveform analysis characterised by using transforms
    • A61B5/726Details of waveform analysis characterised by using transforms using Wavelet transforms
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10TECHNICAL SUBJECTS COVERED BY FORMER USPC
    • Y10STECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10S128/00Surgery
    • Y10S128/92Computer assisted medical diagnostics
    • Y10S128/925Neural network

Definitions

  • the present invention relates generally to plethysmographic signal processing, and more particularly to the application of neural net processes to the classification of plethysmographic signals and the estimation of motion therein.
  • Signal attenuation measurements generally involve transmitting a signal towards or through a tissue medium under analysis, detecting the signal transmitted through or reflected by the medium and computing a parameter value for the medium based on attenuation of the signal by the medium.
  • simultaneous signal attenuation measurement systems multiple signals are simultaneously transmitted (i.e., two or more signals are transmitted during at least one measurement interval) to the medium and detected in order to obtain information regarding the medium.
  • Such attenuation measurement systems are used in various applications in various industries.
  • optical signals are utilized to monitor the composition of respiratory and anesthetic gases, and to analyze tissue or a blood sample with regard to oxygen saturation (Sp02 level), analyte values (e.g., related to certain hemoglobins) or other composition related values.
  • oxygen saturation Sp02 level
  • analyte values e.g., related to certain hemoglobins
  • the case of pulse oximetry is illustrative. Some pulse oximeters extract information regarding patient physiological conditions such as the patient's pulse rate and an oxygen saturation level of the patient's blood, or related analyte values, via analysis of plethysmographic signals or waveforms corresponding to different wavelengths of light transmitted through or reflected from the patient's tissue.
  • pulse oximeters generally include a probe for attaching to a patient's tissue site such as a finger, earlobe, nasal septum, or foot.
  • the probe is used to transmit pulsed optical signals of at least two wavelengths, typically red and infrared, to the patient's tissue site.
  • the different wavelengths of light used are often referred to as the channels of the pulse oximeter (e.g., the red and infrared channels).
  • the optical signals are attenuated by the patient tissue site and subsequently are received by a detector that provides an analog electrical output signal representative of the received optical signals.
  • the attenuated optical signals as received by the detector are often referred to as the transmitted signals.
  • the electrical signal can be processed to obtain plethysmographic signals for each channel and the plethysmographic signals may be analyzed to obtain information regarding patient physiological conditions. Extraction of patient physiological conditions from the plethysmographic signals can be quite effective using a well positioned sensor and when the patient or subject is resting. However motion artifacts can easily swamp the desired information included in the plethysmographic signals when the patient is moving around and/or performing muscular contractions. Some motion artifacts can severely impair the signals, whereas other types can be filtered out or do not significantly effect the desired information included in the plethysmographic signals.
  • the present invention is directed to the use of multiple signal domains in conjunction with neural net processing to achieve pattern classification of different types of motion artifacts and also to classify different patterns of pulse and plethysmographic waveforms that can occur under different physiological conditions.
  • the implementation of a neural-net pattern recognition stage within a pulse oximeter addresses a number of the aforementioned problems thereby achieving an improved instrument.
  • motion classification is performed using features extracted from the time, spectral (e.g. power spectral), bispectral and cepstral domains. These features and the spectral (e.g.
  • a neural network also referred to herein as a neural-net
  • a neural network also referred to herein as a neural-net
  • Another form of information that can be used is coefficients derived from a Wavelet filter bank.
  • the information from the different sources described above is weighted and processed via a neural-net and the output classification of the neural-net is then used to help with filtering out of motion artifacts and in the decision to use selective portions of high- pass frequency information (AC) in Sp02 calculations.
  • AC high- pass frequency information
  • the motion classification can be used to improve the signal extraction in noisy conditions, and to aid in selection of the most appropriate signal extraction method.
  • the neural-net classifier can be used to determine the type of plethysmographic signal, such as normal adult, infant, or that produced from different types of heart conditions. Identification of the signal type will aid in measurement and in extraction of the signal from noise.
  • One aspect of the invention is in describing the different types of physiological and noise conditions for which plethysmographic waveforms are collected for training the neural-net classifier. In order to improve the robustness of the classifier the inputs to the neural-net during training are subjected to 'fuzzification' wherein input values are slightly perturbed and reordered.
  • This process increases the size of the training set and also has the effect of making the neural-net less rigid in its ability to classify patterns.
  • the training of the neural-net can also be enhanced by use of a fuzzy logic controller which is used to adjust learning parameters and to speed up convergence during back-propagation learning.
  • Another aspect of the invention is that the neural-net is trained to respond to different respiration activities - both rate and depth. Pertinent information from spectral coefficients, and Wavelet transforms on the different input wavelengths (typically red and infrared, but also additional wavelengths which may be used) are presented to the neural- net. Optionally derived Sp02 estimates can also be presented. All this information can be used to obtain further more robust Sp02 estimates.
  • a pulse oximeter includes a first optical signal source that is operable to emit an optical signal characterized by a first wavelength (e.g., red) and a second optical signal source that is operable to emit an optical signal characterized by a second wavelength (e.g., infrared) different than the first wavelength.
  • a detector operable to receive the first and second optical signals after they are attenuated by a patient tissue site provides a detector output signal representative of the attenuated first and second optical signals.
  • a processor is enabled to obtain first and second time domain plethysmographic signals from the detector output signal.
  • the processor also classifies one or more of the first and second time domain plethysmographic signals using a neural network.
  • the neural network receives input coefficients derived from one or more transforms of the first and/or second time domain plethysmographic signals.
  • the transforms may, for example, include spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filter bank transforms.
  • a method of processing a plethysmographic signal obtained from a patient in a first signal domain includes transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain.
  • the different signal domains may, for example, include spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filtered domains. Transformation of the first domain plethysmographic signal results in a plurality of transformed plethysmographic signals with each transformed plethysmographic signal being in one of the different signal domains. A plurality of sets of coefficients are selected, with each set of coefficients being derived from a corresponding one of the transformed plethysmographic signals. The sets of coefficients are input to a neural network, and the plethysmographic signal is classified based on an output from the neural network.
  • spectral e.g., power spectral
  • bispectral e.g., bispectral
  • cepstral e.g., cepstral
  • Wavelet filtered domains e.g., Wavelet filtered domains. Transformation of the first domain plethysmographic signal results in a plurality of transformed ple
  • a method of training a neural network to classify a plethysmographic signal obtained from a patient includes selecting a plurality of first domain plethysmographic signal data sets associated with a plurality of different types of predetermined signal conditions from a database of plethysmographic signal data sets.
  • the first domain plethysmographic signal data sets are transformed to other signal domains different than the first domain to obtain a corresponding plurality of transformed plethysmographic signal data sets.
  • the first domain may, for example, be the time domain and the other signal domains include, for example, spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filtered domains.
  • a plurality of sets of coefficients are extracted from the transformed plethysmographic signal data sets, with each set of coefficients being extracted from a corresponding one of the transformed plethysmographic signal data sets.
  • the sets of extracted coefficients are used as inputs to the neural network, and weighting values associated with connections between neurons in the neural network are adjusted in accordance with a learning procedure.
  • the learning procedure may, for example, be a backpropagation learning procedure or a simulated annealing learning procedure. Where desired, the backpropagation learning procedure may be implemented with fuzzy logic control.
  • a method of providing information relating to a physiological condition of a patient based on at least one plethysmographic signal obtained from the patient in a first signal domain includes transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain.
  • the different signal domains may, for example, include spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filtered domains. Transformation of the first domain plethysmographic signal results in a plurality of transformed plethysmographic signals with each transformed plethysmographic signal being in one of the different signal domains.
  • the plethysmographic signal is classified based on an output from a neural network.
  • the output of the neural network is based on input coefficients derived from one or more of the transformed plethysmographic signals.
  • a technique for determining the physiological condition of the patient is selected.
  • the physiological condition of the patient may, for example, comprise a pulse rate.
  • the physiological condition of the patient may, for example, comprise an Sp02 value or a respiration index.
  • the plethysmographic signals are preferably transformed using at least a Wavelet filter bank transform.
  • FIG. 1 is a block diagram of one embodiment of a pulse oximeter employing multi-domain motion estimation and plethysmographic signal recognition using fuzzy neural-nets in accordance with the present invention
  • FIG. 2 is a block diagram showing one embodiment of a method of implementing multi-domain motion estimation and plethysmographic signal recognition using fuzzy neural-nets in accordance with the present invention
  • FIG. 3 is a schematic diagram illustrating one embodiment of a neural network architecture that may be employed in accordance with the present invention.
  • FIG. 4 is a plot of exemplary red and infrared plethysmographic signals that include motion artifacts
  • FIG. 5 is a block diagram showing one embodiment of a neural network training method in accordance with the present invention
  • FIGS. 6A-6D are plots showing exemplary plethysmographic signal conditions that may be included within a database of plethysmographic training sets.
  • the pulse oximeter 10 is configured for use in determining the pulse rate of a patient as well as one or more blood analyte levels in the patient, such as an SP02 level. It should be appreciated that multi- domain motion estimation and plethysmographic signal recognition using fuzzy neural- nets in accordance with the present invention may be implemented in pulse oximeters that are configured differently from the pulse oximeter depicted in FIG.
  • the pulse oximeter 10 includes a pair of optical signal sources 20a, 20b for emitting a corresponding pair of light signals 30a, 30b centered at different predetermined center wavelengths ⁇ ⁇ , ⁇ 2 through a suitable tissue site of a patient and on to a detector 40 (e.g., a photo-sensitive diode).
  • the optical signal sources 20a, 20b and detector 40 may be included in a positioning device 50, or probe, to facilitate alignment of the light signals 30a, 30b with the detector 40.
  • the positioning device 50 may be of clip-type or flexible strip configuration adapted for selective attachment to a suitable patient tissue site (e.g., a finger, an ear lobe, a foot, or the nose of the patient).
  • a suitable patient tissue site e.g., a finger, an ear lobe, a foot, or the nose of the patient.
  • the center wavelengths ⁇ i, ⁇ 2 required depend upon the blood analyte level to be determined.
  • ⁇ ⁇ may be in the Red wavelength range and ⁇ may be in the infrared wavelength range.
  • the pulse oximeter 10 may be readily implemented with more optical signal sources (e.g., four) depending upon the number of different blood analyte levels to be measured.
  • the optical signal sources 20a, 20b are activated by a corresponding plurality of drive signals 60a, 60b to emit the light signals 30a, 30b.
  • the drive signals 60a, 60b are supplied to the optical signal sources 20a, 20b by a corresponding plurality of drive signal sources 70a, 70b.
  • the drive signal sources 70a, 70b may be connected with a digital processor 80, which is driven with a clock signal 90 from a master clock 100.
  • the digital processor 80 may be programmed to define modulation waveforms, or drive patterns, for each of the optical signal sources 20a, 20b.
  • the digital processor 80 may provide separate digital trigger signals 110a, 110b to the drive signal sources 70a-d, which in turn generate the drive signals 60a, 60b.
  • the digital trigger signals 110a, 110b may be configured to provide for multiplexing of the drive signals 60a, 60b, and in turn the light signals 30a, 30b, in accordance with a multiplexing scheme (e.g., time division, frequency division, or code division multiplexing).
  • the drive signal sources 70a, 70b, processor 80, memory device 82 and clock 100 may all be housed in a monitor unit 120. While the illustrated embodiment shows the optical signal sources 20a, 20b physically interconnected with the positioning device 50 (e.g., mounted within the positioning device 50 or mounted within a connector end of a cable that is selectively connectable with the positioning device 50), it should be appreciated that the optical signal sources 20a, 20b may also be disposed within the monitor unit 120. In the latter case, the light signals 30a, 30b emitted from the optical signal sources 20a, 20b may be directed from the monitor unit 120 via one or more optical fibers to the positioning device 50 for transmission through the tissue site.
  • the drive signal sources 70a, 70b may comprise a single drive signal generator unit that supplies each of the drive signals 60a, 60b to the optical signal sources 20a, 20b.
  • Transmitted light signals 130a, 130b i.e., the portions of light signals 30a, 30b exiting the tissue
  • the detector 40 detects the intensities of the transmitted signals 130a, 130b and outputs a current signal 140 wherein the current level is indicative of the intensities of the transmitted signals 130a, 130b.
  • the current signal 140 output by the detector 40 comprises a multiplexed signal in the sense that it is a composite signal including information about the intensity of each of the transmitted signals 130a, 130b.
  • the current signal 140 may, for example, be time division multiplexed, wavelength division multiplexed, or code division multiplexed.
  • the current signal 140 is directed to an amplifier 150, which may be housed in the monitor unit 120 as is shown.
  • the amplifier 150 may instead be included in a probe/cable unit that is selectively connectable with the monitor unit 120.
  • the amplifier 150 converts the current signal 140 to a voltage signal 160 wherein a voltage level is indicative of the intensities of the transmitted signals 130a, 130b.
  • the amplifier 150 may also be configured to filter the current signal 140 from the detector 40 to reduce noise and aliasing.
  • the amplifier 150 may include a bandpass filter to attenuate signal components outside of a predetermined frequency range encompassing modulation frequencies of the drive signals 60a, 60b. Since the current signal 140 output by the detector 40 is a multiplexed signal, the voltage signal 160 is also a multiplexed signal, and thus, the voltage signal 160 must be demultiplexed in order to obtain signal portions corresponding with the intensities of the transmitted light signals 130a, 130b.
  • the digital processor 80 may be provided with demodulation software for demultiplexing the voltage signal 160. In order for the digital processor 80 to demodulate the voltage signal 160, it must first be converted from analog to digital.
  • A/D converter 170 receives the analog voltage signal 160 from the amplifier 150, samples the voltage signal 160, and converts the samples into a series of digital words 180 (e.g., eight, sixteen or thirty-two bit words), wherein each digital word is representative of the level of the voltage signal 160 (and hence the intensities of the transmitted light signals 130a, 130b) at a particular sample instance.
  • the A/D converter 170 should provide for sampling of the voltage signal 160 at a rate sufficient to provide for accurate tracking of the shape of the various signal portions comprising the analog voltage signal 160 being converted.
  • the A/D converter 170 may provide for a sampling frequency at least twice the frequency of the highest frequency drive signal 60a, 60b, and typically at an even greater sampling rate in order to more accurately represent the analog voltage signal.
  • the series of digital words 180 is provided by the A D converter 170 to the processor 80 to be demultiplexed. More particularly, the processor 80 may periodically send an interrupt signal 190 (e.g., once per every eight, sixteen or thirty-two clock cycles) to the A/D converter 170 that causes the A/D converter 170 to transmit one digital word 180 to the processor 80.
  • the demodulation software may then demultiplex the series of digital words 180 in accordance with an appropriate method (e.g., time, wavelength, or code) to obtain digital signal portions indicative of the intensities of each of the transmitted light signals 130a, 130b.
  • the demultiplexed digital signal portions comprise time domain plethysmographic signals corresponding to the center wavelengths ⁇ i, ⁇ 2 (e.g., red and infrared) of the optical signal sources 20a, 20b.
  • the red and infrared time domain plethysmographic signals may then be processed by the processor 80 to obtain desired patient physiological condition related information therefrom such as the patient's pulse rate and SP02 level.
  • a fuzzy neural-net multi-domain motion estimation and plethysmographic recognition signal processing method (200) that may be implemented on a pulse oximeter 10 such as illustrated in FIG. 1.
