TWI764233B - Alzheimer's disease assessment system - Google Patents

Abstract

An Alzheimer’s disease assessment system is provided to solve the problem of a known inability to accurately diagnose Alzheimer’s disease. The system includes a data capture module obtaining a plurality of text samples of several subjects, a data preprocessing module is used to execute data pre-processing on the plurality of text samples to generate a clean data, a feature extraction module obtaining at least one importance feature data from the clean data, a data dimension reduction module is used to reduce the dimension of each at least one importance feature data to generate a training data, a processor module uses the training data as a train dataset for a neural network classifier to build a predictive model of Alzheimer’s disease, input a pending text sample into the predictive model to generate a predictive result, and an input module is used to output the predictive result.

Description

Alzheimer's disease assessment system

The present invention relates to an evaluation system, especially a system for evaluating whether a subject suffers from Alzheimer's disease by applying machine learning technology.

The development of Alzheimer's disease has swept the world, and there is a gradual upward trend. In recent years, studies have found that Alzheimer's disease not only causes memory loss, but may also cause cognitive deficits in people's visual space or other behaviors such as executive function. It is known that the way to judge whether a subject has Alzheimer's disease is generally made by physicians using scales and the clinical data of the subjects. The subjective consciousness of the doctor will cause the diagnosis to be biased, resulting in the problem of inaccurate diagnosis.

In view of this, there is still a need to improve the conventional methods of assessing whether a subject has Alzheimer's disease.

In order to solve the above problems, the purpose of the present invention is to provide an Alzheimer's disease assessment system, which can generate an Alzheimer's disease prediction model through machine learning technology.

Another object of the present invention is to provide an Alzheimer's disease assessment system, which is capable of assessing the disease degree of a subject based on the Alzheimer's disease prediction model.

The elements and components described in the full text of the present invention use the quantifier "a" or "an", It is for convenience only and to provide a general meaning of the scope of the invention; it should be read in the present invention to include one or at least one, and a single concept also includes the plural unless it is obvious that it is meant otherwise.

The "Processor Module" mentioned in the whole of the present invention refers to any electronic chip with functions of data storage, calculation and signal generation, or electronic equipment having the electronic chip. For example, the electronic chip can be a central processing unit (CPU), a microcontroller (MCU), a digital signal processor (DSP), a field programmable logic gate array (FPGA), or a system-on-chip (SoC); the The electronic device can be a Programmable Logic Controller (PLC) or Arduino UNO, which can be selected by those with ordinary knowledge in the art based on computing performance, price, volume constraints or functional requirements.

，1

i

q，其中，q：係為該臨床數據數個資料的其中一種，或該腦部核磁共振數據數個資料的其中一種；p(x _i ,y)：係表示為類別y中出現特徵x _i 的數目，與該訓練集的數目之比，該類別y係包含該病症程度數據，該特徵x _i 係包含該臨床數據或該腦部核磁共振數據；p(x _i )：係表示為特徵x _i 在該訓練集中出現的機率；p(y)：係表示為該訓練集中屬於類別y的機率；y：係表示為該訓練集的類別。 The Alzheimer's disease assessment system of the present invention includes: a data acquisition module for obtaining a test sample from each of a plurality of subjects; a data preprocessing module for performing execution on the plurality of test samples a data preprocessing to generate a clean data, the data preprocessing module has a noise filtering program, and the noise filtering program is used for performing a K-average algorithm on the plurality of test samples, so as to obtain the data The test samples are divided into several groups, and the disease degree that occurs most frequently in each group is used as an indicator, and the test samples that do not belong to the disease degree of the respective group are deleted from the group to generate the clean data; a feature extraction module for executing a feature extraction algorithm based on mutual information algorithm on the clean data to obtain at least one important feature data from the clean data; a data dimension reduction module for Performing a linear discriminant analysis algorithm to reduce the dimension of at least one important feature data of the plurality of subjects to generate a training data; and a processor module coupled to the data acquisition module, the data pre-processing module A processing module, the feature extraction module and the data dimension reduction module, the processor module sequentially executes the data extraction module, the data preprocessing module, the feature extraction module and the data reduction module A dimension module to obtain the training data, the processor module uses the training data as a training set of a learning vector quantization classifier of a class of neural classifiers to establish an Alzheimer's disease prediction model, the processor model The group obtains a test sample to be detected from the data acquisition module, and inputs the test sample to be detected into the Alzheimer's disease prediction model to generate a prediction result, and the processor module transmits the prediction result to An output module makes the output module output the prediction result; the feature extraction algorithm is based on the following formula:

