Búsqueda Imágenes Maps Play YouTube Noticias Gmail Drive Más »
Iniciar sesión
Usuarios de lectores de pantalla: deben hacer clic en este enlace para utilizar el modo de accesibilidad. Este modo tiene las mismas funciones esenciales pero funciona mejor con el lector.

Patentes

  1. Búsqueda avanzada de patentes
Número de publicaciónCN1477581 A
Tipo de publicaciónSolicitud
Número de solicitudCN 03132141
Fecha de publicación25 Feb 2004
Fecha de presentación1 Jul 2003
Fecha de prioridad1 Jul 2003
También publicado comoCN1234092C
Número de publicación03132141.0, CN 03132141, CN 1477581 A, CN 1477581A, CN-A-1477581, CN03132141, CN03132141.0, CN1477581 A, CN1477581A
Inventores周志华
Solicitante南京大学
Exportar citaBiBTeX, EndNote, RefMan
Enlaces externos:  SIPO, Espacenet
Predictive modelling method application to computer-aided medical diagnosis
CN 1477581 A
Resumen
The present invention discloses a prediction model-building method for computer-aided medical diagnosis. It utilizes the medical symptom detection equipment to obtain the symptom of object to be diagnosed to form symptomatic vector, then utilizes the prediction model to make treatment so as to can obtain prediction result. It also provides its concrete steps for implementing said invented prediction model-building method.
Reclamaciones(2)  traducido del chino
1.一种适用于计算机辅助医疗诊断的预测建模方法,包括通过医学症状检测设备获取待诊对象的症状,然后将症状进行量化得到症状向量[t1,t2,…,tn],其中tn表示第n个症状值,症状向量交给预测模型处理,即可得到预测结果及解释的数字化表示形式,其特征是该方法包括以下步骤:(1)若预测模型未训练好,则执行步骤(2),否则转到步骤(6);(2)利用历史病例产生初始训练数据集;(3)利用初始训练数据集训练出一个神经网络集成;(4)利用神经网络集成对初始训练数据集进行处理以产生规则训练数据集;(5)利用规则学习技术从规则训练数据集中产生规则模型;(6)利用规则模型进行预测并给出结果及解释;(7)结束。 A method suitable for computer-aided medical diagnosis prediction modeling method, comprising obtaining a medical condition by detecting an object device to be patient symptoms, and symptoms of the symptoms quantized vector [t1, t2, ..., tn], where tn represents n-th value of symptoms, symptom vector to the prediction model processing, interpretation and prediction results can be obtained digitized representation, characterized in that the method comprises the steps of: (1) If the prediction model is not well trained, step (2 ), otherwise go to step (6); (2) the use of historical cases to generate the initial training data set; (3) using the initial training data set to train a neural network ensemble; (4) the use of neural network ensemble initial training data set processing the training data set to generate a rule; (5) the use of rule generation rule learning techniques focus from training data model rule; (6) the use of the rules of the prediction model and the results are given and explanation; (7) end.
2.根据权利要求1所述的适用于计算机辅助医疗诊断的预测建模方法,其特征是:在(4)中,利用神经网络集成产生用于建立规则模型的规则训练数据集L1的步骤是:(1)将L1置为空集;(2)从初始训练数据集L0中获取一个症状向量及其类别;(3)为每个类别分别设置一个计数器,用来记录神经网络给出的同类别预测结果的数目;(4)将所有计数器清零;(5)将控制参数k置为1,k是一个大于等于1但小于等于神经网络集成中神经网络的个数N;(6)取得神经网络集成中第k个神经网络对待诊症状向量给出的预测结果Fk;(7)将Fk所对应的类别的计数器加1;(8)将k加1;(9)判断k是否小于等于神经网络集成中神经网络的个数N,如果是则表明还有其他神经网络尚未考察,转到步骤(6);否则执行步骤(10);(10)对所有计数器中的值进行比较,找出值最大的计数器,并将其对应的类别作为当前症状向量的新类别;如果有多个计数器中的值均为最大值,则以这些计数器对应的类别中出现机会最大的疾病种类作为当前症状向量的新类别;(11)将当前症状向量及其新类别加入L1;(12)判断L0中是否还有未考察的症状向量,如果有则转到步骤(2);否则进入步骤(13);(13)结束。 2. The applicable claim 1 in predictive modeling of computer-aided medical diagnosis, which is characterized by: in (4), the use of neural network ensemble L1 step rule for establishing the rules of the training data set generation model is : (1) the L1 is set to the empty set; (2) to obtain a vector of its category from the initial symptoms of the training data set L0; and (3) are set for each category as a counter to record the neural network with given The number of class prediction results; (4) all counter is cleared; (5) the control parameter k is set to 1, k is a number greater than 1 but less than or equal equal neural network ensemble neural network number N; (6) to obtain neural network ensemble neural network for the first k vectors are given to treat the symptoms diagnosed predictions Fk; (7) the corresponding category Fk counter plus 1; (8) the k plus 1; if (9) to determine k less than or equal The number of neural network ensemble neural network N, if it means that there is yet another neural network study, go to step (6); otherwise step (10); (10) all counter values, find the maximum value of the counter, and the corresponding category as the current vector of a new category of symptoms; if there are multiple counter values are maximum, the best chance of disease emergence of these types of places counters corresponding category as current symptoms The new category vector; (11) the current vector and its new category of symptoms L1; (12) to determine whether there are not investigated L0 symptoms vector, if go to step (2); otherwise, step (13) ; (13) end.
Descripción  traducido del chino
一种适用于计算机辅助医疗诊断的预测建模方法 One for computer-aided medical diagnosis predictive modeling methods

