WO2001087155A1

WO2001087155A1 - Method and apparatus for the detection and diagnosis of cancer, specifically breast cancer using diffusion mri

Info

Publication number: WO2001087155A1
Application number: PCT/US2001/013487
Authority: WO
Inventors: Hadassah Degani; Yael Paran
Original assignee: Yeda Research And Development Co., Ltd.; Fleit, Lois
Priority date: 2000-05-18
Filing date: 2001-04-26
Publication date: 2001-11-22
Also published as: AU2001257313A1

Abstract

Method and apparatus for detection and diagnosis of cancer, specifically breast cancer using an MRI machine for conducting a high resolution MRI of tissue to produce and record an image set in digital form. A controller (22) controls the MRI machine (20) to set the time of diffusion to a short constant time and diffusion strength gradients to predetermined values. A calculator (26) determines b values based on the diffusion strength gradients. A processor (28) obtains best fitted parameters using a preselected fitting algorithm to fit signals in the set of images, pixel by pixel.

Description

METHOD AND APPARATUS FOR THE DETECTION AND DIAGNOSIS OF CANCER, SPECIFICALLY BREAST CANCER USING DIFFUSION MRI

BACKGROUND OF THE INVENTION

Field of the Invention

The present invention relates to a non-invasive/safe imaging method and apparatus for the detection and diagnosis of cancer, especially breast cancer using diffusion MRI. More particularly, the invention relates to a system including novel software for the unique processing of signals derived from diffusion MRI conducted according to the teachings of the invention.

Description of the Prior Art

> ^■. Cancer, a term referring .to more than 100 forms of the disease, is most successfully treated if detected early. For this reason, many physicians recommend that pepple over the age of 40 have annual health checkups, which can often detect the disease beforejf produces any^'symptoms. Millions of medical diagnostic procedures are carried out each year to detect cancer. Diagnostic procedures employ afyide range of blood tests, through medical imaging procedures to biopsies and other invasive procedures.

Almostevery tissue in the body can spawn malignancies; some several types. Prostate and breast cancer are the leading cancers in males and females, respectively. Although each form of cancer has. unique features, the basic processes that produce these diverse tumors, as well as'-the need for early detection and treatment in order to eradicate the disease, appear to be similar.

In addition to the development of genetic and molecular tests, it is important to develop new non invasive/safe imaging methods which accurately localize the malignant tissue and demonstrates the dimension, the spread and the unique pathophysiologic characteristics of the disease. Despite efforts to develop a method that has the capability to provide specific image contrast that can be then interpreted in terms of specific values of physiologic parameters typical to cancer, and significantly different from those of normal or benign tissues, no one has yet succeeded.

One approach that has been put forth is use of examination by Magnetic Resonance Imaging (MRI) with the aid of a contrast material. This procedure has been found to be extremely sensitive in the detection of invasive cancer. However, the larger problem with breast MRI has been the differentiation of benign conditions from malignancy.

In an earlier development a method for breast MRI diagnosis called the Three Time Point (3TP) method was advanced in which the changes in tissue contrast upon injection of intravenous Gd-DTPA were used to predict benign or malignant lesions with an accuracy up to 97% depending on the type of breast lesion. By this prior technique, breast lesions can be distinctly identified by Gd-based contrast enhanced MRI. The architectural features revealed in the enhanced regions have been suggested to improve the specificity of breast cancer diagnosis. The dynamics of the contrast enhancement using empiric and model based approaches has been also shown to improve the specificity of breast cancer diagnosis. The method is based on recording high resolution contrast enhanced images at three different time points (hence the term 3TP), judiciously chosen so as to maximize the diagnostic information. With the aid of a model based image processing algorithm, the time evolution of the enhancement was related to the parameters that can characterize malignant tumors and differentiate them from benign ones, namely micro-capillary surface area x permeability and extracellular volume fraction (EVF). The results of the first comprehensive clinical testing of this method in the Department of Radiology, University of Wisconsin (46 lesions with a final diagnosis: 25 malignant lesions and 21 benign lesions) showed very high sensitivity and specificity.

SUMMARY OF THE INVENTION

By the present invention, a novel method and apparatus has been developed, utilizing "non-contrast", for diagnosing malignancy based on diffusion MRI. The novel method and apparatus enable mapping, at high resolution, angiogenic and morphological parameters similar to those obtained by contrast enhanced MRI. The novel method and apparatus are based on measurements of, the movement of water molecules in different tissue compartments, which are processed in a unique manner. Development of a non-contrast method has several advantages in lowering the cost of breast MRI: 1. by eliminating the need for a contrast material and an injection procedure, 2. by increasing throughput in the MRI scanner, 3. by permitting non-monitored screening examination to be performed. Furthermore, for some patients who can not tolerate the contrast agent, such as, pregnant patients or those with demonstrated allergy, breast MRI becomes a feasible examination. The availability of MR mammography to women of all ages, including high risk and pregnant women, will help reduce breast cancer mortality by early detection and reliable diagnosis.

In vivo water diffusion in tissues is a complex process influenced by microscopic and macroscopic morphologic heterogeneity and by variations in biochemical composition and physical properties such as osmotic pressure, flow and exchange rate. As the spatial resolution of the MRI increases, a higher homogeneity of the distribution of the cells and of the interstitial spaces within each voxel is reached, increasing the accuracy of mapping the intraceilular and extracellular properties and the intravascular compartment.