  • the signal processing method (200) may, for example, be implemented in computer software instructions executable by the processor 80.
  • the signal processing method (200) may be implemented in hardware, such as where the processor 80 comprises a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC) or the like.
  • the signal processing method (200) starts with obtaining (210) two continuous time domain plethysmographic signals such as red and infrared plethysmographic signals.
  • the red and infrared plethysmographic signals are digitized (220) by sampling the signals at a suitable frequency (e.g., at least about 50 Hz). Typical red and infrared time domain plethysmographic signals that have been sampled at 50 Hz are shown in FIG. 4. The signals depicted in FIG. 4 include motion artifacts. While the method (200) is illustrated with two input plethysmographic signals, in other embodiments, the signal processing method (200) may be configured for processing only one plethysmographic signal or processing more than two plethysmographic signals.
  • the digitized time domain red and infrared plethysmographic signals are directed to a smoothing module (230) wherein they are smoothed via a suitable smoothing window (e.g. Hanning, Hamming, or Kaiser). Smoothing the digitized plethysmographic signals achieves improved frequency estimation and prevents frequency spreading from discontinuities that would be seen with a rectangular time window. However, smoothing may not be advantageous in all instances and thus may not be incorporated in other embodiments of the method (200). Thereafter, the red and infrared plethysmographic signals are transformed (240) from the time domain to other suitable signal domains. Such signal domains include the cepstral domain, a Wavelet filtered domain, and various spectral domains.
  • the bispectrum can be used to obtain a coherence index that is useful to characterize non-linearities in time series via phase relations of harmonic components.
  • this means that the bispectrum has advantages for use in recognition of the pulse signature among various noise components that may be present in the time domain plethysmographic signals. Transformation (240) of the time domain red and infrared plethysmographic signals may be accomplished in a number of manners. As illustrated in FIG.
  • the digitized and smoothed time domain plethysmographic signals may first be processed in parallel via a complex FFT processing module (241).
  • the complex FFT processing module (241) outputs spectral domain red and infrared plethysmographic signals. If desired, the results of the FFT calculation may be scaled to help prevent floating point errors in subsequent computations.
  • the digitized time domain red and infrared plethysmographic signals are also directed to a Wavelet filter bank processing module (242).
  • the Wavelet filter bank processing module (242) applies a Wavelet transform to the time domain plethysmographic signals.
  • the cepstrum processing module (243) transforms the spectral domain plethysmographic signals output by the complex FFT processing module (241) to cepstral domain plethysmographic signals.
  • One manner of obtaining the cepstral domain plethysmographic signals is to first compute logarithmic scaled power spectra from the spectral domain plethysmographic signals and then apply a second stage complex FFT.
  • the cepstrum processing module (243) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared cepstral domain coefficient sets). Each of the cepstral domain coefficient sets are directed from the cepstrum processing module (243) to the input layer (251) of the neural network processing module (250).
  • the power spectrum processing module (244) computes red and infrared power spectrums from the spectral domain plethysmographic signals output by the complex FFT processing module (241).
  • the red and infrared power spectrums may be computed by squaring and summing the appropriate real and imaginary frequency components obtained by the FFT.
  • the power spectrum processing module (243) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared power spectral domain coefficient sets). Each of power spectral domain coefficient sets are directed from the power spectrum processing module (243) to the input layer (251) of the neural network processing module (250).
  • the bispectrum processing module (244) computes red and infrared bispectrums from the spectral domain plethysmographic signals output by the complex FFT processing module (241).
  • F( ⁇ ) the red and infrared bispectrums are defined in accordance with the following expression:
  • the bispectrum processing module (244) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared bispectral domain coefficient sets). Each of bispectral domain coefficient sets are directed from the bispectrum processing module (244) to the input layer (251) of the neural network processing module (250). It should be noted that in other embodiments, transformation (240) of the digitized time domain plethysmographic signals may not involve all of the domains shown in FIG. 2. For example, only one of the cepstral, power spectral, bispectral, or Wavelet filtered domains may be employed.
  • transformation (240) may only involve application of a Wavelet filter to the digitized time domain plethysmographic signals without employing the complex FFT processing module (241) to derive only Wavelet filtered time domain coefficients.
  • the various sets of red and infrared Wavelet filtered domain, cepstral domain, power spectral domain, and bispectral domain coefficients may be concatenated in vector form for presentation to the neural network input stage.
  • T e assembled vector basically represents the information present to the time epoch of the smoothing window.
  • Each successive assembled vector or frame represents different successive time samples of the input signal, in sync with the smoothing window time shift increment (typically half the duration of the smoothing window).
  • Each red and infrared vector is associated with a single frame of sample instances in the digitized time domain plethysmographic signals.
  • processing of the red and infrared plethysmographic signals in accordance with the method (200) depicted in FIG. 2 is preferably undertaken each time a number m of new sample instances is/are received upon digitizing (220) the input plethysmographic signals.
  • m may be one, two, or more sample instances, and m may be predetermined or may vary depending upon factors such as classification of the signal in accordance with the method (200).
  • each frame may correspond with a window of sample instances.
  • the window length may be fixed.
  • the current sample instance and n past sample instances may be used in performing the various calculations involved in the transformation (240) of the plethysmographic signals into the various signal domains.
  • the predetermined number n may, for example, be determined empirically based on tests conducted using known plethysmographic data sets or it may be established during training of the neural network processing module (250).
  • the window length may also vary.
  • the current sample instance and a varying number of past sample instances may be used in performing the various calculations involved in the transformation (240) of the plethysmographic signals into the various signal domains.
  • the number of past sample instances included in the varying length window may, for example, be varied depending upon factors such as classification of the signal in accordance with the method (200).
  • current calculations will not be influenced by older sample instances that are discarded once outside the window length.
  • each frame may be recursive.
  • the current sample instance and all past sample instances are included in each successive frame .
  • current calculations will be influenced by older sample instances since no sample instances are discarded when the frames are of the recursive type.
  • the influence of older sample instances on current calculations can be reduced, if desired, by weighting older sample instances less than current sample instances.
  • the various sets or vectors of red and infrared Wavelet filtered domain, cepstral domain, power spectral domain, and bispectral domain coefficients associated with each respective frame are directed to the input layer (251) of the neural network processing module (250).
  • the input layer (251) may also receive additional inputs from an additional signal characteristics processing module (260).
  • the additional signal characteristics may be derived by the additional signal characteristics processing module (260) from one or more of the red and infrared digitized signals, spectral domain signals, cepstral domain signals, power spectral signals, and bispectral signals.
  • Such additional signal characteristics include: (1) an RMS energy measure; (2) a spikiness measure; (3) a spectral jitter measure; (4) a spectral shimmer measure; (5) a spectral smear measure; (6) cepstral peak jitter measure; and (7) a cepstral peak position measure.
  • the input layer (251) assembles the various coefficients and additional signal characteristics, if any, received thereby and directs such inputs (as appropriately weighted) to one or more hidden layers (252) of the neural network (250).
  • the hidden layer(s) (252) receive the appropriately weighted input coefficients and additional signal characteristics, if any, and direct such values (as appropriately weighted) therefrom to an output layer (253) of the neural network (250).
  • inputs from at least three data frames are processed simultaneously by the neural network (250) to classify the red and infrared plethysmographic signals by plethysmographic signal type (271) and motion artifact type (272), as well as to generate a respiration index (273) (e.g., depth and rate).
  • plethysmographic signal type 271
  • motion artifact type 272
  • respiration index 273
  • the processor 80 of the pulse oximeter 10 may utilize the information in various manners. For example, based on the plethysmographic signal type and motion type, the processor 80 may undertake different filtering of the plethysmographic signals before deriving physiological conditions from the signals.
  • the processor 80 may choose to examine the cepstral transform to extract the signal pulse component. For a severe clenching motion the processor 80 may decide to use DC tracking to determine Sp02 and choose not to attempt to extract the pulse frequency. If the net output classification indicated a highly irregular plethysmographic signal type but little motion artifact then the processor 80 might extract pulse rate frequency and Sp02 values from the time domain. By way of further example, based on the plethysmographic type and motion type, the processor 80 may choose to adjust how patient physiological conditions are derived from the plethysmographic signals.
  • the neural network classifier can be trained on abnormal signals (e.g., extreme arrhythmia's and different heart conditions) so that it will be able to alert medical personnel to occurrence of unusual waveforms possibly due to the onset of a critical physiological condition.
  • abnormal signals e.g., extreme arrhythmia's and different heart conditions
  • FIG. 3 there is shown a schematic representation of one embodiment of a neural network 300 architecture that may be employed for plethysmographic recognition, respiration activity and motion estimation in accordance with the method (200) illustrated in FIG. 2.
  • the neural network 300 includes an input layer 302, one or more hidden layers 304, and an output classification layer 306.
  • the input layer 302 receives input values in the form one or more frames 308 of input values produced at different times from the current frame tO backwards to frame t-n, where n is an integer. As depicted, in one embodiment there are three frames (tO, t-1, t-2). The time interval between frames 308 may be linear, as depicted, or it may be non-linear. Each frame 308 may be made up of input values from a variety of signal transforms (e.g., cepstral, bispectral, Wavelet, power spectral) as well as additional signal characteristics, if any. Each value in the frames 308 activates a separate node (also referred to herein as neuron) in the neural network 300 input layer 302.
  • a separate node also referred to herein as neuron
  • Each node in the input layer 302 activates each node in the first hidden layer 304 interconnected thereto.
  • Each node in the last hidden layer 304 activates each node in the output classification layer 306 interconnected thereto.
  • the output classification layer 306 includes a number of output nodes. The total number of output nodes typically corresponds to the number of different types of classification required. In this regard, the output nodes may generally be classified as being plethysmographic type output nodes 310 or motion type output nodes 312.
  • the output nodes provide indications as to the presence of the particular type of plethysmographic signal or motion in the red and infrared plethysmographic signals by firing at various levels.
  • a level approaching a specified high value indicates a strong presence of the particular type of plethysmographic signal or motion with which the node is associated and a level approaching a specified low value (e.g., 0) indicates a weak presence of the particular type of plethysmographic signal or motion with which the node is associated.
  • a level approaching a specified low value e.g., 0
  • the red and infrared plethysmographic signals are classified by the various levels of the plethysmographic type and motion type output nodes 310, 312.
  • respiration rate output node there may be a respiration rate output node and a respiration depth output node.
  • the values (e.g., from 0 to 1) of the respiration rate and depth nodes comprise the respiration index of the patient.
  • comparing the Wavelet filter bank coefficients from both the red and infrared input signals over time may offer the neural network 300 the required information necessary to determine respiration activity, since the comparison (which the neural network 300 may do in a number of ways) can provide Sp02 estimates. It is the pattern of such Sp02 estimates over time that provides information to build a respiration index.
  • allowing the neural network 300 to compare the other spectral transforms from both the red and infrared input signals assists in extracting information that provides Sp02 estimates.
  • the neural network 300 architecture as described is a feed-forward network, and when there is only a single hidden layer 304, it is a three-layer feed-forward network.
  • Another form of the neural network 300 can be configured by adding unit delay operators 316 at the hidden layer 304.
  • the schematic connections of the unit delay operators 316 are depicted in dashed lines to indicate that they may be optionally included in order to achieve a recursive neural network 300 architecture.
  • the neural network 300 should be trained prior to use in order to establish the appropriate weights applied to the various interconnections between the various nodes.
  • One method of training the neural network is the backpropagation algorithm.
  • the backpropagation algorithm may be supervised by a fuzzy logic controller processing module.
  • the fuzzy logic controller is referenced in FIG.
  • the fuzzy logic controller (280) helps speed up the convergence of the backpropagation learning, and in some cases enables convergence to take place, by controlling the learning parameters with a fuzzy rule set. Essentially some heuristics are applied dependent on the change of error (CE - related to error gradient) and the change of CE (CCE - related to second-order error gradient). This has been shown to accelerate learning.
  • the input to the neural network 300 may also be "fuzzified" by slight perturbing and reordering input values. Such a technique can increase the robustness of the neural network 300 to recognize patterns in noise. Referring now to FIG.
  • the neural network training method (500) establishes the appropriate weighting values for the interconnections between the neurons in the neural network architecture.
  • the neural network training method (500) is implemented on a computer system (e.g., a desktop or laptop computer). After a trained neural network is generated on the computer, the trained neural network can then be installed on a number of pulse oximeters.
  • a computer system e.g., a desktop or laptop computer.
  • the neural network training method (500) begins with selecting (510) a number of plethysmographic signal data sets that are associated with a number of different types of predetermined signal conditions.
  • the plethysmographic signal data sets may be stored in a database of exemplary plethysmographic signal data sets that is stored on a hard drive of the computer or other storage media accessible by the computer.
  • different types of predetermined signal conditions include: (1) normal adult plethysmographic signal; (2) normal baby plethysmographic signal; (3) fixed frequency tapping motion without plethysmographic signal; (4) fixed frequency tapping motion with plethysmographic signal; (5) moving frequency tapping motion without plethysmographic signal; (6) moving frequency tapping motion with plethysmographic signal; (7) clenching motion without plethysmographic signal; (8) clenching motion with plethysmographic signal; (9) baby-kicking motion without plethysmographic signal; (10) baby-kicking motion with plethysmographic signal; (11) irregular plethysmographic signal associated with patient arrhythmia; (12) low perfusion condition plethysmographic signal; (13) plethysmographic signal with fast breathing; (14) plethysmographic signal with shallow breathing; (15) plethysmographic signal with extended breath holding; and (16) no plethysmographic
  • the plethysmographic signal data sets are stored in the form of time domain signals.
  • the selected plethysmographic signal data sets are transformed (520) from the time domain to the other signal domains (e.g., cepstral, power spectral, bispectral, Wavelet filtered domains) to be utilized as inputs to the neural network processing module (250). Transformation (520) results in a corresponding number of transformed plethysmographic signal data sets.
  • sets or vectors of coefficients are extracted (530) from the transformed plethysmographic signal data sets.
  • the steps of transforming (520) and extracting (530) may, for example, be accomplished as described in connection with the transforming step (240) of the method (200) of FIG. 2.
  • additional signal characteristics may be derived (540) from the plethysmographic signal data sets and the various transformed signal domains.
  • the additional signal characteristics that are derived should be the same ones as are intended to be employed by the neural network processing module (250) in addition to the various signal domain coefficients.
  • the sets or vectors of extracted coefficients and the additional signal characteristics are then used (550) as inputs to train the neural network.
  • a learning procedure is used to adjust (560) the weighting values associated with the various interconnections between neurons in the neural network until the weighting values are optimized.
  • the trained neural network may have difficultly recognizing and classifying a particular plethysmographic signal encountered in the field.
  • it may be desirable to add such plethysmographic signals e.g., new plethysmographic signals of interest
  • new plethysmographic signals of interest may be stored in the memory device(s) 82 of the pulse oximeter(s) 10.
  • new plethysmographic signals of interest may be retrieved (570) from the memory device(s) 82 of the pulse oximeter(s) and added (580) to the database of plethysmographic signal data sets.