一、技术领域本发明涉及一种计算机辅助医疗诊断装置,特别涉及一种利用神经网络集成技术和规则学习技术的高精度、高可理解性预测建模方法。 First, the present invention relates to a computer-aided medical diagnostic devices, and more particularly to high-precision, high intelligibility prediction method using neural network modeling technology integration and rule learning techniques.

二、背景技术 II BACKGROUND ART

随着计算机技术的发展,计算机辅助医疗诊断装置由于不受疲劳、情绪等因素的影响,已成为重要的辅助诊断手段。 With the development of computer technology, computer-aided medical diagnostic device unaffected due to fatigue, mood and other factors, has become an important means of diagnosis. 计算机辅助医疗诊断装置通常是利用一些预测建模方法对历史病例进行分析,从而建立预测模型,然后再用该预测模型来对新病例进行诊断,其结果提交给医学专家进行进一步的分析确诊,从而在一定程度上减轻医学专家的工作负担。 Computer-aided medical diagnostic devices usually use some predictive modeling methods for analysis of historical cases, in order to establish the prediction model, and then use the model to predict the diagnosis of new cases, and the results submitted to the medical experts for further analysis confirmed order to some extent, reduce the workload of medical experts. 因此,预测建模方法是计算机辅助医疗诊断装置的关键。 Therefore, predictive modeling approach is the key to computer-aided medical diagnostic device. 一方面,由于医疗诊断务求精确,因此适用的预测建模方法必须具有很高的精度;另一方面,由于医疗诊断事关被诊者的身体健康和生命安全,因此适用的预测建模方法必须具有很高的可理解性,即在作出诊断结论之后还需要能提供对诊断的解释,这不仅是被诊者及其家属的需要,还是医学专家检查诊断过程的需要。 On the one hand, due to medical diagnostics to ensure accurate and therefore must apply predictive modeling method has high accuracy; on the other hand, due to the medical diagnosis is a matter of the health and safety of those attending, so apply predictive modeling methods must with high intelligibility, that diagnosis is made after the conclusion also need to be able to provide diagnostic interpretation, which is not only to be diagnosed and their families need, or need medical expert diagnosis process. 然而,现有技术如神经网络等虽然具有高精度,但不具有高可理解性;而规则学习等虽然具有高可理解性,但却不具有高精度,这就对计算机辅助医疗诊断装置的性能造成了不利影响。 However, the prior art such as neural networks, although with high precision, but does not have high intelligibility; the rule learning, while having high intelligibility, but do not have high accuracy, which on the performance of computer-assisted medical diagnostic apparatus adversely affected.

三、发明内容 III SUMMARY OF THE INVENTION

本发明的目的是针对现有技术难以产生适用于计算机辅助医疗诊断装置的高精度、高可理解性预测模型的问题,提供一种高精度、高可理解性的预测建模方法,以辅助提高计算机辅助医疗诊断装置的性能。 The purpose of the present invention is suitable for difficult problems of computer-aided medical diagnostic device with high accuracy and high intelligibility prediction model for the existing technology to provide a high-precision, high comprehensibility predictive modeling methods to assist improve Computer Aided medical diagnostic apparatus.