The method and apparatus of the present invention are based on diffusion MRI conducted employing specified preselected diffusion MRI sequences. The resulting data is processed in a unique manner to obtain measurements of several histologic and physiologic parameters that enable detection and diagnosis of breast lesions: the intra- and extracellular apparent diffusion coefficients, the "pseudo- diffusion coefficient" which reflects the capillary blood flow, and the intraceilular, extracellular and intravascular volume fractions. The invention provides the preselected conditions and protocol necessary for determining the critical information and provides the image processing algorithms to obtain a final diagnosis and portray the diagnosis in an image representation. The inventive method and apparatus have been tested on human breast cancer implants in nude mice. The novel method and apparatus employs high- resolution diffusion MRI with a broad range of diffusion gradient strengths and low diffusion time duration. Processing and analysis are performed at pixel resolution. The analysis is based on tissue compartmentalization to three major compartments, intraceilular, extracellular and intravascular and on a model-based equation of the effect of the water motion in these compartments on the MRI signal. The novel method and apparatus have the capability to map tumor microvessels density, cell density and arrangement, water properties and compartmentalization. From this information, breast cancer detection and differentiation between benign and malignant lesions becomes feasible with a high degree of accuracy.

Other and further advantages and objects of the present invention will become more evident from the following detailed description of the invention when taken together with the appended drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

Figures 1 A to 11 shows an analysis of diffusion-weighted images of MCF7 tumor implanted in the mammary fat pad of a nude mouse with Figure A showing an EVF map obtained by a uniexponential fitting using a range of color-coded scale: 0 - 1. Figure 1B shows an ADCc map obtained as in Figure 1A using a range of color-coded scale: 0 - 3 x 10^"3 mm²/s. Figure 1C is a correlation coefficient (R²) map of the uniexponential fitting using a range of color-coded scale: 0 - 1. Figure 1 D is an EVF map obtained by a biexponential fitting using a range of color-coded scale: 0-1. Figure 1E is an ADCe map obtained as in Figure 1D using a range of color-coded scale: 0 - 3 x 10^"3 mm²/s. Figure 1F is a correlation coefficient (R²) map of the biexponential fitting using a range of color-coded scale: 0 - 1. Figure 1G is a map showing the subtraction of the EVF values mapped in Figure 1A and Figure 1 D in the tumor ROI using a range of color-coded scale: -1 to Figure 1H is a graph or Histogram of the EVF difference map shown in Figure 1G. Figure 11 is a T₂-weighted spin-echo image. The diffusion-weighted images were recorded and analyzed with a spatial resolution of 0.39 x 0.39 x 2 mm. The boundary of the tumor and the mouse body is marked in white. The bar represents a length of 0.5 cm. Figures 2A to 2E shows maps and graphs of the distribution of EVF and ADCe in MCF7 tumor. Figure 2A shows an EVF map obtained by a biexponential fitting using a range of color-coded scale: 0 - 1. Figure 2B shows an ADCe map obtained as in Figure 2A using a range of color-coded scale: 0 - 3 x 10^"3 mm²/s. Figure 2C shows a histogram of EVF distribution in Figure 2A. Figure 2D shows a histogram of ADCe distribution in Figure 2B. Figure 2E is a graph showing correlation between ADCe and EVF and wherein the slope of the linear line was 0.95 x 10^"3 mm²/sec with R coefficient of 0.46. Data recording and processing are as described with the spatial resolution of 0.39 x 0.39 x 2 mm. The bar in Figures 2A and 2B represents a length of 0.5 cm.

Figure 3A to 3C show the distribution of EVF values in histological sections of MCF7 tumors implanted in the fat pad of nude mice. Figure 3A shows a viable region with EVF of 0.4, the bar size shown in this Figure is 50μm. Figure 3B shows a partial necrosis with EVF of 0.8, the bar size shown in this Figure is 50μm. Figure 3C show a histogram of EVF values in viable regions (40 fields in 6 tumors). The histological sections were stained with Hematoxylin-Eosin, and the EVF values were determined as described.

Figure 4 is a flow chart of the method and apparatus of the present invention.

Figure 5 is a block diagram showing the apparatus of the present invention.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENT

The method and apparatus of the present invention is based on diffusion MRI which is a very common and established measurement. In diffusion MRI a set of images are recorded in a certain way with the images varied from one image to the other by either the time of diffusion ( the time fixed to follow the diffusion) or the diffusion gradient strength ( during the diffusion time a gradient is applied and its strength can vary), or both. These two important conditions (diffusion time and gradient strength) determine the analysis and the final results. According to the present invention, the diffusion measurement is performed in a specific manner, namely, applying a short and constant diffusion time and varying the diffusion gradient strengths to include high strengths. Through the practice of the present invention, the water diffusion properties in the different tissue compartments (intraceilular apd extracellular compartments ) can be separated, and then, by analyzing the diffusion images at the different diffusion gradient strengths, maps are obtained of the volume fraction of each compartment and the ADCs ( apparent diffusion coefficients in each compartment). In a specific Example set forth herein, the invention has been demonstrated in animals for two compartments: the extracellular and intraceilular part. The extracellular compartment, however, can be further divided to intravascular and extracellular parts, and the diffusion imaging can be performed so that also the intravascular volume fraction can be mapped. This is very important for diagnosis as was demonstrated in the prior art 3TP method. Notwithstanding, the EVF is also very important, and more accurate in the diffusion measurements. From the clinical aspect, based on the prior 3TP method, it is known that all benign breast tumors have high and homogenous extracellular volume fraction. Therefore, even just an extracellular volume fraction (EVF) map produced by the practice of the present invention will be very useful for breast diagnosis.