, 1

i

q , wherein, q : is one of the several data of the clinical data, or one of the several data of the brain MRI data; p ( x _i , y ): represented as the occurrence of the feature x _i in the category y The number of , and the ratio of the number of the training set, the category y contains the disease degree data, the feature x _i contains the clinical data or the brain MRI data; p ( x _i ): is represented as feature x The probability of _i appearing in the training set; p ( y ): represents the probability of belonging to the category y in the training set; y : represents the category of the training set.

Accordingly, the Alzheimer's disease assessment system of the present invention can, through the data preprocessing module, eliminate the test samples that are likely to cause misjudgment from the plurality of test samples to generate the clean data, and the processor By executing the feature extraction module, the module obtains at least one important feature data for predicting Alzheimer's disease from the clean data, and establishes an Alzheimer's disease prediction model accordingly. In this way, the Alzheimer's disease assessment system of the present invention has the effect of improving the accuracy of predicting Alzheimer's disease.

Wherein, each of the test samples includes at least one of a clinical data and a brain MRI data, the brain MRI data includes a disease degree data, and the disease degree data has several disease degrees. In this way, the system can comprehensively evaluate various data of the subjects to predict whether or not to suffer from Alzheimer's disease, which has the effect of improving the accuracy of predicting Alzheimer's disease.

Wherein, the data preprocessing module has a data cleaning program, and the data cleaning program is used to perform a data cleaning operation on the plurality of test samples, so as to confirm the correlation between the clinical data and the brain MRI data of the plurality of test samples. Data missing status, if the data of one of the test samples If the missing ratio is higher than a threshold, the test sample is removed from the plurality of test samples to generate the clean data. In this way, the system has the effect of removing the test samples with an excessively high proportion of missing data from the several test samples, so as to improve the accuracy of the Alzheimer's disease prediction model.

Wherein, the data preprocessing module has a consistency program, and the consistency program is used to unify the data units of the respective clinical data of the plurality of test samples, and unify the respective brain MRI data of the plurality of test samples. data unit to produce this clean data. In this way, the data units of the same type of data in the several test samples are unified to avoid affecting the accuracy of the Alzheimer's disease prediction model.

Wherein, the data preprocessing module has a covariate correction program, and the covariate correction program is used to check the age data and education data in the clinical data of each test sample, if the age data of one test sample exceeds an age range value , or the educational data of the test sample is lower than an educational level, the test sample is deleted from the plurality of test samples to generate the clean data. In this way, the system has the effect of deleting the test samples with unreasonable data in the several test samples, so as to avoid affecting the accuracy of the Alzheimer's disease prediction model.

Wherein, the data preprocessing module has a normalization program, and the normalization program is used to perform a data normalization operation on the plurality of test samples, so as to obtain the respective clinical data and brain MRI data of the plurality of test samples Scale to the [0,1] interval to produce this clean data. In this way, the system has the effect of scaling the respective data units of the several test samples within the same range to avoid affecting the accuracy of the Alzheimer's disease prediction model.

〔this invention〕

1: Data capture module

2: Data preprocessing module

21: Data Cleaner

22: Consistency Procedure

23: Common variable correction procedure

24: Regularization Procedure

25: Filter out noise program

3: Feature extraction module

4: Data dimensionality reduction module

5: Processor module

6: Output module

[FIG. 1] A system block diagram of a preferred embodiment of the present invention.