为实现本发明所述目的,本发明提供一种利用机器学习中的神经网络集成技术和规则学习技术进行预测建模的方法,该方法包括以下步骤:(1)若预测模型未训练好,则执行步骤2,否则转到步骤6;(2)利用历史病例产生初始训练数据集;(3)利用初始训练数据集训练出一个神经网络集成;(4)利用神经网络集成对初始训练数据集进行处理以产生规则训练数据集;(5)利用规则学习技术从规则训练数据集中产生规则模型;(6)利用规则模型进行预测并给出结果及解释;(7)结束。 To achieve the objective of the present invention, the present invention provides a method of modeling to predict the use of machine learning neural network technology integration technology and rule learning, the method comprising the steps of: (a) if the training is not a good predictive model, the step 2. Otherwise, go to Step 6; (2) the use of historical cases to generate the initial training data set; (3) using the initial training data set to train a neural network ensemble; (4) the use of neural network ensemble initial training data set processing the training data set to generate a rule; (5) the use of rule generation rule learning techniques focus from training data model rule; (6) the use of the rules of the prediction model and the results are given and explanation; (7) end.

本发明的优点是为计算机辅助医疗诊断装置提供了一种高精度、高可理解性的预测建模方法,以辅助提高计算机辅助医疗诊断装置的性能。 Advantage of the present invention is to provide a computer-aided medical diagnosis apparatus a high precision, high intelligibility prediction methods, to improve the performance of computer-assisted medical diagnostic aid device.

下面将结合附图对最佳实施例进行详细说明。 Below in conjunction with the accompanying drawings of the preferred embodiments described in detail.

四、附图说明 IV BRIEF DESCRIPTION

图1是计算机辅助医疗诊断装置的工作流程图。 Figure 1 is a flowchart of the operation of computer-aided medical diagnosis apparatus.

图2是本发明方法的流程图。 Figure 2 is a flowchart of a method of the present invention.

图3是用神经网络集成产生规则训练数据集的流程图。 Figure 3 is a flow chart of the rules produce neural network ensemble training data set.

五、具体实施方式 V. DETAILED DESCRIPTION

如图1所示,计算机辅助医疗诊断装置利用医学症状检测设备例如体温、血压测量设备等获取待诊对象的症状例如体温、血压等,然后将症状进行量化以得到症状向量,例如[t1,t2,…,tn],其中t1表示第一个症状值,t2表示第二个症状值,依此类推。 1, a medical diagnostic apparatus utilizing computer-aided medical condition detecting apparatus e.g. body temperature, blood pressure measuring apparatus, etc. to obtain the object to be patient symptoms such as body temperature, blood pressure, then the symptoms will be quantized to obtain symptom vectors, e.g. [t1, t2 , ..., tn], where t1 denotes a symptom value, t2 represents the second symptom values, and so on. 症状向量交给预测模型处理,即可得到预测结果及解释的数字化表示形式,经过文字化处理后,就产生了最后提交给用户的诊断结论及解释。 Vector prediction model to deal with the symptoms, you can get predictions and explanations of digital representation, after word processing, it creates a final submission to the user and interpret diagnostic conclusions.

本发明的方法如图2所示。 The method of the present invention is shown in Figure 2. 步骤10是初始动作。 Step 10 is the initial action. 步骤11判断预测模型是否已经训练好,若已训练好则可处理诊断任务,执行步骤16;否则需进行训练,执行步骤12。 11 steps to determine whether the prediction model has been training well, good training can handle Ruoyi diagnostic tasks, step 16; otherwise, the need for training, step 12. 步骤12利用历史病例产生初始训练数据集,为叙述方便,称初始训练数据集为L0。 Step 12 using historical cases to generate the initial training data set, for narrative convenience, said the initial training data set for L0. L0中包含了每一历史病例所对应的症状向量及其类别,即诊断出的具体疾病类别(“没有疾病”也作为一种类别)。 L0 contains the history of each case and the corresponding vector symptom categories, namely, the specific diagnosis of disease categories ("no disease" is also used as a category). 步骤13利用统计学中常用的可重复取样技术从L0中产生N个数据集,并用这N个数据集中的每一个训练出一个神经网络,这些神经网络就组成了神经网络集成。 Step 13 in the common use of statistical sampling techniques can be repeated N data sets generated from L0 and treated with the N data sets for each train a neural network, the neural network is composed of a neural network ensemble. N是一个用户预设的整数值例如9,它确定了神经网络集成所包含的神经网络个数。 N is an integer value preset by the user, for example 9, which determines the number of neural network neural network ensemble contains. 这里使用的神经网络可以是任何类型的神经网络,只要可以执行预测任务即可,例如可以使用神经网络教科书中介绍的多层前馈BP网络。 As used herein, the neural network may be any type of neural network, the prediction may be performed as long as the task can, for example, using a multi-layer neural network is described in the textbook feedforward BP network. 步骤14利用神经网络集成产生用于建立规则模型的规则训练数据集L1,该步骤将在后面的部分结合图3进行具体介绍。 Step 14 production rules using neural network ensemble training data set L1 model used to establish the rules, this step will be combined in a later section 3 specific introduction.