As noted above, the two parameters or conditions under which the diffusion MRI must be conducted are: a short diffusion time and the same one in all the diffusion imaging set where you vary the strength of a diffusion gradient. It is also important to use high enough strength of diffusion gradients. What is high enough is determined by the tissue properties that can be estimated before the diffusion MRI. In a more specific invention, the method and apparatus are carried out using a time saving protocol to obtain the map of EVF and the ADC of the cells. This protocol uses only the high diffusion gradient images. By the present invention, all the images must be recorded with the same and short diffusion time and using preselected diffusion gradient strengths. The minimum number and the strength of the of the diffusion gradients are predetermined in order to be then able to obtain the maps that describe EVF and the ADCs. The diffusion time may be preselected in the range of from 5 to 20 ms, or alternatively, about less than one-half the average water lifetime in cancer cells of 50 ms. More generically, the diffusion time should be shorter than the average lifetime of water inside cells, preferably at least a factor of 2 shorter or more with no limit for the lowest time (which must be greater than 0). The diffusion gradients are expressed as b values as will be apparent from the following detailed description. The b values depend in great measure on the diffusion rate in the extracellular compartment. In general terms, high b values can be defined as the values in which the diffusion of the extracellular water is almost completely attenuated and resulting signal is due essentially to intraceilular water. It is also important to note that the b value depends on the diffusion time, so as the diffusion time is reduced, the diffusion gradient strength must be increased. These parameters must be "tailored¹ for each tissue being examined, but once found, can be fixed.

A special equation or algorithm is used to describe the signal intensity evolution with a change in the diffusion gradient strength. Some of the parameters in this equation can be measured or calculated independently and the other parameters are those that are found by fitting the signal in the set of images ( pixel by pixel) to this special equation or algorithm, and getting the best fitted parameters. To this end simple fitting algorithms or more complicated ones can be used. However, according to the invention, it is necessary to fit the algorithm to the image so that as a result the fitted parameters are presented also as an image. Thus, every aspect is presented as an image so that the spatial distribution of the final extracted values for the parameters being calculated by the invention are demonstrated in a clear manner. A regular color scale is used to reflect different values.

In the practice of the invention, the set of images that were recorded with different diffusion gradient strengths and were analyzed using the equation or algorithm based on the model used, are actually analyzed on a pixel by pixel basis. The final analysis is then represented in a colored image format and can be placed on the original image . The programs disclosed herein can be combined to a software package that will analyze the diffusion gradient images (obtained as the protocol dictates) by pushing one button (or typing one command) and the final output of the results will be obtained in image presentation of the parameters, as the preferred mode of presentation. However, alternate modes of presentation are within the contemplation of the invention. Such manners of presentation will be evident to those skilled in the art of programming, software and presentations. The apparatus of the present invention includes a computer system operating electronically, optically or both, having a memory, a central processing unit, a display, an input device for generating device event signals and coacting therewith software for use with the computer. The software (in binary or related form) comprises a computer usable medium having computer readable program code thereon including the program logic for implementing the various flow charts and block diagrams described herein. Since the details of computers are well known in the art and because persons skilled in the art have sufficient expertise and knowledge to be capable of implementing flow charts and block diagrams, a detailed description of the specific hardware has been omitted as superfluous and unnecessary to a full and complete understanding and appreciation of the present invention as described above. Those skilled in the art will be able to make and use the apparatus and practice the method of the present invention from the detailed description and teachings contained herein.

In the specific Example that will be described in detail herein, the method and apparatus of the present invention was used to detect MCF7 tumors. The MCF7 tumors implanted in the mammary fat pad were scanned using a diffusion sensitive sequence with a constant diffusion time and increased diffusion gradient strengths. Image processing based on a slow exchange in a multi-system model yielded yields maps of the vascular pseudo diffusion coefficient (ADCv) which characterizes the blood vessels, the intraceilular apparent diffusion coefficient (ADCe) and the equivalent extracellular parameter ADCe, as well as, of the volume fractions of each compartment. A demonstration of mapping the extracellular volume fraction (EVF) and the extracellular apparent diffusion constant (using color coding) in image format of a breast cancer tumor is shown in the Figures of the drawings which will be explained in more detail hereafter.

The EVF was also obtained using a novel and inventive "time saving" method. Diffusion weighted images recorded at b=0 and at three highest b values (in the range 3.2 x 10³ to 5.6 x 10³s/mm² were fitted to a uniexponential decay, yielding EVF and the apparent diffusion constant in the cells (ADCe). The intra-tumoral viable regions demonstrated varying EVF values ranging from 0.2 to 0.7. ADCe also appeared heterogeneous within each tumor and between different tumors, spanning a large range from 0.3 x 10^"3 to 2.0 x 10^"3 mm²/s, with the high values usually found in necrotic regions. These variations are due to differences in the vascularity and tortuosity (shape and arrangement of the extracellular environment). The ADCe appeared to be homogeneous within the same tumor due to the similar properties of the cancer cells is the same type of tumor. The results were validated by histopathological characterization of the tumors. The specific Example will now be described in detail.

Example of Diffusion MRI of MCF7 Human Breast Cancer; Mapping Physiologic and Morphologic Parameters.

The method and apparatus of the present invention was carried out using high-resolution diffusion MRI that utilized a broad range of diffusion gradient strengths and a short and constant diffusion time, and was applied in vivo to study MCF7 human breast cancer implanted in the mammary fat pad of nude mice. Processing and analysis at pixel resolution, assuming a two compartment system at slow exchange, yielded calculated images of the intra- and extracellular water apparent diffusion coefficient (ADCe and ADCe, respectively) and of the extracellular volume fraction (EVF). ADCe values were homogeneous and similar for all tumors with a mean ±SD of (0.27±0.03)x10^~3 mm²/s (~27°C). ADCe values were heterogeneous, spanning a broad range from 0.3x10^"3 to 2.0x10^"3 mm²/s (~27°C). EVF values calculated from analysis of the high b-value images or from the whole data set were similar and exhibited heterogeneous distribution in the range of 0.2 to 1 , confirmed also by analysis of histological stained slices. The heterogeneous distribution of EVF and ADCe emphasizes the importance of mapping these parameters at high spatial resolution in tumor studies. As noted, the present invention carries out a high-resolution diffusion MRI of MCF7 human breast cancer implanted in the mammary fat pad of nude mice. The selection of the conditions for this example, and the processing and analysis at pixel resolution enabled mapping of the extracellular volume fraction (EVF) along with the intraceilular ADC (ADCe) and extracellular ADC (ADCe). By an improvement to the generic invention, it is possible, according to a more specific method and apparatus of the present invention, and as demonstrated by this example, to use a limited number of images obtained at high diffusion gradient strengths to provide a time-saving method for mapping EVF.