In order to make the above-mentioned and other objects, features and advantages of the present invention more obvious and easy to understand, the preferred embodiments of the present invention are exemplified below, and are described in detail as follows in conjunction with the accompanying drawings: please refer to Figure 1, It is a preferred embodiment of the Alzheimer's disease assessment system of the present invention, and includes a data acquisition module 1, a data preprocessing module 2, a feature extraction module 3, and a data dimension reduction module 4 , a processor module 5 and an output module 6, the data acquisition module 1, the data preprocessing module 2, the feature extraction module 3, the data dimension reduction module 4 and the output module 6 are respectively coupled to the processor module 5 .

The data acquisition module 1 is used for acquiring a test sample of a plurality of subjects. Each test sample may include at least one of a clinical data (Clinical) and a brain magnetic resonance data (fMRI). The partial MRI data includes a condition level data. In this embodiment, the data acquisition module 1 may include a biochemical analyzer, an immune analyzer, an MRI machine, a blood analyzer, a bacteria analyzer, and a microorganism analyzer, but not limited.

In detail, the clinical data may include blood glucose before meals (GLU-AC), alanine transaminase (GPT-ALT), total cholesterol (T-CHOL), triglycerides (TG), high density lipoprotein ( HDL-C), low density lipoprotein (LDL-C), negatively charged low density lipoprotein (L5), lipoprotein E type (APO E), angiotensin converting enzyme (ACE), left arm index (LABI), Right Bare Arm Index (RABI), Left Brachial Ankle Pulse Wave Velocity (LbaPWV), Right Brachial Ankle Pulse Wave Velocity (RbaPWV), Cognitive Impairment Screening Inventory (CASI), Brief Intelligence Test (MMSE), Orientation (Orientation), recent memory (Remote memory), concentration (Concentration), gender, age, height, weight and education and other information.

The brain MRI data is the correlation analysis data generated by performing MRI on the hippocampal memory blocks of each subject's left and right brains, which can be understood by those with ordinary knowledge in the art. In this embodiment, The brain MRI data may include L.Calc-L, R.Calc-L, L.Fus-L, R.Fus-L, L.Hes-L, R.Hes-L, L.InfO-L, R.InfO-L, L.InfT-L, L.InfT-R, L. Ling-L, R.Ling-L, L.MidO-L, R.MidO-L, L.MidT-L, R.MidT-L, L.MidT-R, R.MidT-R, L.SupT- L, L.SupT-R, L.pHip-L, R.pHip-L, R.InfT-L, R.InfT-R, R.SupT-L and R.SupT-R etc.

The disease level data is used to represent the disease level of Alzheimer's disease. In this embodiment, the disease level data has several disease levels, including no dementia (NC), mild cognitive impairment (MCI), no dementia Confirmed or to be observed (AD0.5), mild dementia (AD1.0), moderate dementia (AD2.0), severe dementia (AD3.0), profound dementia (AD4.0) and terminal dementia Intelligence (AD5.0) and other disease degree.

The data preprocessing module 2 is used for performing a data preprocessing (Data Preprocessing) on the plurality of test samples to generate a clean data. Specifically, the data preprocessing may include a data cleaning program 21, a consistency program 22, a total variable correction program 23, a normalization program 24, and a noise filtering program 25. In this embodiment, the The data preprocessing module 2 is executed according to the sequence of the data cleaning procedure 21 , the consistency procedure 22 , the covariate correction procedure 23 , the normalization procedure 24 and the noise filtering procedure 25 to generate the clean data. The data cleaning program 21 is used for performing a data cleaning operation on the plurality of test samples to confirm the data missing status of the clinical data and the brain MRI data of the plurality of test samples. If the data missing ratio of the test sample is higher than a threshold, the test sample is deleted from the plurality of test samples. In this embodiment, the threshold can be set to 30%, but not limited thereto.