图2的步骤15利用L1训练出规则模型。 Step 2 of 15 rules out the use of L1 training model. 规则模型是一个由很多条IF-Then或类似形式的规则组成的预测模型,它由某种规则学习方法从某个训练数据集(这里就是L1)中训练出来。 Rule model is a predictive model by many IF-Then strip or similar form of rules, which consists of some kind of rule learning from a training data set (this is L1) in trained. 这里可以使用任何类型的规则学习方法,只要其产生的模型可以执行预测任务即可,例如可以使用机器学习教科书中介绍的RIPPER、C4.5 Rule等。 Here you can use any type of rule learning method, as long as it produces the model to predict the task can be performed, for example, can be used RIPPER machine learning textbooks introduced, C4.5 Rule and so on. 步骤16接收待诊断的症状向量。 Step 16 receives the symptoms to be diagnosed vector. 步骤17将症状向量提交给训练好的规则模型进行预测。 Step 17 will be submitted to the trained symptom vector prediction rule model. 步骤18给出规则模型产生的预测结果及预测过程中使用的规则,这些规则就组成了对该预测结果的解释。 Rules prediction and forecasting process steps used in 18 models produced given the rules, these rules on the composition of the interpretation of the predicted results. 步骤19是结束状态。 Step 19 is the end state.

由于本发明的方法建立的预测模型是规则模型,因此其具有高理解性;又由于该方法利用了具有高精度的神经网络集成来产生建立规则模型的训练数据集,这可以视为对初始数据集进行了去噪、增强等良性处理,因此建立的规则模型也具有高精度。 Since the method of the present invention, the prediction model is established rule model, and therefore it has high comprehension; and because the method utilizes neural network ensemble with high accuracy to produce build rule model training data set, which can be considered as the initial data sets a de-noising, enhancement benign treatment, so the rule model also has high accuracy.

图3详细说明了图2的步骤14,其作用是利用神经网络集成来产生用于建立规则模型的规则训练数据集L1。 Figure 3 details the step 14 of FIG. 2, its role is to integrate the use of neural networks to produce rules for the establishment of a training data set L1 rule model. 图3的步骤140是起始状态。 Step 140 of FIG. 3 is the initial state. 步骤141将L1置为空集。 Step 141 L1 is set to the empty set. 步骤142从图2的步骤12产生的初始训练数据集L0中获取一个症状向量及其类别。 Step 142 from step 12 of FIG. 2 to generate an initial set of training data acquired in a symptom vector L0 and categories. 步骤143为每个类别分别设置一个计数器,这些计数器用来记录有多少个神经网络给出的预测结果是该类别,这里的各类别分别对应了诊断出的具体疾病类别(“没有疾病”也作为一种类别)。 Step 143 for each category are provided a counter, which counter is used to predict the results of the number of recorded neural networks is given in this category, where each category corresponding to a diagnosis of the particular disease category ("no disease" also as one category). 步骤144将所有计数器清零。 Step 144 clears all counters. 步骤145将控制参数k置为1,k是一个大于等于1但小于等于图2中步骤13的N的一个整数值,它用来指示当前考察的神经网络的序号。 Step 145 the control parameter k is set to 1, k is equal to a greater than 1 but less than 2 in step 13 is equal to an integer value of FIG N, which is used to indicate the number of the current investigation of neural networks. 步骤146取得神经网络集成中第k个神经网络对待诊症状向量给出的预测结果,为叙述方便,称该结果为Fk。 Steps 146 to obtain the first neural network ensemble k treat neural networks to predict the results of diagnostic symptoms vector given for narrative convenience, saying that the result is Fk. 步骤147将Fk所对应的类别的计数器加一。 Step 147 Fk corresponding category plus a counter. 步骤148将k加一。 Step 148 k is incremented by one. 步骤149判断k是否小于等于神经网络集成中神经网络的个数,即图2中步骤13的N,如果是则表明还有其他神经网络尚未考察,转到步骤146;否则就执行步骤150。 Step 149 is less than the number of judges equal to k neural network ensemble neural networks, two steps that map N 13, and if it means that there are other yet to examine the neural network, go to step 146; otherwise, step 150.