Tumor Model

MCF7 cells were cultured routinely in DMEM supplemented with 6% fetal calf serum (FCS) as known in the art. The cells were inoculated in the mammary fat pad of female CD1-NU athymic mice following a procedure known in the art. Seven tumors in different mice with a mean ± SD tumor size of 2.1 ± 0.9 cm³ (range 0.7 to 2.9 cm³) were investigated.

The mice were anaesthetized prior to the MRI measurements by an intraperitoneal injection of Ketamine HCL (85 μg/g wt, Ketaset, Fort Dodge Laboratories, Iowa, USA) and Xylazin (20 μg/g wt, Bayer, Leverkusen, Germany). Further low doses (Ketamine 8.5 μg/g wt, Xylazin 2 μg/g wt) of anaesthesia were added subcutaneously via an infusion line every thirty minutes. The temperature was maintained at (27 ± 1)°C. The body temperature of the mice (32 ± 1)°C was reduced by the anaesthesia to (27 ± 1)°C.

MRI

¹H MR images were recorded at 4.7 Tesla using a Bruker Biospec 4.7/30 spectrometer and a Bruker RF resonator with 7 cm diameter. Shielded gradient coils were utilized with a maximum gradient strength of 20 gauss/cm for a rise time of 200 μs. A PGSE sequence (Pulse Gradient Spin Echo) was applied with echo time (TE) of 34.5 or 39 ms, repetition time (TR) of 2400 ms, time interval separating the onset of the diffusion gradients (Δ) of 16 ms or 20 ms, duration of diffusion gradient (δ) of 13 ms, field of view of 5 cm, slice thickness of either 1 mm with a matrix of 64 x 64 (17 slices in five tumors) or 2 mm with a matrix of 128 x 128 (10 slices in two tumors). Each tumor was scanned using 11 diffusion-encoding gradients (G_d) applied in the read direction. The b values were calculated by including the cross term

between the diffusion and the imaging gradients: b = γ²(G aδ²(A-— ^) + 2AG_rβG_dδ)

where y is the gyromagnetic ratio, G_r is the amplitude of the imaging gradient in the read direction and β is the duration of G_r. The b values ranged between 0 to 5.6 x 10³ s/mm² with five b values larger than 3.2 x 10³ s/mm². The number of averages used to obtain the images increased from 2 to 10 as the strength of the diffusion gradients increased, maintaining a signal-to-noise ratio larger than 25. The final calculations took into account this variation in the number of averages. Each tumor was also scanned with a T₂-weighted spin-echo sequence with TE/TR = 80/3200 ms, field of view of 5 cm, slice thickness of either 1 mm with a matrix of 256 x 128 (17 slices in five tumors) or 2 mm with a matrix of 128 x 128 (10 slices in two tumors).

Histology

At the end of scanning, each tumor was surgically removed and fixed in a 4% formaldehyde solution and embedded in paraffin. Representative 4 μm thick slices were stained with Hematoxylin-Eosin (H&E) to provide a comparison of the MR images to histopathology. Typical regions (500 x 300 μm, n=40) from the histologic slices were photographed by Olympus DP10 camera using Olympus BX50WI microscope. These images were transferred to a computer and were analyzed using Adobe Photoshop version 5. The fractions of EVF and CVF were determined on the basis of the difference in coloring using automatic selection of pixels within a defined color range. In addition, correlation between the gross histopathologic features in the MR images and the histological sections were performed by visual inspection.

Model and Image Analysis

The analysis of the diffusion-weighted images was based on the assumption that each tumor pixel contains two several major compartments: such as intraceilular and extracellular. The constant diffusion time (Δ - δ/3) in each run (11.7 or 15.7 ms) was chosen to be much shorter than the average water lifetime in cancer cells of 50 ms. This justified analysis of the data at the slow exchange limit. Thus, the attenuation of the PGSE signal intensity as a function of the strength of the diffusion gradients (G_d), for a constant diffusion time at the slow exchange limit, is given by the following equation:

I(b /I(0) = ∑INiKi exp(-bDi ∑NiKi [1] Where i labels the compartment N is the compartment volume and D the apparent diffusion coefficient (ADC) where:

p is the proton density in the compartment, and Ti and T₂ are the intrinsic water relaxation times.

For a tissue with two compartments: intraceilular and extracellular

1(b) ₌ V_eK_e e _V(-bD P) + V_CK_C exp(-bD P)

[1b]

1(0) v_eκ_e +v_cκ_c

The subscripts c and e designate the intraceilular and the extracellular compartments, respectively.

It is possible to express this normalized attenuation in terms of the EVF defined

as ^ — or the CVF, both clearly related (EVF + CVF = 1 ), yielding:

' e ⁺ c

1(b) ₌ (EVF)K_e exp(-bD^) + (l -EVF)K_c ex_V -bD^) 1(0) (EVF)K_e + (1 - EVF)K_C

At high b values the signal of the fast-diffusing extracellular water is almost completely attenuated and only the intraceilular, slowly diffusing water contributes to the measured signal intensity. Therefore, the normalized signal attenuation as a function of high b values decays approximately in a uniexponential manner with two free parameters CVF and ADCe. K_e and K_c were estimated from MRI measurements in MCF7 tumors of necrotic and dense viable regions, respectively. The ratio pc Pe ⁼ 0-9 was determined by analyzing density-weighted spin echo images (TE/TR= 9ms/15s). The parameters Tι_e= 3280 ms, T-ι_c= 2300 ms, T_2e= 84 ms and T_2c= 60 ms were known from previous prior art measurements in these tumors. EVF and ADCe were then calculated from data obtained at b=0 and at high b values (five values above 3.2 x 10³ s/mm²) fitted to a uniexponential decay. Further analysis included all the b value data points that were fitted to a biexponential decay (Eq. [2]) using a Levenberg- Marquardt least-squares algorithm. Attempts to fit all three parameters (ADCe, ADCe and EVF) failed to yield sensible values in many pixels. Therefore, the ADCe value obtained from all tumors by the uniexponential fitting, was used as a fixed parameter and the ADCe and EVF as free parameters. The fitting quality was demonstrated in correlation coefficient (R²) maps according to known prior art techniques.