The consistency program 22 is used to perform consistency analysis on the plurality of test samples after the data cleaning program 21 is executed, so as to analyze whether the plurality of test samples have contradictory or incompatible data. In this embodiment, , the consistency program 22 is used for unifying the data unit of the respective clinical data of the plurality of test samples, and unifying the data unit of the respective brain MRI data of the plurality of test samples.

The covariate correction program 23 is used for checking several test samples after the consistency program 22 is executed, so as to check the age data and education data in the clinical data of each test sample, if the age data of one of the test samples exceeds An age range value, and the educational data of the test sample is lower than an educational level, the test sample is deleted from the plurality of test samples. In this embodiment, the age range value can be set to 55-80 years old, and the education level can be set to a national elementary school, but not limited thereto.

其中，X _nom ：係為該臨床數據執行正規化後的數值，或該腦部核磁共振數據執行正規化後的數值；X：係為該數個臨床數據中目前要進行運算的資料之數值，或該數個腦部核磁共振數據中目前要進行運算的資料之數值；X _max ：係為該數個臨床數據相同資料中的最大值，或該數個腦部核磁共振數據相同資料中的最大值；X _min ：係為該數個臨床數據相同資料中的最小值，或該數個腦部核磁共振數據相同資料中的最小值。 The normalization program 24 is used for performing a data normalization operation on the plurality of test samples after the covariate correction program 23 is executed, so as to obtain the clinical data and the brain MRI data of the plurality of test samples. It is scaled to the [0,1] interval. Specifically, the normalization program 24 can respectively perform normalization operations on the clinical data and the brain MRI data through the following formulas:

Wherein, X _nom : is the value after the normalization of the clinical data, or the value after the normalization of the brain MRI data; X : is the value of the data currently to be calculated among the several clinical data, Or the value of the data currently to be calculated in the brain MRI data; X _max : is the maximum value in the same data of the several clinical data, or the maximum value in the same data of the brain MRI data Xmin : the _minimum value in the same data of the several clinical data, or the smallest value in the same data of the several brain MRI data.

The noise filtering program 25 is used for performing a noise filtering operation on the plurality of test samples after the normalization program 24 is executed, so as to group the plurality of test samples according to the disease degree data. In this embodiment, the noise filtering program 25 performs an unsupervised learning data mining algorithm on the plurality of test samples, such as a K-means Clustering algorithm. , to divide the number of test samples into several For groups, the disease degree that occurs most frequently in each group is used as an index, and the test samples of the disease degree that do not belong to the respective groups are deleted from the group.

，1

i

q，其中，q：係表示為該臨床數據數個資料的其中一種，或該腦部核磁共振數據數個資料的其中一種；p(x _i ,y)：係表示為類別y中出現特徵x _i 的數目，與該訓練集的數目之比，該類別y係包含該病症程度數據，該特徵x _i 係包含該臨床數據或該腦部核磁共振數據；p(x _i )：係表示為特徵x _i 在該訓練集中出現的機率；p(y)：係表示為該訓練集中屬於類別y的機率；y：係表示為該訓練集的類別。 The feature extraction module 3 is used for executing a feature extraction algorithm on the clean data to obtain at least one important feature data from the clean data. Specifically, the feature extraction algorithm can generate the following formula based on a mutual information (Mutual Information, MI) algorithm to obtain the at least one important feature data from the clean data:

, 1

i

q , where q : represents one of the several data of the clinical data, or one of the several data of the brain MRI data; p ( x _i , y ): represents the appearance of the feature x in the category y The number of _i , the ratio of the number of the training set, the category y contains the disease degree data, the feature x _i contains the clinical data or the brain MRI data; p ( x _i ): is represented as a feature The probability that x _i appears in the training set; p ( y ): represents the probability of belonging to the category y in the training set; y : represents the category of the training set.