图3的步骤150对所有计数器中的值进行比较,找出值最大的计数器,并将其对应的类别作为当前症状向量的新类别;如果有多个计数器中的值均为最大值,则以这些计数器对应的类别中出现机会最大的疾病种类作为当前症状向量的新类别。 Step 3 150 for all counter values are compared to find the maximum value of the counter, and the corresponding category as the current vector of a new category of symptoms; if there are multiple counter values are maximum, places These types of diseases, the biggest opportunity counters corresponding category appears as a new category of current symptoms of the vector. 步骤151将当前症状向量及其新类别加入L1。 Step 151 current symptoms vector and its new category L1. 步骤152判断L0中是否还有未考察的症状向量,如果有则转到步骤142;否则就进入步骤153,即图3的结束状态。 Step 152 determines whether or not there L0 examine symptoms vector, if there is go to step 142; otherwise, proceeds to step 153, i.e., the end state of FIG. 3.

Citada por
Patente citante Fecha de presentación Fecha de publicación Solicitante Título
CN100589754C14 Jul 200617 Feb 2010深圳迈瑞生物医疗电子股份有限公司Electrocardiogram computer-aided diagnosis system
US776960031 Oct 20073 Ago 2010Clinical Decision SupportDisease management system and method
US778059514 May 200424 Ago 2010Clinical Decision Support, LlcPanel diagnostic method and system
US799326731 Oct 20079 Ago 2011Clinical Decision Support, LlcDisease management system including a no response method
US801513829 Oct 20076 Sep 2011Clinical Decision Support, LlcComputerized medical self-diagnostic and treatment advice system
US801958220 Abr 200413 Sep 2011Clinical Decision Support, LlcAutomated diagnostic system and method
US805551623 Ago 20108 Nov 2011Clinical Decision Support, LlcPanel diagnostic method and system
US806037830 Oct 200715 Nov 2011Clinical Decision Support, LlcDisease management system and method including question version
US806663631 Oct 200729 Nov 2011Clinical Decision Support, LlcDisease management system and method including pain code
US830146725 Oct 201130 Oct 2012Clinical Decision Support, LlcPanel diagnostic method and system including active and passive strategies
US833740931 Oct 200725 Dic 2012Clinical Decision Support LlcComputerized medical diagnostic system utilizing list-based processing
US839221731 Oct 20075 Mar 2013Clinical Decision Support, LlcDisease management system and method including preview mode
US862847020 Sep 201114 Ene 2014Clinical Decision Support, LlcDisease management system and method including medication therapy self-management
US863087531 Oct 200714 Ene 2014Clinical Decision Support, LlcDisease management system and health assessment method
US866310431 Oct 20074 Mar 2014Clinical Decision Support, LlcDisease management system and method including therapeutic alterations permission level
US868269430 Oct 200725 Mar 2014Clinical Decision Support, LlcDisease management system and method including permission database
US872797631 Oct 200720 May 2014Clinical Decision Support, LlcDisease management system operating on a network
US872797931 Oct 200720 May 2014Clinical Decision Support, LlcDisease management system and method including significant symptom filtering
US873196829 Oct 201220 May 2014Clinical Decision Support, LlcPanel diagnostic method and system including automated diagnostic analysis
US874079029 Oct 20073 Jun 2014Clinical Decision Support, LlcDisease management system and method
US900511931 Oct 200714 Abr 2015Clinical Decision Support, LlcComputerized medical diagnostic and treatment advice system including network access
US908187924 Oct 200514 Jul 2015Clinical Decision Support, LlcMatrix interface for medical diagnostic and treatment advice system and method
USRE4343324 Mar 200929 May 2012Clinical Decision Support, LlcComputerized medical diagnostic and treatment advice system
USRE4354824 Mar 200924 Jul 2012Clinical Decision Support, LlcComputerized medical diagnostic and treatment advice system
Clasificaciones
Clasificación internacionalG06F19/00
Eventos legales
FechaCódigoEventoDescripción
25 Feb 2004C06Publication
5 May 2004C10Request of examination as to substance
28 Dic 2005C14Granted
6 Oct 2010C17Cessation of patent right