The similarity between the EVF maps was assessed by subtracting the EVF image derived from the biexponential fitting from that obtained by the uniexponential fitting. The resultant EVF difference values were plotted as a map, and also, as a histogram of the number of pixels for each difference (ranging from -1 to 1). Analysis of these histograms in terms of the position of the maximum, the shape and the half width at half height served to statistically assess the similarity between the EVF maps of all_. the tumors.

Estimation of the sensitivity of the EVF values to errors in the physical parameters p, Ti , T₂ and ADCe (in the biexponential fitting) was performed by calculating an error propagation ratio (EPR, defined by:

.. _ fractional change in the EVF value fractional change in the parameter used to obtain EVF

EPR values were calculated for three typical EVF values of 0.3, 0.5 and 0.8.

Results

MCF7 tumors implanted in the mammary fat pad were scanned using a diffusion sensitive PGSE sequence and were analyzed as described above. Image processing yielded maps of ADCe, ADCe and EVF as demonstrated in Fig. 1 for a typical tumor. The distribution of necrosis and viable regions in this tumor was deduced from the signal intensity distribution in T₂-weighted images (Fig. 11), with necrosis exhibiting high intensity and viable tumor regions intermediate intensity (gray pixels in Fig. 11). Diffusion weighted images recorded at b=0 and at the five highest b values (in the range 3.2 x 10³to 5.6 x 10³s/mm²), where the signal attenuation was assumed to be solely due to the intraceilular water, were fitted to a uniexponential decay, yielding EVF and ADCe maps (Fig. 1 A-B). The correlation coefficient of the uniexponential fitting was higher in viable tumor regions (R² > 0.8) than in necrotic regions (R² < 0.7) (Fig. 1C). This is consistent with the fact that at the high diffusion gradients the signal intensity in the viable tumor regions (with a high fraction of cell volume) is higher than that in necrotic regions, which is close to the noise level. Calculation of ADCe in the viable tumor regions using uniexponential fitting yielded similar values over the whole tumor and in tumors implanted in different mice, with a mean ± SD of (0.27 ± 0.03) x 10^"3 mm²/s (7 tumors with a total of 27 slices). The EVF -maps derived from this fitting exhibited a diverse distribution within the same tumors and in different tumors, in accord with the uneven packing of cells throughout the tumor. Necrotic regions, usually localized in the center, were identified by their high EVF values of 0.8 - 1 (Fig. 1A).

Upon further analyzing the diffusion-weighted images by fitting the signal intensity as a function of all b values to a biexponential decay (Eq. [2]), using the mean ADCe (0.27 x 10^"3 mm²/s) as a fixed parameter and ADCe and EVF as free parameters, maps were obtained of EVF, ADCe and the correlation coefficient R². Maps of EVF, ADCe, and the correlation coefficient R² of the biexponential fitting are demonstrated in Fig. 1D to 1F. The EVF map derived from the biexponential fitting exhibited a non-uniform distribution of this parameter similar to the EVF map derived from the uniexponential fitting (Fig. 1). Necrotic regions were clearly identified by their very high EVF values. The intra-tumoral viable regions demonstrated varying EVF values ranging from 0.2 to 0.7 (Fig. 1D, 2A, 2C). ADCe also appeared heterogeneous within each tumor and between different tumors, spanning a large range from 0.3 x 10^"3 to 2.0 x 10^"3 mm²/s, with the high values usually found in necrotic regions (Fig. 1E, 2B, 2D). These variations are presumably due to differences in the nature and tortuosity (shape and arrangement) of the extracellular environment (Fig. 3). The correlation coefficient (R²) of the biexponential fitting was high (above 0.8) in all tumors (Fig. 1F).

The subtraction of the EVF maps (EVFuniexponentiai - EVFbiexponentiai, Fig. 1G) demonstrated the presence of similar EVF values (difference =~ 0) in a large fraction of the pixels. Furthermore, histograms of difference maps (Fig. 1H) exhibited a bell- shaped form with the maximum very close to zero at an average value for all tumors of 0.06 ± 0.06 (mean ± SD, 27 slices) and a mean half width at half height of 0.10 ± 0.03 (mean ± SD, 27 slices), ranging from 0.04 to 0.14.

The EVF and the ADCe reflect the nature of the interstitial environment and may therefore show a correlation. In regions with high EVF, usually necrotic, ADCe was high while in viable regions, with a low to moderate EVF, ADCe was lower (Fig. 1D, 1E, 2A-D). However, a pixel by pixel linear correlation of these two parameters (Fig. 2E) showed either a low to moderate R coefficient (ranging for 15 slices between 0.40 to 0.75 with a mean ± SD slope of 1.6 ± 0.7 mm²/s) or no correlation (12 slices). The poor correlation was particularly emphasized in viable tumor regions (Fig. 2E).