Abstract

Pulse oximetry is improved through classification of plethysmographic signals by processing the plethysmographic signals using a neural network that receives input coefficients from multiple signal domains including, for example, spectral, bispectral, cepstral and Wavelet filtered signal domains. In one embodiment, a plethysmographic signal obtained from a patient is transformed (240) from a first domain to a plurality of different signal domains (242, 243, 244, 245) to obtain a corresponding plurality of transformed plethysmographic signals. A plurality of sets of coefficients derived from the transformed plethysmographic signals are selected and directed to an input layer (251) of a neural network (250). The plethysmographic signal is classified by an output layer (253) of the neural network (250) that is connected to the input layer (251) by one or more hidden layers (252).

Description

MULTI-DOMAIN MOTION ESTIMATION AND PLETHYSMOGRAPHIC RECOGNITION USING FUZZY NEURAL-NETS
FIELD OF THE INVENTION The present invention relates generally to plethysmographic signal processing, and more particularly to the application of neural net processes to the classification of plethysmographic signals and the estimation of motion therein.
BACKGROUND OF THE INVENTION Signal attenuation measurements generally involve transmitting a signal towards or through a tissue medium under analysis, detecting the signal transmitted through or reflected by the medium and computing a parameter value for the medium based on attenuation of the signal by the medium. In simultaneous signal attenuation measurement systems, multiple signals are simultaneously transmitted (i.e., two or more signals are transmitted during at least one measurement interval) to the medium and detected in order to obtain information regarding the medium. Such attenuation measurement systems are used in various applications in various industries. For example, in the medical or health care field, optical (i.e., visible spectrum or other wavelength) signals are utilized to monitor the composition of respiratory and anesthetic gases, and to analyze tissue or a blood sample with regard to oxygen saturation (Sp02 level), analyte values (e.g., related to certain hemoglobins) or other composition related values. The case of pulse oximetry is illustrative. Some pulse oximeters extract information regarding patient physiological conditions such as the patient's pulse rate and an oxygen saturation level of the patient's blood, or related analyte values, via analysis of plethysmographic signals or waveforms corresponding to different wavelengths of light transmitted through or reflected from the patient's tissue. In particular, pulse oximeters generally include a probe for attaching to a patient's tissue site such as a finger, earlobe, nasal septum, or foot. The probe is used to transmit pulsed optical signals of at least two wavelengths, typically red and infrared, to the patient's tissue site. The different wavelengths of light used are often referred to as the channels of the pulse oximeter (e.g., the red and infrared channels). The optical signals are attenuated by the patient tissue site and subsequently are received by a detector that provides an analog electrical output signal representative of the received optical signals. The attenuated optical signals as received by the detector are often referred to as the transmitted signals. The electrical signal can be processed to obtain plethysmographic signals for each channel and the plethysmographic signals may be analyzed to obtain information regarding patient physiological conditions. Extraction of patient physiological conditions from the plethysmographic signals can be quite effective using a well positioned sensor and when the patient or subject is resting. However motion artifacts can easily swamp the desired information included in the plethysmographic signals when the patient is moving around and/or performing muscular contractions. Some motion artifacts can severely impair the signals, whereas other types can be filtered out or do not significantly effect the desired information included in the plethysmographic signals. Furthermore, depending upon the severity and type of motion artifacts present in the plethysmographic signals, some techniques for extracting the desired patient physiological conditions may not be appropriate and alternative techniques may need to be employed. Another potential problem that can occur when attempting to make a pulse oximetry system robust to motion artifacts is that heart arrhythmia or rapid pulse variations might possibly be sensed as motion effects and cause motion rejection steps to be applied which may be inappropriate in these cases and could cause false pulse-rate and Sp02 readings. SUMMARY OF THE INVENTION Accordingly, the present invention is directed to the use of multiple signal domains in conjunction with neural net processing to achieve pattern classification of different types of motion artifacts and also to classify different patterns of pulse and plethysmographic waveforms that can occur under different physiological conditions. The implementation of a neural-net pattern recognition stage within a pulse oximeter addresses a number of the aforementioned problems thereby achieving an improved instrument. In accordance with the present invention, motion classification is performed using features extracted from the time, spectral (e.g. power spectral), bispectral and cepstral domains. These features and the spectral (e.g. power spectral), bispectral, and cepstral coefficients are input to a neural network (also referred to herein as a neural-net) to perform recognition of plethysmographic waveforms, and the type of motion or combination of pulsatile waveform and motion artifacts. Another form of information that can be used is coefficients derived from a Wavelet filter bank. The information from the different sources described above is weighted and processed via a neural-net and the output classification of the neural-net is then used to help with filtering out of motion artifacts and in the decision to use selective portions of high- pass frequency information (AC) in Sp02 calculations. The classification from the neural- net is used to determine the existence and severity and type of motion artifact. The motion classification can be used to improve the signal extraction in noisy conditions, and to aid in selection of the most appropriate signal extraction method. Further the neural-net classifier can be used to determine the type of plethysmographic signal, such as normal adult, infant, or that produced from different types of heart conditions. Identification of the signal type will aid in measurement and in extraction of the signal from noise. One aspect of the invention is in describing the different types of physiological and noise conditions for which plethysmographic waveforms are collected for training the neural-net classifier. In order to improve the robustness of the classifier the inputs to the neural-net during training are subjected to 'fuzzification' wherein input values are slightly perturbed and reordered. This process increases the size of the training set and also has the effect of making the neural-net less rigid in its ability to classify patterns. The training of the neural-net can also be enhanced by use of a fuzzy logic controller which is used to adjust learning parameters and to speed up convergence during back-propagation learning. Another aspect of the invention is that the neural-net is trained to respond to different respiration activities - both rate and depth. Pertinent information from spectral coefficients, and Wavelet transforms on the different input wavelengths (typically red and infrared, but also additional wavelengths which may be used) are presented to the neural- net. Optionally derived Sp02 estimates can also be presented. All this information can be used to obtain further more robust Sp02 estimates. Thus the neural-net provides a variety of information which can be used to classify respiration conditions. According to another aspect of the invention, a pulse oximeter includes a first optical signal source that is operable to emit an optical signal characterized by a first wavelength (e.g., red) and a second optical signal source that is operable to emit an optical signal characterized by a second wavelength (e.g., infrared) different than the first wavelength. A detector operable to receive the first and second optical signals after they are attenuated by a patient tissue site provides a detector output signal representative of the attenuated first and second optical signals. A processor is enabled to obtain first and second time domain plethysmographic signals from the detector output signal. The processor also classifies one or more of the first and second time domain plethysmographic signals using a neural network. The neural network receives input coefficients derived from one or more transforms of the first and/or second time domain plethysmographic signals. The transforms may, for example, include spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filter bank transforms. According to a further aspect of the present invention, a method of processing a plethysmographic signal obtained from a patient in a first signal domain (e.g., time) includes transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain. The different signal domains may, for example, include spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filtered domains. Transformation of the first domain plethysmographic signal results in a plurality of transformed plethysmographic signals with each transformed plethysmographic signal being in one of the different signal domains. A plurality of sets of coefficients are selected, with each set of coefficients being derived from a corresponding one of the transformed plethysmographic signals. The sets of coefficients are input to a neural network, and the plethysmographic signal is classified based on an output from the neural network. According to yet another aspect of the present invention, a method of training a neural network to classify a plethysmographic signal obtained from a patient includes selecting a plurality of first domain plethysmographic signal data sets associated with a plurality of different types of predetermined signal conditions from a database of plethysmographic signal data sets. The first domain plethysmographic signal data sets are transformed to other signal domains different than the first domain to obtain a corresponding plurality of transformed plethysmographic signal data sets. In this regard, the first domain may, for example, be the time domain and the other signal domains include, for example, spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filtered domains. A plurality of sets of coefficients are extracted from the transformed plethysmographic signal data sets, with each set of coefficients being extracted from a corresponding one of the transformed plethysmographic signal data sets. The sets of extracted coefficients are used as inputs to the neural network, and weighting values associated with connections between neurons in the neural network are adjusted in accordance with a learning procedure. The learning procedure may, for example, be a backpropagation learning procedure or a simulated annealing learning procedure. Where desired, the backpropagation learning procedure may be implemented with fuzzy logic control. According to one more aspect of the present invention, a method of providing information relating to a physiological condition of a patient based on at least one plethysmographic signal obtained from the patient in a first signal domain (e.g., time) includes transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain. The different signal domains may, for example, include spectral (e.g., power spectral), bispectral, cepstral, and Wavelet filtered domains. Transformation of the first domain plethysmographic signal results in a plurality of transformed plethysmographic signals with each transformed plethysmographic signal being in one of the different signal domains. The plethysmographic signal is classified based on an output from a neural network. The output of the neural network is based on input coefficients derived from one or more of the transformed plethysmographic signals. Based on the classification, a technique for determining the physiological condition of the patient is selected. The physiological condition of the patient may, for example, comprise a pulse rate. Where at least two plethysmographic signals corresponding to different optical wavelengths are transformed and classified, the physiological condition of the patient may, for example, comprise an Sp02 value or a respiration index. Where the physiological condition of the patient comprises a respiration index, the plethysmographic signals are preferably transformed using at least a Wavelet filter bank transform. These and other aspects and advantages of the present invention will be apparent upon review of the following Detailed Description when taken in conjunction with the accompanying figures.
DESCRIPTION OF THE DRAWINGS For a more complete understanding of the present invention and further advantages thereof, reference is now made to the following Detailed Description, taken in conjunction with the drawings, in which: FIG. 1 is a block diagram of one embodiment of a pulse oximeter employing multi-domain motion estimation and plethysmographic signal recognition using fuzzy neural-nets in accordance with the present invention; FIG. 2 is a block diagram showing one embodiment of a method of implementing multi-domain motion estimation and plethysmographic signal recognition using fuzzy neural-nets in accordance with the present invention; FIG. 3 is a schematic diagram illustrating one embodiment of a neural network architecture that may be employed in accordance with the present invention; FIG. 4 is a plot of exemplary red and infrared plethysmographic signals that include motion artifacts; FIG. 5 is a block diagram showing one embodiment of a neural network training method in accordance with the present invention; and FIGS. 6A-6D are plots showing exemplary plethysmographic signal conditions that may be included within a database of plethysmographic training sets.
DETAILED DESCRIPTION Referring now to Fig. 1, there is shown a block diagram of one embodiment of a pulse oximeter 10 in which multi-domain motion estimation and plethysmographic signal recognition using fuzzy neural-nets may be implemented. The pulse oximeter 10 is configured for use in determining the pulse rate of a patient as well as one or more blood analyte levels in the patient, such as an SP02 level. It should be appreciated that multi- domain motion estimation and plethysmographic signal recognition using fuzzy neural- nets in accordance with the present invention may be implemented in pulse oximeters that are configured differently from the pulse oximeter depicted in FIG. 1 as well as in other environments wherein plethysmographic signals are processed in order to obtain desired information relating to patient physiological conditions from the plethysmographic signals. The pulse oximeter 10 includes a pair of optical signal sources 20a, 20b for emitting a corresponding pair of light signals 30a, 30b centered at different predetermined center wavelengths λ \, λ 2 through a suitable tissue site of a patient and on to a detector 40 (e.g., a photo-sensitive diode). The optical signal sources 20a, 20b and detector 40 may be included in a positioning device 50, or probe, to facilitate alignment of the light signals 30a, 30b with the detector 40. For example, the positioning device 50 may be of clip-type or flexible strip configuration adapted for selective attachment to a suitable patient tissue site (e.g., a finger, an ear lobe, a foot, or the nose of the patient). The center wavelengths λ i, λ 2 required depend upon the blood analyte level to be determined. For example, in order to determine an SP02 level, λ ι may be in the Red wavelength range and λ may be in the infrared wavelength range. It should be appreciated that the pulse oximeter 10 may be readily implemented with more optical signal sources (e.g., four) depending upon the number of different blood analyte levels to be measured. The optical signal sources 20a, 20b are activated by a corresponding plurality of drive signals 60a, 60b to emit the light signals 30a, 30b. The drive signals 60a, 60b are supplied to the optical signal sources 20a, 20b by a corresponding plurality of drive signal sources 70a, 70b. The drive signal sources 70a, 70b may be connected with a digital processor 80, which is driven with a clock signal 90 from a master clock 100. The digital processor 80 may be programmed to define modulation waveforms, or drive patterns, for each of the optical signal sources 20a, 20b. In this regard, there may be a separate memory device 82 interfaced with the digital processor 80 (or the memory device 82 may be incorporated in the processor 80) on which various software instructions executable by the processor 80 are stored. More particularly, the digital processor 80 may provide separate digital trigger signals 110a, 110b to the drive signal sources 70a-d, which in turn generate the drive signals 60a, 60b. In this regard, the digital trigger signals 110a, 110b may be configured to provide for multiplexing of the drive signals 60a, 60b, and in turn the light signals 30a, 30b, in accordance with a multiplexing scheme (e.g., time division, frequency division, or code division multiplexing). The drive signal sources 70a, 70b, processor 80, memory device 82 and clock 100 may all be housed in a monitor unit 120. While the illustrated embodiment shows the optical signal sources 20a, 20b physically interconnected with the positioning device 50 (e.g., mounted within the positioning device 50 or mounted within a connector end of a cable that is selectively connectable with the positioning device 50), it should be appreciated that the optical signal sources 20a, 20b may also be disposed within the monitor unit 120. In the latter case, the light signals 30a, 30b emitted from the optical signal sources 20a, 20b may be directed from the monitor unit 120 via one or more optical fibers to the positioning device 50 for transmission through the tissue site. Furthermore, the drive signal sources 70a, 70b may comprise a single drive signal generator unit that supplies each of the drive signals 60a, 60b to the optical signal sources 20a, 20b. Transmitted light signals 130a, 130b (i.e., the portions of light signals 30a, 30b exiting the tissue) are detected by the detector 40. The detector 40 detects the intensities of the transmitted signals 130a, 130b and outputs a current signal 140 wherein the current level is indicative of the intensities of the transmitted signals 130a, 130b. As may be appreciated, the current signal 140 output by the detector 40 comprises a multiplexed signal in the sense that it is a composite signal including information about the intensity of each of the transmitted signals 130a, 130b. Depending upon the nature of the drive signals 60a, 60b, the current signal 140 may, for example, be time division multiplexed, wavelength division multiplexed, or code division multiplexed. The current signal 140 is directed to an amplifier 150, which may be housed in the monitor unit 120 as is shown. As an alternative, the amplifier 150 may instead be included in a probe/cable unit that is selectively connectable with the monitor unit 120. The amplifier 150 converts the current signal 140 to a voltage signal 160 wherein a voltage level is indicative of the intensities of the transmitted signals 130a, 130b. The amplifier 150 may also be configured to filter the current signal 140 from the detector 40 to reduce noise and aliasing. By way of example, the amplifier 150 may include a bandpass filter to attenuate signal components outside of a predetermined frequency range encompassing modulation frequencies of the drive signals 60a, 60b. Since the current signal 140 output by the detector 40 is a multiplexed signal, the voltage signal 160 is also a multiplexed signal, and thus, the voltage signal 160 must be demultiplexed in order to obtain signal portions corresponding with the intensities of the transmitted light signals 130a, 130b. In this regard, the digital processor 80 may be provided with demodulation software for demultiplexing the voltage signal 160. In order for the digital processor 80 to demodulate the voltage signal 160, it must first be converted from analog to digital. Conversion of the analog voltage signal 160 is accomplished with an analog-to-digital (A/D) converter 170, which may also be included in the monitor unit 120. The A/D converter 170 receives the analog voltage signal 160 from the amplifier 150, samples the voltage signal 160, and converts the samples into a series of digital words 180 (e.g., eight, sixteen or thirty-two bit words), wherein each digital word is representative of the level of the voltage signal 160 (and hence the intensities of the transmitted light signals 130a, 130b) at a particular sample instance. In this regard, the A/D converter 170 should provide for sampling of the voltage signal 160 at a rate sufficient to provide for accurate tracking of the shape of the various signal portions comprising the analog voltage signal 160 being converted. For example, the A/D converter 170 may provide for a sampling frequency at least twice the frequency of the highest frequency drive signal 60a, 60b, and typically at an even greater sampling rate in order to more accurately represent the analog voltage signal. The series of digital words 180 is provided by the A D converter 170 to the processor 80 to be demultiplexed. More particularly, the processor 80 may periodically send an interrupt signal 190 (e.g., once per every eight, sixteen or thirty-two clock cycles) to the A/D converter 170 that causes the A/D converter 170 to transmit one digital word 180 to the processor 80. The demodulation software may then demultiplex the series of digital words 180 in accordance with an appropriate method (e.g., time, wavelength, or code) to obtain digital signal portions indicative of the intensities of each of the transmitted light signals 130a, 130b. In this regard, the demultiplexed digital signal portions comprise time domain plethysmographic signals corresponding to the center wavelengths λ i, λ 2 (e.g., red and infrared) of the optical signal sources 20a, 20b. The red and infrared time domain plethysmographic signals may then be processed by the processor 80 to obtain desired patient physiological condition related information therefrom such as the patient's pulse rate and SP02 level. Referring now to FIG. 2, there is shown one embodiment of a fuzzy neural-net multi-domain motion estimation and plethysmographic recognition signal processing method (200) that may be implemented on a pulse oximeter 10 such as illustrated in FIG. 1. In this regard, where the processor 80 comprises a general purpose microprocessor or the like, the signal processing method (200) may, for example, be implemented in computer software instructions executable by the processor 80. In other embodiments, the signal processing method (200) may be implemented in hardware, such as where the processor 80 comprises a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC) or the like. The signal processing method (200) starts with obtaining (210) two continuous time domain plethysmographic signals such as red and infrared plethysmographic signals. The red and infrared plethysmographic signals are digitized (220) by sampling the signals at a suitable frequency (e.g., at least about 50 Hz). Typical red and infrared time domain plethysmographic signals that have been sampled at 50 Hz are shown in FIG. 4. The signals depicted in FIG. 4 include motion artifacts. While the method (200) is illustrated with two input plethysmographic signals, in other embodiments, the signal processing method (200) may be configured for processing only one plethysmographic signal or processing more than two plethysmographic signals. The digitized time domain red and infrared plethysmographic signals are directed to a smoothing module (230) wherein they are smoothed via a suitable smoothing window (e.g. Hanning, Hamming, or Kaiser). Smoothing the digitized plethysmographic signals achieves improved frequency estimation and prevents frequency spreading from discontinuities that would be seen with a rectangular time window. However, smoothing may not be advantageous in all instances and thus may not be incorporated in other embodiments of the method (200). Thereafter, the red and infrared plethysmographic signals are transformed (240) from the time domain to other suitable signal domains. Such signal domains include the cepstral domain, a Wavelet filtered domain, and various spectral domains. In this regard, nth order cumulant spectral domains such as, for example, the power spectrum (n =2), the bispectrum (n= 3), and the trispectrum (n= 4) are generally of interest, with the power spectrum and the bispectrum being particularly useful. In this regard, the bispectrum can be used to obtain a coherence index that is useful to characterize non-linearities in time series via phase relations of harmonic components. In practical terms, this means that the bispectrum has advantages for use in recognition of the pulse signature among various noise components that may be present in the time domain plethysmographic signals. Transformation (240) of the time domain red and infrared plethysmographic signals may be accomplished in a number of manners. As illustrated in FIG. 2, the digitized and smoothed time domain plethysmographic signals may first be processed in parallel via a complex FFT processing module (241). The complex FFT processing module (241) outputs spectral domain red and infrared plethysmographic signals. If desired, the results of the FFT calculation may be scaled to help prevent floating point errors in subsequent computations. The output from the complex FFT processing module