For example, when the feature xi includes only the clinical data _, the feature extraction module 3 calculates the correlation between the clinical data and the disease level data to generate a plurality of first results, with the largest value among them. The clinical data of a first result of , is used as the at least one important feature data. In this embodiment, the clinical data of the first result may be orientation force. Next, the feature extraction module 3 calculates the correlation between the remaining clinical data and the at least one important feature data and the disease degree data to generate several second results, if the second result with the largest value is greater than the first one A result, the clinical data of the second result is also used as the at least one important feature data, in this embodiment, the clinical data of the second result may be education. The feature extraction module 3 repeatedly performs the above steps until the calculated value is not greater than the value of the last generated result, and the at least one important feature data can be obtained from the clean data.

The data dimensionality reduction module 4 is used for dimensionality reduction of at least one important feature data of each of the plurality of subjects to generate a training data. In this embodiment, the data dimensionality reduction module 4 is A linear discriminant analysis (LDA) algorithm can be used to reduce the dimension of at least one important feature data of each of the plurality of subjects.

The processor module 5 is coupled to the data capture module 1, the data preprocessing module 2, the feature capture module 3, the data dimension reduction module 4 and the output module 6. The processor module The group 5 executes the data extraction module 1, the data preprocessing module 2, the feature extraction module 3 and the data dimension reduction module 4 in sequence to generate the training data; the processor module 5 will The training data is used as a training set of a type of neural classifier to establish an Alzheimer's disease prediction model. In this embodiment, the type of neural classifier can be a learning vector quantization classifier (Learning Vector Quantization, LVQ). , the operation mode of the learning vector quantization classifier can be understood by those with ordinary knowledge in the field, and will not be repeated here.

The processor module 5 obtains a test sample to be detected from the data acquisition module 1, and inputs the test sample to be detected into the Alzheimer's disease prediction model to generate a prediction result. The processor module 5 transmits the prediction result to the output module 6, so that the output module 6 outputs the prediction result. For example, but not limited to, the output module 6 can be a display screen.

The Alzheimer's disease assessment system of the present invention can be applied to a computer host, and the processor module 5 can obtain the respective test samples of several subjects by executing the data acquisition module 1, for example, it can be Obtain at least one of the aforementioned clinical data and the aforementioned brain MRI data through equipment such as biochemical analyzers, immune analyzers, MRI machines, blood analyzers, bacterial analyzers, and microbiology analyzers, and can be stored in A storage device of the computer host. After obtaining the plurality of test samples, the processor module 5 executes the data preprocessing module 2, so that the data cleaning program 21, the consistency program 22, and the covariate correction are sequentially executed for the plurality of test samples The program 23, the normalization program 24, and the noise filtering program 25 are used to generate the clean data, and the programs can be stored in a memory device of the computer host.

The processor module 5 executes the feature extraction module after obtaining the clean data 3. To obtain at least one important feature data from the clean data, in this embodiment, when the test sample only contains the clinical data, the important feature data can be included in the clinical data in orientation, education, Recent memory, lipoprotein E typing and negative low density lipoprotein; in addition, when the test sample only contains the brain MRI data, the important feature data can be included in the brain MRI data. L.Fus-L, R.MidT-L, R.pHip-L, L.Hes-L and R.InfO-L; and, when the test sample includes the clinical data and the brain MRI data, the Important characteristic data may include orientation, gender, lipoprotein E-type, L.Hes-L, and concentration.