Inspection of the histological slices of the tumors provided visual confirmation of the presence of viable and necrotic regions (Fig. 3). An assessment of the EVF values in viable tumor regions was obtained by analyzing the stained histology sections to yield the EVF values in arbitrarily selected viable regions (Fig. 3C). Most viable tissues demonstrated an average EVF ranging between 0.40 to 0.55. In few areas of highly dense tissue, the EVF values were low (0.30 to 0.40), while less dense areas that still appeared viable, reached EVF of 0.65. These results are in accord with the EVF distribution derived from the diffusion studies (Fig. 1 and 2).

The sensitivity of the fitted EVF values to errors in the estimated physical parameters was determined by calculating the error propagation ratio using a known prior art technique for three selected EVF values (Table 1). The absolute values for all EPRs consistently increased as EVF decreased. For an EVF value of 0.8, a 10% change in any of the parameters resulted in less than 2% change in this value. Thus, high EVF appeared to be almost independent of variations in the estimated parameters. For EVF of 0.5 the EPR increased but still remained low, with a 10% change in any parameter leading to not more than a 5% change in the EVF value. At the low range of EVF (0.3) there was a significant increase in the EPR for all the estimated parameters. In particular, the biexponential fitting became exceedingly sensitive to the estimated value of ADCe, shifting EVF by 25% for a change of 10%. As the fraction of pixels with this low EVF was very small (Fig. 1 and 2), the conclusion was reached that overall the mapping of the EVF was not markedly affected by small errors in the estimated parameters.

In vivo water diffusion in tissues is a complex process influenced by microscopic and macroscopic morphologic heterogeneity and by variations in biochemical composition and physical properties such as osmotic pressures and exchange rates. As the spatial resolution of the MRI increases, a higher homogeneity of the distribution of the cells and of the interstitial spaces within each voxel is reached, increasing the accuracy of mapping the intraceilular and extracellular properties. Accordingly, by the present invention, it is possible to show that high-resolution diffusion MRI can reveal breast cancer heterogeneity and provide estimated maps of the intraceilular and extracellular diffusion coefficients and volume fractions. Although tissues have a third, intravascular, compartment, the fraction of this compartment in cancers is usually low (ranging between 0.01 to 0.20, and therefore, can be often neglected or integrated as part of the extracellular compartment. This is particularly justified at high b values, where the contribution to the signal intensity by water flowing in the blood vessels or by the intravoxel incoherent motion (MM) vanishes completely.

Image processing and analysis yielded estimated maps of physiologic and morphologic parameters distinct to MCF7 human breast cancer. The two physiologic parameters, ADCe and ADCe, are characteristic cell and tissue parameters, respectively, and the EVF describes a morphologic tissue parameter. ADCe and the lifetime of a water molecule in cultured cells depend on cell size and membrane permeability and are intrinsic parameters of the cells. These parameters presumably do not change markedly in a tumor derived from the cells. The major differences between cell culture and tumors are the morphology, composition and architecture of the extracellular matrix. Thus, the parameters obtained in the spectroscopic diffusion studies of cell cultures can provide basic understanding of the cellular diffusion but can not mimic extracellular behavior of tumors. The tumor spheroid model resembles more closely the tumor in vivo and average values of ADCe, ADCe and EVF can be compared to those of tumors.

The intrinsic nature of ADCe was confirmed here by the similarity of this parameter (about 0.27 x 10^"3 mm²/s at ~27°C) within and among the tumors, all derived from the same cellular origin. However, comparison with results obtained in cell culture studies shows a discrepancy: van Zijl et al " Complete separation of intraceilular and extracellular information in NMR spectra of perfused cells by diffusion-weighted spectroscopy". Proc Natl Acad Sci U S A 1991 ;88:3228-3232 measured ADCe in MCF7 cells of 0.11 x 10^"3 mm²/s (temperature was not specified); while ADCe of 0.47 x 10^"3 mm²/s at 37°C was determined by Pilatus et al "Intraceilular volume and apparent diffusion constants of perfused cancer cell cultures, as measured by NMR", Magn Reson Med 1997;37:825-832 in MCF7 cells and other carcinoma and fibrosarcoma cell lines; ADCe of 0.25 x 10^"3 mm²/s at 22°C was determined in spheroids of EMT6 rat mammary cells by Neeman M, Jarrett KA, Sillerud LO, Freyer JP. "Self-diffusion of water in multicellular spheroids measured by magnetic resonance microimaging". Cancer Res 1991 ;51 :4072-4079; and a very low ADCe of θ!06 x 10^"3 mirrVs at 33°C was determined in glioma cell cultures by Pfeuffer J, Flogel U, Leibfritz D. "Monitoring of cell volume and water exchange time in perfused cells by diffusion-weighted 1H NMR ". NMR Biomed 1998;11 :11-18. Aside from the differences in the temperature, these variations could result from differences in the cell type and in the nature of the matrix in which the cells were embedded. Thus, although the in vitro studies may yield valuable information, it is preferable to measure in vivo the actual diffusion properties of the cells in tumors.

ADCe depends on the composition and tortuosity of the interstitial space. In viable regions with dense packing of cancer cells and complex tortuosity, ADCe appeared to be low and closer to ADCe. In viable regions with possibly loosely packed cells and less tortuosity, ADCe reached values 3-fold higher than ADCe. In necrotic regions, where water diffuses more freely, and there is practically no tortuosity, ADCe was high, reaching values close to that of free water ((2.34 ± 0.08) x 10^"3 mm²/s at (25.5 ± 0.5)°C ). Thus, MCF7 tumors demonstrated heterogeneous distribution of ADCe.