(241) is then directed to a cepstrum processing module (243), and various nth-order cumulant spectral domain processing modules, including, in this embodiment, a power spectrum processing module (244) and a bispectrum processing module (246). As part of the transformation (240), the digitized time domain red and infrared plethysmographic signals are also directed to a Wavelet filter bank processing module (242). The Wavelet filter bank processing module (242) applies a Wavelet transform to the time domain plethysmographic signals. The Wavelet filter bank processing module
(242) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared Wavelet filtered domain coefficient sets). Each of Wavelet filtered domain coefficient sets are directed from the Wavelet filter bank processing module (242) to an input layer (251) of a neural network processing module (250). The cepstrum processing module (243) transforms the spectral domain plethysmographic signals output by the complex FFT processing module (241) to cepstral domain plethysmographic signals. One manner of obtaining the cepstral domain plethysmographic signals is to first compute logarithmic scaled power spectra from the spectral domain plethysmographic signals and then apply a second stage complex FFT. In this regard, more detail concerning cepstral domain processing of plethysmographic signals is provided in United States Patent No. 6,650,918 entitled "CEPSTRAL
DOMAIN PULSE OXIMETRY", the entire disclosure of which is hereby incorporated by reference herein. If desired, the results of the FFT calculation may be scaled to help prevent floating point errors in subsequent computations. The cepstrum processing module (243) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared cepstral domain coefficient sets). Each of the cepstral domain coefficient sets are directed from the cepstrum processing module (243) to the input layer (251) of the neural network processing module (250). The power spectrum processing module (244) computes red and infrared power spectrums from the spectral domain plethysmographic signals output by the complex FFT processing module (241). In this regard, the red and infrared power spectrums may be computed by squaring and summing the appropriate real and imaginary frequency components obtained by the FFT. The power spectrum processing module (243) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared power spectral domain coefficient sets). Each of power spectral domain coefficient sets are directed from the power spectrum processing module (243) to the input layer (251) of the neural network processing module (250). The bispectrum processing module (244) computes red and infrared bispectrums from the spectral domain plethysmographic signals output by the complex FFT processing module (241). In this regard, for a Fourier transform F(ω), the red and infrared bispectrums are defined in accordance with the following expression:
B(ωιrω2) = E[F(ωι)F(ω2)F(ωι + 2)]
where ω\ and ω2 are the frequencies present in the spectrums. The bispectrum processing module (244) outputs sets of coefficients corresponding to each of the input plethysmographic signals (the red and infrared bispectral domain coefficient sets). Each of bispectral domain coefficient sets are directed from the bispectrum processing module (244) to the input layer (251) of the neural network processing module (250). It should be noted that in other embodiments, transformation (240) of the digitized time domain plethysmographic signals may not involve all of the domains shown in FIG. 2. For example, only one of the cepstral, power spectral, bispectral, or Wavelet filtered domains may be employed. Or, a different combination of such domains (e.g., cepstral and power spectral, cepstral and bispectral, etc.) may be employed. Furthermore, in other embodiments transformation (240) may only involve application of a Wavelet filter to the digitized time domain plethysmographic signals without employing the complex FFT processing module (241) to derive only Wavelet filtered time domain coefficients. The various sets of red and infrared Wavelet filtered domain, cepstral domain, power spectral domain, and bispectral domain coefficients may be concatenated in vector form for presentation to the neural network input stage. T e assembled vector basically represents the information present to the time epoch of the smoothing window. Each successive assembled vector or frame represents different successive time samples of the input signal, in sync with the smoothing window time shift increment (typically half the duration of the smoothing window). Each red and infrared vector is associated with a single frame of sample instances in the digitized time domain plethysmographic signals. In this regard, processing of the red and infrared plethysmographic signals in accordance with the method (200) depicted in FIG. 2 is preferably undertaken each time a number m of new sample instances is/are received upon digitizing (220) the input plethysmographic signals. In this regard, m may be one, two, or more sample instances, and m may be predetermined or may vary depending upon factors such as classification of the signal in accordance with the method (200). Thus, each time m sample instances is/are received, a new frame is established. Each frame may correspond with a window of sample instances. In this regard, the window length may be fixed. For example, the current sample instance and n past sample instances (n being a predetermined number) may be used in performing the various calculations involved in the transformation (240) of the plethysmographic signals into the various signal domains. The predetermined number n may, for example, be determined empirically based on tests conducted using known plethysmographic data sets or it may be established during training of the neural network processing module (250). The window length may also vary. For example, the current sample instance and a varying number of past sample instances may be used in performing the various calculations involved in the transformation (240) of the plethysmographic signals into the various signal domains. The number of past sample instances included in the varying length window may, for example, be varied depending upon factors such as classification of the signal in accordance with the method (200). As may be appreciated, current calculations will not be influenced by older sample instances that are discarded once outside the window length. Alternatively, each frame may be recursive. In this regard, the current sample instance and all past sample instances are included in each successive frame . As may be appreciated, current calculations will be influenced by older sample instances since no sample instances are discarded when the frames are of the recursive type. However, the influence of older sample instances on current calculations can be reduced, if desired, by weighting older sample instances less than current sample instances. Each time a frame is established, the various sets or vectors of red and infrared Wavelet filtered domain, cepstral domain, power spectral domain, and bispectral domain coefficients associated with each respective frame are directed to the input layer (251) of the neural network processing module (250). In addition to the various sets of coefficients, the input layer (251) may also receive additional inputs from an additional signal characteristics processing module (260). The additional signal characteristics may be derived by the additional signal characteristics processing module (260) from one or more of the red and infrared digitized signals, spectral domain signals, cepstral domain signals, power spectral signals, and bispectral signals. Such additional signal characteristics include: (1) an RMS energy measure; (2) a spikiness measure; (3) a spectral jitter measure; (4) a spectral shimmer measure; (5) a spectral smear measure; (6) cepstral peak jitter measure; and (7) a cepstral peak position measure. The input layer (251) assembles the various coefficients and additional signal characteristics, if any, received thereby and directs such inputs (as appropriately weighted) to one or more hidden layers (252) of the neural network (250). The hidden layer(s) (252) receive the appropriately weighted input coefficients and additional signal characteristics, if any, and direct such values (as appropriately weighted) therefrom to an output layer (253) of the neural network (250). Preferably, inputs from at least three data frames are processed simultaneously by the neural network (250) to classify the red and infrared plethysmographic signals by plethysmographic signal type (271) and motion artifact type (272), as well as to generate a respiration index (273) (e.g., depth and rate). Once the plethysmographic signals (or frames thereof) are classified, the processor 80 of the pulse oximeter 10 may utilize the information in various manners. For example, based on the plethysmographic signal type and motion type, the processor 80 may undertake different filtering of the plethysmographic signals before deriving physiological conditions from the signals. For example, for a tapping motion the processor 80 may choose to examine the cepstral transform to extract the signal pulse component. For a severe clenching motion the processor 80 may decide to use DC tracking to determine Sp02 and choose not to attempt to extract the pulse frequency. If the net output classification indicated a highly irregular plethysmographic signal type but little motion artifact then the processor 80 might extract pulse rate frequency and Sp02 values from the time domain. By way of further example, based on the plethysmographic type and motion type, the processor 80 may choose to adjust how patient physiological conditions are derived from the plethysmographic signals. In this regard, the neural network classifier can be trained on abnormal signals (e.g., extreme arrhythmia's and different heart conditions) so that it will be able to alert medical personnel to occurrence of unusual waveforms possibly due to the onset of a critical physiological condition. Referring now to FIG. 3, there is shown a schematic representation of one embodiment of a neural network 300 architecture that may be employed for plethysmographic recognition, respiration activity and motion estimation in accordance with the method (200) illustrated in FIG. 2. The neural network 300 includes an input layer 302, one or more hidden layers 304, and an output classification layer 306. The input layer 302 receives input values in the form one or more frames 308 of input values produced at different times from the current frame tO backwards to frame t-n, where n is an integer. As depicted, in one embodiment there are three frames (tO, t-1, t-2). The time interval between frames 308 may be linear, as depicted, or it may be non-linear. Each frame 308 may be made up of input values from a variety of signal transforms (e.g., cepstral, bispectral, Wavelet, power spectral) as well as additional signal characteristics, if any. Each value in the frames 308 activates a separate node (also referred to herein as neuron) in the neural network 300 input layer 302. Each node in the input layer 302 activates each node in the first hidden layer 304 interconnected thereto. There may be one or more hidden layers 304. Each node in the last hidden layer 304 activates each node in the output classification layer 306 interconnected thereto. The output classification layer 306 includes a number of output nodes. The total number of output nodes typically corresponds to the number of different types of classification required. In this regard, the output nodes may generally be classified as being plethysmographic type output nodes 310 or motion type output nodes 312. The output nodes provide indications as to the presence of the particular type of plethysmographic signal or motion in the red and infrared plethysmographic signals by firing at various levels. A level approaching a specified high value (e.g., 1) indicates a strong presence of the particular type of plethysmographic signal or motion with which the node is associated and a level approaching a specified low value (e.g., 0) indicates a weak presence of the particular type of plethysmographic signal or motion with which the node is associated. Thus, the red and infrared plethysmographic signals are classified by the various levels of the plethysmographic type and motion type output nodes 310, 312. In addition to the plethysmographic signal type and motion type output nodes 310, 312, there may also be one or more respiration index output nodes 314. For example, there may be a respiration rate output node and a respiration depth output node. The values (e.g., from 0 to 1) of the respiration rate and depth nodes comprise the respiration index of the patient. In order to obtain a respiration index, comparing the Wavelet filter bank coefficients from both the red and infrared input signals over time may offer the neural network 300 the required information necessary to determine respiration activity, since the comparison (which the neural network 300 may do in a number of ways) can provide Sp02 estimates. It is the pattern of such Sp02 estimates over time that provides information to build a respiration index. Similarly allowing the neural network 300 to compare the other spectral transforms from both the red and infrared input signals assists in extracting information that provides Sp02 estimates. The neural network 300 architecture as described is a feed-forward network, and when there is only a single hidden layer 304, it is a three-layer feed-forward network. Another form of the neural network 300 can be configured by adding unit delay operators 316 at the hidden layer 304. The schematic connections of the unit delay operators 316 are depicted in dashed lines to indicate that they may be optionally included in order to achieve a recursive neural network 300 architecture. Regardless of the neural network 300 architecture, the neural network 300 should be trained prior to use in order to establish the appropriate weights applied to the various interconnections between the various nodes. One method of training the neural network is the backpropagation algorithm. The backpropagation algorithm may be supervised by a fuzzy logic controller processing module. The fuzzy logic controller is referenced in FIG. 2 as 280 and is depicted in dashed lines to indicate that it may optionally be included when training the neural network 300. The fuzzy logic controller (280) helps speed up the convergence of the backpropagation learning, and in some cases enables convergence to take place, by controlling the learning parameters with a fuzzy rule set. Essentially some heuristics are applied dependent on the change of error (CE - related to error gradient) and the change of CE (CCE - related to second-order error gradient). This has been shown to accelerate learning. The input to the neural network 300 may also be "fuzzified" by slight perturbing and reordering input values. Such a technique can increase the robustness of the neural network 300 to recognize patterns in noise. Referring now to FIG. 5, there is shown one embodiment of a method (500) for training the neural network. The neural network training method (500) establishes the appropriate weighting values for the interconnections between the neurons in the neural network architecture. In one embodiment, the neural network training method (500) is implemented on a computer system (e.g., a desktop or laptop computer). After a trained neural network is generated on the computer, the trained neural network can then be installed on a number of pulse oximeters. Although not required, such centralized training of the neural network and installation of the same neural network on multiple pulse oximeters provides for enhanced consistency among the recognition and classification of various plethysmographic signals by different pulse oximeters. The neural network training method (500) begins with selecting (510) a number of plethysmographic signal data sets that are associated with a number of different types of predetermined signal conditions. The plethysmographic signal data sets may be stored in a database of exemplary plethysmographic signal data sets that is stored on a hard drive of the computer or other storage media accessible by the computer. By way of example, different types of predetermined signal conditions include: (1) normal adult plethysmographic signal; (2) normal baby plethysmographic signal; (3) fixed frequency tapping motion without plethysmographic signal; (4) fixed frequency tapping motion with plethysmographic signal; (5) moving frequency tapping motion without plethysmographic signal; (6) moving frequency tapping motion with plethysmographic signal; (7) clenching motion without plethysmographic signal; (8) clenching motion with plethysmographic signal; (9) baby-kicking motion without plethysmographic signal; (10) baby-kicking motion with plethysmographic signal; (11) irregular plethysmographic signal associated with patient arrhythmia; (12) low perfusion condition plethysmographic signal; (13) plethysmographic signal with fast breathing; (14) plethysmographic signal with shallow breathing; (15) plethysmographic signal with extended breath holding; and (16) no plethysmographic signal. Examples of a number of different red and infrared plethysmographic signal conditions are depicted in FIGS. 6A-D. Typically, the plethysmographic signal data sets are stored in the form of time domain signals. Thus, the selected plethysmographic signal data sets are transformed (520) from the time domain to the other signal domains (e.g., cepstral, power spectral, bispectral, Wavelet filtered domains) to be utilized as inputs to the neural network processing module (250). Transformation (520) results in a corresponding number of transformed plethysmographic signal data sets. Thereafter, sets or vectors of coefficients are extracted (530) from the transformed plethysmographic signal data sets. As may be appreciated, the steps of transforming (520) and extracting (530) may, for example, be accomplished as described in connection with the transforming step (240) of the method (200) of FIG. 2. In addition to the extracted coefficients, additional signal characteristics may be derived (540) from the plethysmographic signal data sets and the various transformed signal domains. The additional signal characteristics that are derived should be the same ones as are intended to be employed by the neural network processing module (250) in addition to the various signal domain coefficients. The sets or vectors of extracted coefficients and the additional signal characteristics are then used (550) as inputs to train the neural network. A learning procedure is used to adjust (560) the weighting values associated with the various interconnections between neurons in the neural network until the weighting values are optimized. In this regard, learning procedures such as the backpropagation and simulated annealing may be employed. In some instances, the trained neural network may have difficultly recognizing and classifying a particular plethysmographic signal encountered in the field. In such instances, it may be desirable to add such plethysmographic signals (e.g., new plethysmographic signals of interest) to the set of training signals used to train a neural network for a new set of pulse oximeters or to retrain the neural network installed in existing pulse oximeters. When encountered in the field, new plethysmographic signals of interest may be stored in the memory device(s) 82 of the pulse oximeter(s) 10. Thereafter, new plethysmographic signals of interest may be retrieved (570) from the memory device(s) 82 of the pulse oximeter(s) and added (580) to the database of plethysmographic signal data sets. While various embodiments of the present invention have been described in detail, further modifications and adaptations of the invention may occur to those skilled in the art. However, it is to be expressly understood that such modifications and adaptations are within the spirit and scope of the present invention.

Claims

What is claimed is: 1. A pulse oximeter comprising: a first optical signal source operable to emit an optical signal characterized by a first wavelength; a second optical signal source operable to emit an optical signal characterized by a second wavelength different than said first wavelength; a detector operable to receive said first and second optical signals after said first and second optical signals are attenuated by a patient tissue site of a patient, said detector being further operable to provide a detector output signal representative of said attenuated first and second optical signals; and a processor enabled to obtain first and second time domain plethysmographic signals from the detector output signal and classify at least one of the first and second time domain plethysmographic signals using a neural network, said neural network receiving input coefficients derived from at least one transform of said at least one of said first and second time domain plethysmographic signals.
2. The pulse oximeter of Claim 1 wherein said at least one of said first and second time domain plethysmographic signals is classified by said neural network as being associated with at least one of a plurality of different types of predetermined signal conditions.
3. The pulse oximeter of Claim 2 wherein the plurality of different types of predetermined signal conditions comprise: ( 1 ) Normal adult plethysmographic signal ; (2) Normal baby plethysmographic signal; (3) Fixed frequency tapping motion without plethysmographic signal; (4) Fixed frequency tapping motion with plethysmographic signal; (5) Moving frequency tapping motion without plethysmographic signal; (6) Moving frequency tapping motion with plethysmographic signal; (7) Clenching motion without plethysmographic signal; (8) Clenching motion with plethysmographic signal; (9) Baby-kicking motion without plethysmographic signal; (10) Baby-kicking motion with plethysmographic signal; (11) Irregular plethysmographic signal associated with patient arrhythmia; (12) Low perfusion condition plethysmographic signal; (13) Plethysmographic signal with fast breathing; (14) Plethysmographic signal with shallow breathing; (15) Plethysmographic signal with extended breath holding; and (16) No plethysmographic signal.
4. The pulse oximeter of Claim 2 wherein said neural network is trained prior to field use with data sets representative of each of said plurality of different types of predetermined signal conditions.
5. The pulse oximeter of Claim 4 wherein said neural network comprises a fuzzy neural network that has been trained prior to field use in a manner achieving fuzzification of said pre-trained neural network.
6. The pulse oximeter of Claim 1 wherein said at least one transform comprises at least one of a spectral transform, a bispectral transform, a cepstral transform, and Wavelet filter bank transform.
7. The pulse oximeter of Claim 6 wherein, in addition to said coefficients, said neural network further receives at least one signal characteristic derived from at least one of the first time domain plethysmographic signal, the second time domain plethysmographic signal, and said at least one transform of said at least one of said first and second time domain plethysmographic signals.
8. The pulse oximeter of Claim 7 wherein said at least one signal characteristic comprises at least one of: (1) An RMS energy measure; (2) A spikiness measure; (3) A spectral jitter measure; (4) A spectral shimmer measure; (5) A spectral smear measure; (6) A cepstral peak jitter measure; and (7) A cepstral peak position measure.
9. The pulse oximeter of Claim 1 wherein said neural network comprises one of a feed-forward network and a recursive network.
10. The pulse oximeter of Claim 9 wherein inputs to an intermediate layer of said neural network are used to fuzzify the input coefficients to an initial layer of said neural network.
11. The pulse oximeter of Claim 1 wherein said first wavelength is within the range of infrared light wavelengths and said second wavelength is within the range of red light wavelengths.
12. The pulse oximeter of Claim 1 wherein said pulse oximeter further comprises: a drive system operable to cause operation of said first and second optical signal sources such that each of said first and second optical signal sources emit first and second optical signals, respectively, in accordance with a multiplexing method; a sampler operable to sample the detector output signal at a desired sampling rate and output a signal having a series of sample values representative of said attenuated first and second optical signals; and wherein said processor comprises a digital processor, said digital processor being further operable to demultiplex the series of sample values to obtain said first and second time domain plethysmographic signals.
13. The pulse oximeter of Claim 12 wherein said desired sampling rate is at least 50 Hz.
14. The pulse oximeter of Claim 12 wherein said multiplexing method comprises at least one of frequency division multiplexing, time division multiplexing, and code division multiplexing.
15. A method of processing a plethysmographic signal obtained from a patient, the plethysmographic signal being obtained in a first signal domain, said method comprising the steps of: transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain to obtain a corresponding plurality of transformed plethysmographic signals, each transformed plethysmographic signal being in one of the different signal domains; selecting a plurality of sets of coefficients, each set of coefficients being derived from a corresponding one of the transformed plethysmographic signals; inputting the sets of coefficients to a neural network; and classifying the plethysmographic signal based on an output from the neural network.
16. The method of Claim 15 wherein the first domain comprises the time domain and wherein said step of transforming comprises: transforming the plethysmographic signal from the time domain to the spectral domain to obtain a spectral domain plethysmographic signal; and transforming the spectral domain plethysmographic signal to the cepstral domain to obtain a cepstral domain plethysmographic signal.
17. The method of Claim 16 wherein said step of transforming the plethysmographic signal from the time domain to the spectral domain comprises: performing a Fourier transformation on the time domain plethysmographic signal.
18. The method of Claim 16 wherein said step of transforming the plethysmographic signal from the spectral domain to the cepstral domain comprises: performing a Fourier transformation on the spectral domain plethysmographic signal.
19. The method of Claim 16 wherein said step of transforming further comprises: transforming the plethysmographic signal from the time domain to the bispectral domain to obtain a bispectral domain plethysmographic signal.
20. The method of Claim 19 wherein said step of transforming the plethysmographic signal from the time domain to the bispectral domain is performed in accordance with the following expression:
B[ωιtω2) = E[F{ω1)F(ω2)F(ωι + ω2)] where ωi and ω2 are the frequencies present in spectrums of the time domain plethysmographic signal.
21. The method of Claim 16 wherein said step of transforming further comprises: applying a Wavelet filter bank transform to the time domain plethysmographic signal to obtain a Wavelet filtered domain plethysmographic signal.
22. The method of Claim 15 further comprising: deriving at least one signal characteristic using information included in at least one of the first domain plethysmographic signal or the transformed plethysmographic signals; and inputting the at least one signal characteristic to the neural network in addition to the sets of coefficients.
23. The method of Claim 22 wherein, in said step of deriving, the at least one signal characteristic comprises at least one of: (1) An RMS energy measure; (2) A spikiness measure; (3) A spectral jitter measure; (4) A spectral shimmer measure; (5) A spectral smear measure; (6) A cepstral peak jitter measure; and (7) A cepstral peak position measure.
24. The method of Claim 15 wherein, in said step of selecting, each set of coefficients corresponds with a plurality of frames of its corresponding transformed plethysmographic signal.
25. The method of Claim 15 wherein said step of classifying comprises: associating the output from the neural network with at least one of a plurality of different types of predetermined signal conditions.
26. The method of Claim 25 wherein, in said step of associating, the plurality of different types of predetermined signal conditions comprise: (1) Normal adult plethysmographic signal; (2) Normal baby plethysmographic signal; (3) Fixed frequency tapping motion without plethysmographic signal; (4) Fixed frequency tapping motion with plethysmographic signal; (5) Moving frequency tapping motion without plethysmographic signal; (6) Moving frequency tapping motion with plethysmographic signal; (7) Clenching motion without plethysmographic signal; (8) Clenching motion with plethysmographic signal; (9) Baby-kicking motion without plethysmographic signal; (10) Baby-kicking motion with plethysmographic signal; (11) Irregular plethysmographic signal associated with patient arrhythmia; (12) Low perfusion condition plethysmographic signal; (13) Plethysmographic signal with fast breathing; (14) Plethysmographic signal with shallow breathing; (15) Plethysmographic signal with extended breath holding; and (16) No plethysmographic signal.
27. The method of Claim 25 further comprising: training the neural network prior to field use with data sets representative of each of said plurality of different types of predetermined signal conditions.
28. The method of Claim 27 wherein, in said step of training, the neural network is trained prior to field use in a manner achieving fuzzification of the neural network.
29. The method of Claim 27 wherein, in said step of training, the neural network is trained in accordance with one of a backpropagation learning procedure and a simulated annealing learning procedure.
30. The method of Claim 29 wherein, in said step of training, the backpropagation learning procedure is implemented with fuzzy logic control.
31. The method of claim 15 further comprising: transmitting an optical signal through a tissue site of the patient to obtain the first domain plethysmographic signal.
32. A method of training a neural network to classify a plethysmographic signal obtained from a patient, said method comprising the steps of: selecting a plurality of first domain plethysmographic signal data sets associated with a plurality of different types of predetermined signal conditions from a database of plethysmographic signal data sets; transforming the first domain plethysmographic signal data sets to other signal domains different than the first domain to obtain a corresponding plurality of transformed plethysmographic signal data sets; extracting a plurality of sets of coefficients from the transformed plethysmographic signal data sets, each set of coefficients being extracted from a corresponding one of the transformed plethysmographic signal data sets; using the sets of extracted coefficients as inputs to the neural network; and adjusting weighting values associated with connections between neurons in the neural network in accordance with a learning procedure.
33. The method of Claim 32 wherein, in said step of selecting, the plurality of different types of predetermined signal conditions comprise: (1) Normal adult plethysmographic signal; (2) Normal baby plethysmographic signal; (3) Fixed frequency tapping motion without plethysmographic signal; (4) Fixed frequency tapping motion with plethysmographic signal; (5) Moving frequency tapping motion without plethysmographic signal; (6) Moving frequency tapping motion with plethysmographic signal; (7) Clenching motion without plethysmographic signal; (8) Clenching motion with plethysmographic signal; (9) Baby-kicking motion without plethysmographic signal; (10) Baby-kicking motion with plethysmographic signal; (11) Irregular plethysmographic signal associated with patient arrhythmia; (12) Low perfusion condition plethysmographic signal; (13) Plethysmographic signal with fast breathing; (14) Plethysmographic signal with shallow breathing; (15) Plethysmographic signal with extended breath holding; and (16) No plethysmographic signal.
34. The method of Claim 32 wherein, in said step of transforming the first domain plethysmographic signal data sets to other signal domains, the signal domains different than the first domain comprise spectral, cepstral, bispectral and Wavelet filter bank signal domains.
35. The method of Claim 32 wherein, in said step of adjusting, the weighting values are adjusted in accordance with one of a backpropagation learning procedure and a simulated annealing learning procedure.
36. The method of Claim 32 further comprising: deriving at least one signal characteristic using information included in at least one of the first domain plethysmographic signal data sets and the transformed plethysmographic signal data sets; and using the at least one signal characteristic as an input to the neural network in addition to the extracted coefficients.
37. The method of Claim 36 wherein, in said step of deriving, the at least one signal characteristic comprises at least one of: (1) An RMS energy measure; (2) A spikiness measure; (3) A spectral jitter measure; (4) A spectral shimmer measure; (5) A spectral smear measure; (6) A cepstral peak jitter measure; and (7) A cepstral peak position measure.
38. The method of Claim 32 further comprising: storing plethysmographic signals of interest.
39. The method of Claim 38 wherein, in said step of storing, the plethysmographic signal of interest is stored in a memory device of a pulse oximeter.
40. The method of Claim 38 further comprising: retrieving the stored plethysmographic signal of interest; and adding the plethysmographic signal of interest to the database of plethysmographic signal data sets.
41. A method of providing information relating to a physiological condition of a patient based on at least one plethysmographic signal obtained from the patient, the plethysmographic signal being obtained in a first signal domain, said method comprising the steps of: transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain to obtain a corresponding plurality of transformed plethysmographic signals, each transformed plethysmographic signal being in one of the different signal domains; classifying the plethysmographic signal based on an output from a neural network, wherein the output of the neural network is based on input coefficients derived from at least one of the transformed plethysmographic signals; and selecting a technique for determining the physiological condition of the patient based on the classification.
42. The method of Claim 41 wherein the physiological condition of the patient comprises a pulse rate.
43. The method of Claim 41 wherein at least two plethysmographic signals corresponding to different optical wavelengths are transformed and classified, and wherein the physiological condition of the patient comprises an Sp02 value.
44. The method of Claim 41 wherein at least two plethysmographic signals corresponding to different optical wavelengths are transformed and classified, the plethysmographic signals being transformed using at least a Wavelet filter bank transform, and wherein the physiological condition of the patient comprises a respiration index.
45. The method of Claim 41 wherein the first domain comprise the time domain and the plurality of signal domains different than the first domain comprise spectral, cepstral, bispectral and Wavelet filtered domains.
PCT/US2004/027973 2003-08-27 2004-08-27 Multi-domain motion estimation and plethysmographic recognition using fuzzy neural-nets WO2005020798A2 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US49803203P 2003-08-27 2003-08-27
US60/498,032 2003-08-27