Most cancer cells in tumors are constantly proliferating but some of the cells remain quiescent and some undergo cell death processes (apoptosis, necrosis). Therefore, the cellular and EVF distribution is heterogeneous and varies with time. This heterogeneity is demonstrated here in the large variations in the EVF values ranging from 0.2 to 1.0 (MRI and histological analysis). EVF values obtained by the two analyses (uniexponential and biexponential fitting) were similar. The relatively large changes in the signal intensity at low b values could elicit a bias in the biexponential fit to a higher EVF while the analysis of the high b values as a uniexponential decay could artificially lower EVF due to the neglection of contribution from the extracellular environment. However, the bell-shape distribution around zero of the EVF-difference-histograms, see Figure 1H, suggests that both biases were insignificant. From a practical, time-saving aspect it appears that the uniexponential high b value fitting can be more useful clinically.

The signal attenuation curves were also affected by the nuclear relaxation rates (1/Tι and 1/T₂) and by the proton density in each compartment (Eq. [1]). Measurements of these physical parameters in regions that were highly cellular or predominantly non-cellular (necrotic) yielded approximated intraceilular and extracellular values of these parameters, respectively. To assess the effect of this approximation on the final results, the EPR of the EVFs (Table 1) was calculated. Generally, in most of the tumor volume (EVF > 0.5) inaccuracies in each of the physical parameters of up to 10% modulated by less than 5% the final EVF (determined from the uniexponential or biexponential fitting). The highest EPR was exhibited for the density ratio, however, it is not anticipate that there is more than 10% error in this parameter. In addition, in these regions the changes due to the error in the extracellular T1 and intraceilular T2 were compensated by those of the extracellular T2 and intraceilular T1 , thus reducing the final error in the estimation of the EVF. At the low EVF range (≤0.3) the EPR were generally higher but still variations in the relaxation and density parameters led to a low change (percent- wise) in EVF values. However, the biexponential fitting became in this range highly susceptible to the accuracy of ADCe. As regions of low EVF occupied a small fraction of the whole tumor (<10%) this did not appear to modulate significantly the EVF distribution obtained by the two analyses.

The EVF and the ADCe are associated with the properties of the extracellular compartment. In part of the tumors, EVF and ADCe exhibited a low to moderate linear correlation. The failure to show a correlation occurred predominantly in viable regions, where the diffusion coefficient is sensitive to changes in the tortuosity and the interstitial pressure gradient that are not necessarily influencing EVF. Heterogeneity in the biochemical composition of the extracellular matrix could also affect the ADCe and not the EVF. Thus, it appears that these two parameters may frequently be independent of one another.

Most of the prior art diffusion imaging studies focused on the nervous system, which exhibits distinct cellular and tissue characteristics. In studies of implanted MCF7 breast carcinomas, ADC maps derived from relatively low b values enabled differentiation between viable regions and necrosis. The ADC in the viable regions have been found to range between ~0.5 x 10^"3 to ~0.7 x 10^"3 mm²/s. This range is significantly higher than ADCe noted in the development of the present invention and overlaps with the range of ADCe in viable regions (Fig. 3), demonstrating the predominant influence of the extracellular properties on the ADC at low b values. The average ADC in highly necrotic regions of these tumors of 1.3 x 10^"3 mm²/s (12) is close to ADCe of 1.2 x 10^"3 mm²/s determined in high EVF regions. Clearly, a more detailed resolution between the various regions of the tumors was obtained by the present invention by mapping ADCe and EVF.

In summary, the present invention enables diffusion imaging using preselected specific MRI parameters and model-based image processing to yield mapping at pixel resolution of ADCe, ADCe and EVF. ADCe is an intrinsic feature of cancer cells and may uniquely characterize their size, composition and water permeability. ADCe and EVF and their spatial distribution characterize distinct features of the tumor. Developing high-resolution diffusion MR mammography, from which maps of att these three parameters and of the related vascular parameters will be obtained, provides an effective "non-contrast" method for breast cancer diagnosis. Table !

EPR values for the estimated parameters in the EVF calculation.

EPR EPR EPR

(EVF=0.3) (EVF=0.5) (EVF=0.8)

T1 -0.48 0.35 0.14 extracellular

T2 -0.29 -0.21 -0.09 extracellular

T1 -0.40 -0.29 -0.12 intraceilular

T2 0.41 0.29 0.12 intraceilular

Pc/Pe 0.71 0.51 0.21

ADCc ^a 2.46 -0.42 -0.09

^a Used as constant parameter only in the fitting to the biexponential equation (Eq. [2]).

It should be noted that at high b values, the first term in equation 2 becomes zero ( most of the signal is attenuated) and only with the second term is left. This allows for calculation of EVF and ADCe from a fitting to a single exponent of the high b value data. These two parameters may be sufficient in certain cases to make a diagnosis (clearly EVF map can identify fibroadenomas that are the most common benign breast tumor). If the whole range of b values is used, it allows mapping of also ADCe. This is an additional parameter that can be of diagnostic value. Moreover, if one measures and analyzes the "low b value" data, decomposing these data to intravascular and Extravascular/extracellular it is possible to map the intravascular volume fraction and extravascular/extracellular volume fraction and the ADCe and ADC in the blood capillaries. The resulting maps will be valuable tools in a diagnosis. In Figure 4 a flow chart is shown of the inventive method and apparatus. As shown, in Step S10 the diffusion time is selected and set. As noted previously, the diffusion time is set to a relatively short time. In Step S11 a number of diffusion strength gradients are selected or predetermined, based on the tissue undergoing analysis by the MRI examination, and are encoded appropriately, as is known. In Step S12 the selected diffusion strength gradients are stored, and in Step S13, the first gradient is presented to the MRI apparatus for the conduct of the first MRI run in Step S14. The short diffusion time of Step S10 is also presented to control the MRI run in Step S14. The conclusion of the first run is detected in Step S15 and used to decrement a counter in Step S16 which controls the presentation of the diffusion strength gradients to the MRI. Initially, the counter is set in Step S17 to the total number x of gradients to be presented sequentially to the MRI. When the count reaches 0, the presentation of gradients to the MRI is stopped in Step S18 and the image set resulting from the MRI is complete. The output signals of the images determined by the MRI are recorded in Step S19 as an image set. Responsive to the termination of the MRI, or alternatively, simultaneously with the conduct of the MRI, the processing of the recorded image set is initiated in Step S20. In Step S21 the b values are calculated according to the equation set forth above in the specific Example. Next, in Step S22, using Equation 2 above and the calculated b values, the best fitted parameters to be determined are obtained using a preselected fitting algorithm. The signals in the set of images are fit, pixel by pixel, to obtain sets of signals corresponding to images of the parameters to be determined. The parameters to be determined may, for example, comprise certain histologic and physiologic parameters according to the invention from which a diagnosis can be made. As examples of such parameters are the ADCe and ADCe and the EVF. The output signals from Step S22 are captured in Step S23, and in Step S24 are mapped at high resolution as output parameter images. In step S25 a determination diagnosis is made from the image representations of the mapped images. In the time saving protocol referred to above, only high b values are used in Step S22 to obtain the output images from which the diagnosis is made.