Publications (2)

Publication Number Publication Date
WO2005020798A2 true WO2005020798A2 (en) 2005-03-10
WO2005020798A3 WO2005020798A3 (en) 2005-07-14

Family

ID=34272632

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2004/027973 WO2005020798A2 (en) 2003-08-27 2004-08-27 Multi-domain motion estimation and plethysmographic recognition using fuzzy neural-nets

Country Status (2)

Country Link
US (1) US6931269B2 (en)
WO (1) WO2005020798A2 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2006134197A1 (en) * 2005-06-13 2006-12-21 Consejo Superior De Investigaciones Científicas Method of processing photopletismographic signals obtained from a person or animal and oximeter using said method
WO2011041216A1 (en) * 2009-09-30 2011-04-07 Nellcor Puritan Bennett Llc Systems and method for identifying non-corrupted signal segments for use in determining physiological parameters
CN105380630A (en) * 2014-08-22 2016-03-09 苹果公司 Heart rate path optimizer
CN106798552A (en) * 2015-11-25 2017-06-06 德克萨斯仪器股份有限公司 Heart rate with status switch optimization estimates equipment
US10758185B2 (en) 2015-11-25 2020-09-01 Texas Instruments Incorporated Heart rate estimation apparatus using digital automatic gain control

Families Citing this family (218)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9468378B2 (en) 1997-01-27 2016-10-18 Lawrence A. Lynn Airway instability detection system and method
US8932227B2 (en) 2000-07-28 2015-01-13 Lawrence A. Lynn System and method for CO2 and oximetry integration
US9042952B2 (en) 1997-01-27 2015-05-26 Lawrence A. Lynn System and method for automatic detection of a plurality of SPO2 time series pattern types
US6024089A (en) 1997-03-14 2000-02-15 Nelcor Puritan Bennett Incorporated System and method for setting and displaying ventilator alarms
US20070191697A1 (en) 2006-02-10 2007-08-16 Lynn Lawrence A System and method for SPO2 instability detection and quantification
US9521971B2 (en) 1997-07-14 2016-12-20 Lawrence A. Lynn System and method for automatic detection of a plurality of SPO2 time series pattern types
US6675031B1 (en) 1999-04-14 2004-01-06 Mallinckrodt Inc. Method and circuit for indicating quality and accuracy of physiological measurements
US20060195041A1 (en) 2002-05-17 2006-08-31 Lynn Lawrence A Centralized hospital monitoring system for automatically detecting upper airway instability and for preventing and aborting adverse drug reactions
US9053222B2 (en) 2002-05-17 2015-06-09 Lawrence A. Lynn Patient safety processor
US6748254B2 (en) 2001-10-12 2004-06-08 Nellcor Puritan Bennett Incorporated Stacked adhesive optical sensor
US7190986B1 (en) 2002-10-18 2007-03-13 Nellcor Puritan Bennett Inc. Non-adhesive oximeter sensor for sensitive skin
ATE521279T1 (en) 2003-02-27 2011-09-15 Nellcor Puritan Bennett Ie METHOD AND DEVICE FOR EVALUATION AND PROCESSING PHOTOPLETHYSMOGRAPHIC SIGNALS BY WAVE TRANSFORMATION ANALYSIS
US7222624B2 (en) * 2004-07-02 2007-05-29 Praxair Technology, Inc. Dual sensor oxygen therapy device
US7596398B2 (en) 2005-03-01 2009-09-29 Masimo Laboratories, Inc. Multiple wavelength sensor attachment
US20070078311A1 (en) * 2005-03-01 2007-04-05 Ammar Al-Ali Disposable multiple wavelength optical sensor
US7392075B2 (en) 2005-03-03 2008-06-24 Nellcor Puritan Bennett Incorporated Method for enhancing pulse oximetry calculations in the presence of correlated artifacts
US8597193B2 (en) 2005-05-06 2013-12-03 Vasonova, Inc. Apparatus and method for endovascular device guiding and positioning using physiological parameters
US20090118612A1 (en) 2005-05-06 2009-05-07 Sorin Grunwald Apparatus and Method for Vascular Access
US9204819B2 (en) 2005-05-06 2015-12-08 Vasonova, Inc. Endovenous access and guidance system utilizing non-image based ultrasound
US7657295B2 (en) 2005-08-08 2010-02-02 Nellcor Puritan Bennett Llc Medical sensor and technique for using the same
US7657294B2 (en) 2005-08-08 2010-02-02 Nellcor Puritan Bennett Llc Compliant diaphragm medical sensor and technique for using the same
US7590439B2 (en) 2005-08-08 2009-09-15 Nellcor Puritan Bennett Llc Bi-stable medical sensor and technique for using the same
US20070060808A1 (en) 2005-09-12 2007-03-15 Carine Hoarau Medical sensor for reducing motion artifacts and technique for using the same
US8092379B2 (en) 2005-09-29 2012-01-10 Nellcor Puritan Bennett Llc Method and system for determining when to reposition a physiological sensor
US7899510B2 (en) 2005-09-29 2011-03-01 Nellcor Puritan Bennett Llc Medical sensor and technique for using the same
US7725146B2 (en) 2005-09-29 2010-05-25 Nellcor Puritan Bennett Llc System and method for pre-processing waveforms
US7869850B2 (en) 2005-09-29 2011-01-11 Nellcor Puritan Bennett Llc Medical sensor for reducing motion artifacts and technique for using the same
US7725147B2 (en) 2005-09-29 2010-05-25 Nellcor Puritan Bennett Llc System and method for removing artifacts from waveforms
US7904130B2 (en) 2005-09-29 2011-03-08 Nellcor Puritan Bennett Llc Medical sensor and technique for using the same
US7555327B2 (en) 2005-09-30 2009-06-30 Nellcor Puritan Bennett Llc Folding medical sensor and technique for using the same
US20070077200A1 (en) * 2005-09-30 2007-04-05 Baker Clark R Method and system for controlled maintenance of hypoxia for therapeutic or diagnostic purposes
US7483731B2 (en) 2005-09-30 2009-01-27 Nellcor Puritan Bennett Llc Medical sensor and technique for using the same
US7486979B2 (en) 2005-09-30 2009-02-03 Nellcor Puritan Bennett Llc Optically aligned pulse oximetry sensor and technique for using the same
US8062221B2 (en) 2005-09-30 2011-11-22 Nellcor Puritan Bennett Llc Sensor for tissue gas detection and technique for using the same
US7881762B2 (en) 2005-09-30 2011-02-01 Nellcor Puritan Bennett Llc Clip-style medical sensor and technique for using the same
US8233954B2 (en) 2005-09-30 2012-07-31 Nellcor Puritan Bennett Llc Mucosal sensor for the assessment of tissue and blood constituents and technique for using the same
US7668579B2 (en) 2006-02-10 2010-02-23 Lynn Lawrence A System and method for the detection of physiologic response to stimulation
US8021310B2 (en) 2006-04-21 2011-09-20 Nellcor Puritan Bennett Llc Work of breathing display for a ventilation system
US8073518B2 (en) 2006-05-02 2011-12-06 Nellcor Puritan Bennett Llc Clip-style medical sensor and technique for using the same
US8145288B2 (en) 2006-08-22 2012-03-27 Nellcor Puritan Bennett Llc Medical sensor for reducing signal artifacts and technique for using the same
US8219170B2 (en) 2006-09-20 2012-07-10 Nellcor Puritan Bennett Llc System and method for practicing spectrophotometry using light emitting nanostructure devices
US8359079B2 (en) * 2006-09-21 2013-01-22 Starr Life Sciences Corporation Pulse oximetry system and techniques for deriving cardiac and breathing parameters from extra-thoracic blood flow measurements
US20080076991A1 (en) * 2006-09-21 2008-03-27 Starr Life Sciences Corp. Medical display devices for cardiac and breathing parameters derived from extra-thoracic blood flow measurements
US7922666B2 (en) * 2006-09-21 2011-04-12 Starr Life Sciences Corporation Pulse oximeter based techniques for controlling anesthesia levels and ventilation levels in subjects
US8175671B2 (en) 2006-09-22 2012-05-08 Nellcor Puritan Bennett Llc Medical sensor for reducing signal artifacts and technique for using the same
US8396527B2 (en) 2006-09-22 2013-03-12 Covidien Lp Medical sensor for reducing signal artifacts and technique for using the same
US8190224B2 (en) 2006-09-22 2012-05-29 Nellcor Puritan Bennett Llc Medical sensor for reducing signal artifacts and technique for using the same
US7784461B2 (en) 2006-09-26 2010-08-31 Nellcor Puritan Bennett Llc Three-dimensional waveform display for a breathing assistance system
US7869849B2 (en) 2006-09-26 2011-01-11 Nellcor Puritan Bennett Llc Opaque, electrically nonconductive region on a medical sensor
US7574245B2 (en) 2006-09-27 2009-08-11 Nellcor Puritan Bennett Llc Flexible medical sensor enclosure
US7890153B2 (en) 2006-09-28 2011-02-15 Nellcor Puritan Bennett Llc System and method for mitigating interference in pulse oximetry
US7796403B2 (en) 2006-09-28 2010-09-14 Nellcor Puritan Bennett Llc Means for mechanical registration and mechanical-electrical coupling of a faraday shield to a photodetector and an electrical circuit
US7680522B2 (en) 2006-09-29 2010-03-16 Nellcor Puritan Bennett Llc Method and apparatus for detecting misapplied sensors
US8175667B2 (en) 2006-09-29 2012-05-08 Nellcor Puritan Bennett Llc Symmetric LED array for pulse oximetry
US7476131B2 (en) 2006-09-29 2009-01-13 Nellcor Puritan Bennett Llc Device for reducing crosstalk
US8160668B2 (en) * 2006-09-29 2012-04-17 Nellcor Puritan Bennett Llc Pathological condition detector using kernel methods and oximeters
US7684842B2 (en) 2006-09-29 2010-03-23 Nellcor Puritan Bennett Llc System and method for preventing sensor misuse
US8068891B2 (en) 2006-09-29 2011-11-29 Nellcor Puritan Bennett Llc Symmetric LED array for pulse oximetry
WO2008055173A2 (en) * 2006-10-30 2008-05-08 Cox Paul G Methods and systems for non-invasive, internal hemorrhage detection
US7496453B2 (en) * 2006-11-07 2009-02-24 The Hong Kong Polytechnic University Classification of herbal medicines using wavelet transform
US20080221418A1 (en) * 2007-03-09 2008-09-11 Masimo Corporation Noninvasive multi-parameter patient monitor
US8280469B2 (en) 2007-03-09 2012-10-02 Nellcor Puritan Bennett Llc Method for detection of aberrant tissue spectra
US8265724B2 (en) 2007-03-09 2012-09-11 Nellcor Puritan Bennett Llc Cancellation of light shunting
US7894869B2 (en) 2007-03-09 2011-02-22 Nellcor Puritan Bennett Llc Multiple configuration medical sensor and technique for using the same
EP2476369B1 (en) 2007-03-27 2014-10-01 Masimo Laboratories, Inc. Multiple wavelength optical sensor
GB2447940B (en) * 2007-03-28 2012-04-04 Thales Holdings Uk Plc A motion classification device
US20080262326A1 (en) * 2007-04-19 2008-10-23 Starr Life Sciences Corp. Signal Processing Method and Apparatus for Processing a Physiologic Signal such as a Photoplethysmography Signal
US8374665B2 (en) 2007-04-21 2013-02-12 Cercacor Laboratories, Inc. Tissue profile wellness monitor
KR100883185B1 (en) 2007-06-29 2009-02-13 장형종 System and Method for Detecting Abnormality of Biosignal using Neural Network with Weighted Fuzzy Membership Funtions
US8346328B2 (en) 2007-12-21 2013-01-01 Covidien Lp Medical sensor and technique for using the same
US8352004B2 (en) 2007-12-21 2013-01-08 Covidien Lp Medical sensor and technique for using the same
US8366613B2 (en) 2007-12-26 2013-02-05 Covidien Lp LED drive circuit for pulse oximetry and method for using same
US8577434B2 (en) 2007-12-27 2013-11-05 Covidien Lp Coaxial LED light sources
US8452364B2 (en) 2007-12-28 2013-05-28 Covidien LLP System and method for attaching a sensor to a patient's skin
US8442608B2 (en) 2007-12-28 2013-05-14 Covidien Lp System and method for estimating physiological parameters by deconvolving artifacts
US20090171171A1 (en) * 2007-12-31 2009-07-02 Nellcor Puritan Bennett Llc Oximetry sensor overmolding location features
US8199007B2 (en) 2007-12-31 2012-06-12 Nellcor Puritan Bennett Llc Flex circuit snap track for a biometric sensor
US8070508B2 (en) 2007-12-31 2011-12-06 Nellcor Puritan Bennett Llc Method and apparatus for aligning and securing a cable strain relief
US8897850B2 (en) 2007-12-31 2014-11-25 Covidien Lp Sensor with integrated living hinge and spring
US20090171173A1 (en) * 2007-12-31 2009-07-02 Nellcor Puritan Bennett Llc System and method for reducing motion artifacts in a sensor
US8092993B2 (en) 2007-12-31 2012-01-10 Nellcor Puritan Bennett Llc Hydrogel thin film for use as a biosensor
US8437822B2 (en) 2008-03-28 2013-05-07 Covidien Lp System and method for estimating blood analyte concentration
US8112375B2 (en) 2008-03-31 2012-02-07 Nellcor Puritan Bennett Llc Wavelength selection and outlier detection in reduced rank linear models
EP2283443A1 (en) 2008-05-07 2011-02-16 Lynn, Lawrence A. Medical failure pattern search engine
US20090326402A1 (en) * 2008-06-30 2009-12-31 Nellcor Puritan Bennett Ireland Systems and methods for determining effort
US20090324033A1 (en) * 2008-06-30 2009-12-31 Nellcor Puritan Bennett Ireland Signal Processing Systems and Methods for Determining Slope Using an Origin Point
US8862194B2 (en) * 2008-06-30 2014-10-14 Covidien Lp Method for improved oxygen saturation estimation in the presence of noise
US7887345B2 (en) 2008-06-30 2011-02-15 Nellcor Puritan Bennett Llc Single use connector for pulse oximetry sensors
US8660799B2 (en) 2008-06-30 2014-02-25 Nellcor Puritan Bennett Ireland Processing and detecting baseline changes in signals
US8071935B2 (en) 2008-06-30 2011-12-06 Nellcor Puritan Bennett Llc Optical detector with an overmolded faraday shield
US8077297B2 (en) 2008-06-30 2011-12-13 Nellcor Puritan Bennett Ireland Methods and systems for discriminating bands in scalograms
US7880884B2 (en) 2008-06-30 2011-02-01 Nellcor Puritan Bennett Llc System and method for coating and shielding electronic sensor components
US8295567B2 (en) 2008-06-30 2012-10-23 Nellcor Puritan Bennett Ireland Systems and methods for ridge selection in scalograms of signals
US8827917B2 (en) * 2008-06-30 2014-09-09 Nelleor Puritan Bennett Ireland Systems and methods for artifact detection in signals
US7944551B2 (en) * 2008-06-30 2011-05-17 Nellcor Puritan Bennett Ireland Systems and methods for a wavelet transform viewer
EP2326239B1 (en) 2008-07-03 2017-06-21 Masimo Laboratories, Inc. Protrusion for improving spectroscopic measurement of blood constituents
US8679027B2 (en) 2008-07-15 2014-03-25 Nellcor Puritan Bennett Ireland Systems and methods for pulse processing
US8082110B2 (en) 2008-07-15 2011-12-20 Nellcor Puritan Bennett Ireland Low perfusion signal processing systems and methods
US8761855B2 (en) 2008-07-15 2014-06-24 Nellcor Puritan Bennett Ireland Systems and methods for determining oxygen saturation
US8358213B2 (en) 2008-07-15 2013-01-22 Covidien Lp Systems and methods for evaluating a physiological condition using a wavelet transform and identifying a band within a generated scalogram
US20100016692A1 (en) * 2008-07-15 2010-01-21 Nellcor Puritan Bennett Ireland Systems and methods for computing a physiological parameter using continuous wavelet transforms
US8285352B2 (en) 2008-07-15 2012-10-09 Nellcor Puritan Bennett Llc Systems and methods for identifying pulse rates
US8226568B2 (en) * 2008-07-15 2012-07-24 Nellcor Puritan Bennett Llc Signal processing systems and methods using basis functions and wavelet transforms