A block diagram of the apparatus is shown in Figure 5 and consists of an MRI machine 20 having a controller 22 to set the diffusion time and the preselected diffusion strength gradients. The output of the MRI machine is a set of images in digital format that is recorded in recorder 24. A calculator 26 is connected to receive the set of images to calculate the b values as discussed above. A processor 28 receives the b values and uses them in the algorithm of Equation 2 to obtain the desired parameters using a best fitted algorithm as explained above. A mapper 30 receives the output of the processor and maps the parameters according to the processed signals. Finally, a visual output 32 is made such as by a printer to show the mapped parameters which may be color coded to easily distinguish features.

Although the present invention has been shown and described in terms of preferred embodiments, nevertheless changes and modifications are possible which do not depart from the teachings herein. Such changes and modifications that are evident to one skilled in the art from the teachings herein are deemed to fall within the purview of the invention as claimed.

Claims

WHAT IS CLAIMED IS:

1. A method for analyzing tissue comprising the step of:

(a) conducting a high resolution diffusion MRI of a tissue using a short constant time of diffusion and a plurality of diffusion strength gradients to obtain a recorded image set,

(b) determining b values for the recorded image set based on the diffusion strength gradients,

(c) obtaining best fitted parameters using a fitting algorithm to fit signals in the set of images, pixel by pixel, to an algorithm expressed as

I(b)/ (0) = Σ[NiKi exp(-bDi)]/ ΣNiKi

Where i labels the compartment, N is the compartment volume and D the apparent diffusion coefficient (ADC) and where:

p is the proton density in the compartment, and T_\ and T₂ are the intrinsic water relaxation times.

2. The method of claim 1 including the further step of analyzing the mapped high resolution images of the parameters to make a diagnosis of the condition of the tissue.

3. The method of claim 1 wherein the b values are determined according to an algorithm expressed as

: b = γ² (G_dδ² (Δ — ) + 2AG_rβG_dδ) where y is the gyromagnetic ratio, G_r is the

amplitude of the imaging gradient in the read direction and β is the duration of G_r.

4. The method of claim 1 wherein the Extracellular volume fraction is mapped as the principal parameter of interest.

5. The method of claim 1 wherein the parameters mapped include the intraceilular and extracellular apparent diffusion coefficients, the "psuedo-diffusion coefficient", the intraceilular, extracellular and intravascular volume fractions.

6. The method of claim 1 wherein the time of diffusion is shorter than the average lifetime of water inside cells, preferably at least a factor of 2 shorter or more with no limit for the lowest time (which must be greater than 0).

7. The method of claim 1 wherein the b values are those values in which the diffusion of the extracellular water is almost completely attenuated and resulting signal is due essentially to intraceilular water.

8. Apparatus for analyzing tissue comprising,

(a) an MRI machine for conducting a high resolution MRI of tissue to produce and record an image set in digital form,

(b) a controller for controlling the MRI machine to set the time of diffusion to a short constant time and diffusion strength gradients to predetermined values,

(c) a calculator for determining b values based on the diffusion strength gradients,

(d) a processor for obtaining best fitted parameters using a preselected fitting algorithm to fit signals in said set of images, pixel by pixel, to an algorithm expressed as

I(b)/I(0) = ∑rViKi exp(-bDi)]/ ΣNiKi

K_t = -(l - exp(- ex - )

p is the proton density in the compartment, and Ti and T are the intrinsic water relaxation times. (d) a mapper for mapping the output parameters as high resolution images.

9. The apparatus according to claim 8 wherein for two compartments, extracellular and intraceilular, designated e and c respectively, the algorithm of step (d) is expressed as:

1(b) ₌ (EVF)K_e Qxp(-bPf^p ) + (l -EVF)K_c exp(-bPf^p ) 1(0) ^" (EVF)K_e + (1 - EVF)K_C

10. The apparatus of claim 8 wherein the b values are determined according to an algorithm expressed as

: b = γ²(G²,δ²(A — ) + 2AG_rβG_dδ) where χ is the gyromagnetic ratio, G_f G_d is the

amplitude of the imaging gradient in the read direction and β is the duration of G^.

11. The apparatus of claim 8 wherein the Extracellular volume fraction is mapped as the principal parameter of interest.

12. The apparatus of claim 8 wherein the parameters mapped include the intraceilular and extracellular apparent diffusion coefficients, the "psuedo- diffusion coefficient", the intraceilular, extracellular and intravascular volume fractions.

13. The apparatus of claim 8 wherein the time of diffusion is shorter than the average lifetime of water inside cells, preferably at least a factor of 2 shorter or more with no limit for the lowest time (which must be greater than 0).

14. The apparatus of claim 8 wherein the b values are those values in which the diffusion of the extracellular water is almost completely attenuated and resulting signal is due essentially to intraceilular water.