US20100016676A1 (en) 2008-07-15 2010-01-21 Nellcor Puritan Bennett Ireland Systems And Methods For Adaptively Filtering Signals
US8370080B2 (en) * 2008-07-15 2013-02-05 Nellcor Puritan Bennett Ireland Methods and systems for determining whether to trigger an alarm
US8506498B2 (en) 2008-07-15 2013-08-13 Nellcor Puritan Bennett Ireland Systems and methods using induced perturbation to determine physiological parameters
US8385675B2 (en) * 2008-07-15 2013-02-26 Nellcor Puritan Bennett Ireland Systems and methods for filtering a signal using a continuous wavelet transform
US8660625B2 (en) * 2008-07-15 2014-02-25 Covidien Lp Signal processing systems and methods for analyzing multiparameter spaces to determine physiological states
US8203704B2 (en) 2008-08-04 2012-06-19 Cercacor Laboratories, Inc. Multi-stream sensor for noninvasive measurement of blood constituents
US8364220B2 (en) 2008-09-25 2013-01-29 Covidien Lp Medical sensor and technique for using the same
US8410951B2 (en) 2008-09-30 2013-04-02 Covidien Lp Detecting a signal quality decrease in a measurement system
US8696585B2 (en) * 2008-09-30 2014-04-15 Nellcor Puritan Bennett Ireland Detecting a probe-off event in a measurement system
US8914088B2 (en) 2008-09-30 2014-12-16 Covidien Lp Medical sensor and technique for using the same
US8423112B2 (en) 2008-09-30 2013-04-16 Covidien Lp Medical sensor and technique for using the same
US8417309B2 (en) 2008-09-30 2013-04-09 Covidien Lp Medical sensor
US9155493B2 (en) 2008-10-03 2015-10-13 Nellcor Puritan Bennett Ireland Methods and apparatus for calibrating respiratory effort from photoplethysmograph signals
US20100087714A1 (en) * 2008-10-03 2010-04-08 Nellcor Puritan Bennett Ireland Reducing cross-talk in a measurement system
US9011347B2 (en) 2008-10-03 2015-04-21 Nellcor Puritan Bennett Ireland Methods and apparatus for determining breathing effort characteristics measures
WO2010053845A1 (en) * 2008-11-05 2010-05-14 Nellcor Puritan Bennett Llc System and method for facilitating observation of monitored physiologic data
US8788002B2 (en) 2009-02-25 2014-07-22 Valencell, Inc. Light-guiding devices and monitoring devices incorporating same
WO2010098912A2 (en) 2009-02-25 2010-09-02 Valencell, Inc. Light-guiding devices and monitoring devices incorporating same
US8452366B2 (en) 2009-03-16 2013-05-28 Covidien Lp Medical monitoring device with flexible circuitry
US8221319B2 (en) 2009-03-25 2012-07-17 Nellcor Puritan Bennett Llc Medical device for assessing intravascular blood volume and technique for using the same
US8509869B2 (en) 2009-05-15 2013-08-13 Covidien Lp Method and apparatus for detecting and analyzing variations in a physiologic parameter
US20100298728A1 (en) * 2009-05-20 2010-11-25 Nellcor Puritan Bennett Ireland Signal Processing Techniques For Determining Signal Quality Using A Wavelet Transform Ratio Surface
US8364225B2 (en) * 2009-05-20 2013-01-29 Nellcor Puritan Bennett Ireland Estimating transform values using signal estimates
US8634891B2 (en) 2009-05-20 2014-01-21 Covidien Lp Method and system for self regulation of sensor component contact pressure
US8444570B2 (en) * 2009-06-09 2013-05-21 Nellcor Puritan Bennett Ireland Signal processing techniques for aiding the interpretation of respiration signals
US20100324827A1 (en) * 2009-06-18 2010-12-23 Nellcor Puritan Bennett Ireland Fluid Responsiveness Measure
US20100331716A1 (en) * 2009-06-26 2010-12-30 Nellcor Puritan Bennett Ireland Methods and apparatus for measuring respiratory function using an effort signal
US8290730B2 (en) 2009-06-30 2012-10-16 Nellcor Puritan Bennett Ireland Systems and methods for assessing measurements in physiological monitoring devices
US8311601B2 (en) 2009-06-30 2012-11-13 Nellcor Puritan Bennett Llc Reflectance and/or transmissive pulse oximeter
US20100331715A1 (en) * 2009-06-30 2010-12-30 Nellcor Puritan Bennett Ireland Systems and methods for detecting effort events
US9010634B2 (en) 2009-06-30 2015-04-21 Covidien Lp System and method for linking patient data to a patient and providing sensor quality assurance
US8505821B2 (en) 2009-06-30 2013-08-13 Covidien Lp System and method for providing sensor quality assurance
US8636667B2 (en) 2009-07-06 2014-01-28 Nellcor Puritan Bennett Ireland Systems and methods for processing physiological signals in wavelet space
US8391941B2 (en) 2009-07-17 2013-03-05 Covidien Lp System and method for memory switching for multiple configuration medical sensor
US20110021941A1 (en) * 2009-07-23 2011-01-27 Nellcor Puritan Bennett Ireland Systems and methods for respiration monitoring
US20110021892A1 (en) * 2009-07-23 2011-01-27 Nellcor Puritan Bennett Ireland Systems and methods for respiration monitoring
US8478376B2 (en) * 2009-07-30 2013-07-02 Nellcor Puritan Bennett Ireland Systems and methods for determining physiological information using selective transform data
US8346333B2 (en) 2009-07-30 2013-01-01 Nellcor Puritan Bennett Ireland Systems and methods for estimating values of a continuous wavelet transform
US8594759B2 (en) * 2009-07-30 2013-11-26 Nellcor Puritan Bennett Ireland Systems and methods for resolving the continuous wavelet transform of a signal
US8628477B2 (en) 2009-07-31 2014-01-14 Nellcor Puritan Bennett Ireland Systems and methods for non-invasive determination of blood pressure
US8755854B2 (en) 2009-07-31 2014-06-17 Nellcor Puritan Bennett Ireland Methods and apparatus for producing and using lightly filtered photoplethysmograph signals
US8417310B2 (en) 2009-08-10 2013-04-09 Covidien Lp Digital switching in multi-site sensor
US8428675B2 (en) 2009-08-19 2013-04-23 Covidien Lp Nanofiber adhesives used in medical devices
US8923945B2 (en) * 2009-09-24 2014-12-30 Covidien Lp Determination of a physiological parameter
US8840562B2 (en) * 2009-09-24 2014-09-23 Covidien Lp Signal processing warping technique
EP2480997A2 (en) * 2009-09-24 2012-08-01 Nellcor Puritan Bennett LLC Determination of a physiological parameter
US8400149B2 (en) * 2009-09-25 2013-03-19 Nellcor Puritan Bennett Ireland Systems and methods for gating an imaging device
US20110098933A1 (en) * 2009-10-26 2011-04-28 Nellcor Puritan Bennett Ireland Systems And Methods For Processing Oximetry Signals Using Least Median Squares Techniques
US9839381B1 (en) 2009-11-24 2017-12-12 Cercacor Laboratories, Inc. Physiological measurement system with automatic wavelength adjustment
US9119925B2 (en) 2009-12-04 2015-09-01 Covidien Lp Quick initiation of respiratory support via a ventilator user interface
DE112010004682T5 (en) 2009-12-04 2013-03-28 Masimo Corporation Calibration for multi-level physiological monitors
US8335992B2 (en) 2009-12-04 2012-12-18 Nellcor Puritan Bennett Llc Visual indication of settings changes on a ventilator graphical user interface
US8924878B2 (en) 2009-12-04 2014-12-30 Covidien Lp Display and access to settings on a ventilator graphical user interface
US8499252B2 (en) 2009-12-18 2013-07-30 Covidien Lp Display of respiratory data graphs on a ventilator graphical user interface
US9262588B2 (en) 2009-12-18 2016-02-16 Covidien Lp Display of respiratory data graphs on a ventilator graphical user interface
US9050043B2 (en) 2010-05-04 2015-06-09 Nellcor Puritan Bennett Ireland Systems and methods for wavelet transform scale-dependent multiple-archetyping
US8834378B2 (en) 2010-07-30 2014-09-16 Nellcor Puritan Bennett Ireland Systems and methods for determining respiratory effort
EP2637568B1 (en) 2010-11-08 2017-04-12 Vasonova, Inc. Endovascular navigation system
US8521247B2 (en) 2010-12-29 2013-08-27 Covidien Lp Certification apparatus and method for a medical device computer
US9113830B2 (en) 2011-05-31 2015-08-25 Nellcor Puritan Bennett Ireland Systems and methods for detecting and monitoring arrhythmias using the PPG
US9597022B2 (en) 2011-09-09 2017-03-21 Nellcor Puritan Bennett Ireland Venous oxygen saturation systems and methods
US9675274B2 (en) 2011-09-23 2017-06-13 Nellcor Puritan Bennett Ireland Systems and methods for determining respiration information from a photoplethysmograph
US8880576B2 (en) 2011-09-23 2014-11-04 Nellcor Puritan Bennett Ireland Systems and methods for determining respiration information from a photoplethysmograph
US9693709B2 (en) 2011-09-23 2017-07-04 Nellcot Puritan Bennett Ireland Systems and methods for determining respiration information from a photoplethysmograph
US9402554B2 (en) 2011-09-23 2016-08-02 Nellcor Puritan Bennett Ireland Systems and methods for determining respiration information from a photoplethysmograph
US9119597B2 (en) 2011-09-23 2015-09-01 Nellcor Puritan Bennett Ireland Systems and methods for determining respiration information from a photoplethysmograph
US8755871B2 (en) 2011-11-30 2014-06-17 Covidien Lp Systems and methods for detecting arrhythmia from a physiological signal
US9693736B2 (en) 2011-11-30 2017-07-04 Nellcor Puritan Bennett Ireland Systems and methods for determining respiration information using historical distribution
US9247896B2 (en) 2012-01-04 2016-02-02 Nellcor Puritan Bennett Ireland Systems and methods for determining respiration information using phase locked loop
US8880155B2 (en) 2012-02-24 2014-11-04 Covidien Lp Hypovolemia diagnosis technique
US9179876B2 (en) 2012-04-30 2015-11-10 Nellcor Puritan Bennett Ireland Systems and methods for identifying portions of a physiological signal usable for determining physiological information
US9345447B2 (en) 2012-05-07 2016-05-24 Vasonova, Inc. Right atrium indicator
EP2849635A4 (en) * 2012-05-18 2016-01-20 Univ Florida PATIENT IN-TtHE-LOOP PARTICIPATORY CARE AND MONITORING
US10362967B2 (en) 2012-07-09 2019-07-30 Covidien Lp Systems and methods for missed breath detection and indication
US9560978B2 (en) 2013-02-05 2017-02-07 Covidien Lp Systems and methods for determining respiration information from a physiological signal using amplitude demodulation
US9554712B2 (en) 2013-02-27 2017-01-31 Covidien Lp Systems and methods for generating an artificial photoplethysmograph signal
US9687159B2 (en) 2013-02-27 2017-06-27 Covidien Lp Systems and methods for determining physiological information by identifying fiducial points in a physiological signal
US9276031B2 (en) 2013-03-04 2016-03-01 Apple Inc. Photodiode with different electric potential regions for image sensors
US9741754B2 (en) 2013-03-06 2017-08-22 Apple Inc. Charge transfer circuit with storage nodes in image sensors
US9974468B2 (en) 2013-03-15 2018-05-22 Covidien Lp Systems and methods for identifying a medically monitored patient
EP2792297B1 (en) * 2013-04-15 2016-03-02 Stichting IMEC Nederland System and method for the analysis of biopotential signals using motion artifact removal techniques
US10022068B2 (en) 2013-10-28 2018-07-17 Covidien Lp Systems and methods for detecting held breath events
US9596423B1 (en) 2013-11-21 2017-03-14 Apple Inc. Charge summing in an image sensor
US20150148632A1 (en) * 2013-11-26 2015-05-28 David Alan Benaron Calorie Monitoring Sensor And Method For Cell Phones, Smart Watches, Occupancy Sensors, And Wearables
US9473706B2 (en) 2013-12-09 2016-10-18 Apple Inc. Image sensor flicker detection
WO2015102588A1 (en) * 2013-12-30 2015-07-09 Apple Inc. User identification system based on plethysmography
US9848820B2 (en) 2014-01-07 2017-12-26 Covidien Lp Apnea analysis system and method
US9955894B2 (en) 2014-01-28 2018-05-01 Covidien Lp Non-stationary feature relationship parameters for awareness monitoring
US10285626B1 (en) * 2014-02-14 2019-05-14 Apple Inc. Activity identification using an optical heart rate monitor
WO2015127281A1 (en) 2014-02-20 2015-08-27 Covidien Lp Systems and methods for filtering autocorrelation peaks and detecting harmonics
US9686485B2 (en) 2014-05-30 2017-06-20 Apple Inc. Pixel binning in an image sensor
US9950129B2 (en) 2014-10-27 2018-04-24 Covidien Lp Ventilation triggering using change-point detection
US10888280B2 (en) 2016-09-24 2021-01-12 Sanmina Corporation System and method for obtaining health data using a neural network
US10299734B2 (en) * 2015-10-02 2019-05-28 Covidien Lp Medical device with adaptive power consumption
US9912883B1 (en) 2016-05-10 2018-03-06 Apple Inc. Image sensor with calibrated column analog-to-digital converters
CN111682039B (en) 2016-09-23 2021-08-03 苹果公司 Stacked back side illumination SPAD array
EP3574344A2 (en) 2017-01-25 2019-12-04 Apple Inc. Spad detector having modulated sensitivity
US10656251B1 (en) 2017-01-25 2020-05-19 Apple Inc. Signal acquisition in a SPAD detector
US10962628B1 (en) 2017-01-26 2021-03-30 Apple Inc. Spatial temporal weighting in a SPAD detector
US10622538B2 (en) 2017-07-18 2020-04-14 Apple Inc. Techniques for providing a haptic output and sensing a haptic input using a piezoelectric body
US10440301B2 (en) 2017-09-08 2019-10-08 Apple Inc. Image capture device, pixel, and method providing improved phase detection auto-focus performance
EP3688718A1 (en) 2017-11-15 2020-08-05 Google LLC Unsupervised learning of image depth and ego-motion prediction neural networks
CN108338777A (en) * 2018-02-08 2018-07-31 深圳还是威健康科技有限公司 A kind of pulse signal determination method and device
EP3752060A4 (en) * 2018-02-17 2021-11-24 Trilinear BioVentures, LLC System and method for obtaining health data using a neural network
CN108670209A (en) * 2018-03-29 2018-10-19 中国科学院微电子研究所 A kind of method and system of automatic identification Chinese medicine pulse
WO2019212820A1 (en) * 2018-05-04 2019-11-07 Hi Llc Interferometric frequency-swept source and detector in a photonic integrated circuit
US11019294B2 (en) 2018-07-18 2021-05-25 Apple Inc. Seamless readout mode transitions in image sensors
US10848693B2 (en) 2018-07-18 2020-11-24 Apple Inc. Image flare detection using asymmetric pixels
US11205419B2 (en) * 2018-08-28 2021-12-21 International Business Machines Corporation Low energy deep-learning networks for generating auditory features for audio processing pipelines
US11233966B1 (en) 2018-11-29 2022-01-25 Apple Inc. Breakdown voltage monitoring for avalanche diodes
US11534112B2 (en) * 2020-04-21 2022-12-27 Covidien Lp AI based method for determining oxygen saturation levels
US11672934B2 (en) 2020-05-12 2023-06-13 Covidien Lp Remote ventilator adjustment
US11563910B2 (en) 2020-08-04 2023-01-24 Apple Inc. Image capture devices having phase detection auto-focus pixels
US11546532B1 (en) 2021-03-16 2023-01-03 Apple Inc. Dynamic correlated double sampling for noise rejection in image sensors
CN113312996B (en) * 2021-05-19 2023-04-18 哈尔滨工程大学 Detection and identification method for aliasing short-wave communication signals

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5533511A (en) * 1994-01-05 1996-07-09 Vital Insite, Incorporated Apparatus and method for noninvasive blood pressure measurement
US6035223A (en) * 1997-11-19 2000-03-07 Nellcor Puritan Bennett Inc. Method and apparatus for determining the state of an oximetry sensor
US6280381B1 (en) * 1999-07-22 2001-08-28 Instrumentation Metrics, Inc. Intelligent system for noninvasive blood analyte prediction
US6512936B1 (en) * 1999-07-22 2003-01-28 Sensys Medical, Inc. Multi-tier method of classifying sample spectra for non-invasive blood analyte prediction
US6697654B2 (en) * 1999-07-22 2004-02-24 Sensys Medical, Inc. Targeted interference subtraction applied to near-infrared measurement of analytes
US6788965B2 (en) * 2001-08-03 2004-09-07 Sensys Medical, Inc. Intelligent system for detecting errors and determining failure modes in noninvasive measurement of blood and tissue analytes

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6512937B2 (en) * 1999-07-22 2003-01-28 Sensys Medical, Inc. Multi-tier method of developing localized calibration models for non-invasive blood analyte prediction

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5533511A (en) * 1994-01-05 1996-07-09 Vital Insite, Incorporated Apparatus and method for noninvasive blood pressure measurement
US6035223A (en) * 1997-11-19 2000-03-07 Nellcor Puritan Bennett Inc. Method and apparatus for determining the state of an oximetry sensor
US6280381B1 (en) * 1999-07-22 2001-08-28 Instrumentation Metrics, Inc. Intelligent system for noninvasive blood analyte prediction
US6512936B1 (en) * 1999-07-22 2003-01-28 Sensys Medical, Inc. Multi-tier method of classifying sample spectra for non-invasive blood analyte prediction
US6697654B2 (en) * 1999-07-22 2004-02-24 Sensys Medical, Inc. Targeted interference subtraction applied to near-infrared measurement of analytes
US6788965B2 (en) * 2001-08-03 2004-09-07 Sensys Medical, Inc. Intelligent system for detecting errors and determining failure modes in noninvasive measurement of blood and tissue analytes

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2006134197A1 (en) * 2005-06-13 2006-12-21 Consejo Superior De Investigaciones Científicas Method of processing photopletismographic signals obtained from a person or animal and oximeter using said method
ES2276594A1 (en) * 2005-06-13 2007-06-16 Consejo Superior Investig. Cientificas Method of processing photopletismographic signals obtained from a person or animal and oximeter using said method
WO2011041216A1 (en) * 2009-09-30 2011-04-07 Nellcor Puritan Bennett Llc Systems and method for identifying non-corrupted signal segments for use in determining physiological parameters
CN105380630A (en) * 2014-08-22 2016-03-09 苹果公司 Heart rate path optimizer
CN109106359A (en) * 2014-08-22 2019-01-01 苹果公司 Heart rate path optimization device
CN109106359B (en) * 2014-08-22 2022-03-01 苹果公司 Heart rate path optimizer
CN106798552A (en) * 2015-11-25 2017-06-06 德克萨斯仪器股份有限公司 Heart rate with status switch optimization estimates equipment
US10758185B2 (en) 2015-11-25 2020-09-01 Texas Instruments Incorporated Heart rate estimation apparatus using digital automatic gain control
CN106798552B (en) * 2015-11-25 2021-05-04 德克萨斯仪器股份有限公司 Heart rate estimation device with state sequence optimization
US11129538B2 (en) 2015-11-25 2021-09-28 Texas Instruments Incorporated Heart rate estimation apparatus with state sequence optimization

Also Published As

Publication number Publication date
WO2005020798A3 (en) 2005-07-14
US20050049470A1 (en) 2005-03-03
US6931269B2 (en) 2005-08-16

Similar Documents

Publication Publication Date Title
US6931269B2 (en) Multi-domain motion estimation and plethysmographic recognition using fuzzy neural-nets
US20040039273A1 (en) Cepstral domain pulse oximetry
US6650918B2 (en) Cepstral domain pulse oximetry
US6094592A (en) Methods and apparatus for estimating a physiological parameter using transforms
US5524631A (en) Passive fetal heart rate monitoring apparatus and method with enhanced fetal heart beat discrimination
US6067462A (en) Signal processing apparatus and method
EP0335357B1 (en) Improved method and apparatus for detecting optical pulses
EP1611847A1 (en) Validating pulse oximetry signals in the potential presence of artifact
Stetson Determining heart rate from noisy pulse oximeter signals using fuzzy logic
JP2010234059A (en) Method for determining state of oximetry sensor
CA2330572A1 (en) Method and apparatus for non-invasive diagnosis of cardiovascular and related disorders
CN114974566B (en) Cognitive function assessment method and system
Habbu et al. Noninvasive blood glucose estimation using pulse based cepstral coefficients
AU720609B2 (en) Method and apparatus for measuring pulse rate and saturation
US7037261B2 (en) Fundamental frequency estimation of Electro-Islet Graphy
KR102627661B1 (en) Method for analyzing photoplethysmography data and recording medium storing program to implement the method
RU2442529C1 (en) Means of diagnosis the cardiovascular system
KR20240015709A (en) METHOD AND APPARATUS FOR NON-INVASIVE HbA1c OR GLUCOSE LEVEL MEASUREMENT USING MACHINE LEARNING
EP1082051B1 (en) Frequency estimation of electro-islet-graphy
Alexandridi et al. An integrated system for the diagnosis of cardiac pathology through the analysis of heartbeat interval variability

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A2

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW

AL Designated countries for regional patents

Kind code of ref document: A2

Designated state(s): BW GH GM KE LS MW MZ NA SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LU MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
122 Ep: pct application non-entry in european phase