The processor module 5 performs linear discriminant analysis through the data dimensionality reduction module 4 to reduce the dimension of the important feature data to generate a training data, and the processor module 5 uses the training data as the learning vector quantization classifier training set to build an Alzheimer's disease prediction model. The processor module 5 may select the accuracy as an indicator of the performance of the Alzheimer's disease prediction model. Specifically, the processor module 5 sets a first state (the subject actually has Alzheimer's disease, and the prediction result of the Alzheimer's disease prediction model is that the subject has Alzheimer's disease) ), a second state (the subject does not actually have Alzheimer's disease, but the prediction result of the Alzheimer's disease prediction model is that the subject has Alzheimer's disease), a third state (the subject has Alzheimer's disease) The test subject does not actually have Alzheimer's disease, and the prediction result of the Alzheimer's disease prediction model is that the subject does not have Alzheimer's disease) and a fourth state (the subject actually has Alzheimer's disease) Alzheimer's disease, but the prediction result of the Alzheimer's disease prediction model is that the subject does not have Alzheimer's disease). Thus, the accuracy is the ratio of the sum of the first state and the third state to the sum of the first state, the second state, the third state and the fourth state.

The accuracy results of the Alzheimer's disease prediction model established by the present invention and the prediction models generated by using other types of neural-like classifiers can be shown in the following Table 1, from which the prediction accuracy of the present invention can be known. The accuracy is better than using other types of neural-like classifiers.

Table 1: Prediction model accuracy comparison table

On the other hand, in the Alzheimer's disease assessment system of the present invention, by first executing the data preprocessing module 2 on the test sample, and then executing the feature extraction module 3, compared to executing the feature extraction module first 3. Execute the data preprocessing module 2 again, which can have better prediction accuracy of Alzheimer's disease, and the above accuracy results can be shown in the following list 2:

To sum up, the Alzheimer's disease assessment system of the present invention can, through the data preprocessing module, eliminate test samples that may cause misjudgment from the plurality of test samples to generate the clean data. By executing the feature extraction module, the processor module obtains at least one important feature data for predicting Alzheimer's disease from the clean data, and establishes an Alzheimer's disease prediction model accordingly. In this way, the Alzheimer's disease assessment system of the present invention has the ability to improve the prediction of Alzheimer's disease The power of accuracy.

Although the present invention has been disclosed by the above-mentioned preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make various changes and modifications relative to the above-mentioned embodiments without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the patent application attached hereto.

1: Data capture module

2: Data preprocessing module

21: Data Cleaner

22: Consistency Procedure

23: Common variable correction procedure

24: Regularization Procedure

25: Filter out noise program

3: Feature extraction module

4: Data dimensionality reduction module

5: Processor module

6: Output module

Claims (6)

An Alzheimer's disease assessment system, comprising: a data acquisition module for obtaining a test sample of a plurality of subjects; a data preprocessing module for executing a data on the plurality of test samples preprocessing to generate a clean data, the data preprocessing module has a noise filtering program, and the noise filtering program is used for performing a K-average algorithm on the plurality of test samples, so as to obtain the plurality of test samples The samples are divided into several groups, and the disease degree that occurs most frequently in each group is used as an indicator, and the test samples that do not belong to the disease degree of the respective group are deleted from the group to generate the clean data; a feature The extraction module is used for executing a feature extraction algorithm based on mutual information algorithm on the clean data, so as to obtain at least one important feature data from the clean data; a data dimension reduction module is used for executing a line A sex discriminant analysis algorithm performs dimensionality reduction on at least one important feature data of each of the plurality of subjects to generate a training data; and a processor module is coupled to the data acquisition module and the data preprocessing module set, the feature extraction module and the data dimension reduction module, the processor module executes the data extraction module, the data preprocessing module, the feature extraction module and the data dimension reduction module in sequence group, to obtain the training data, the processor module uses the training data as a training set of a learning vector quantization classifier of a class of neural classifiers to establish an Alzheimer's disease prediction model, the processor module consists of The data acquisition module obtains a test sample to be tested, and inputs the test sample to be tested into the Alzheimer's disease prediction model to generate a prediction result, and the processor module transmits the prediction result to an output module to make the output module output the prediction result; the feature extraction algorithm is based on the following formula:

, 1

i

q , wherein, q : is one of the several data of the clinical data, or one of the several data of the brain MRI data; p ( x _i , y ): is represented as the occurrence of the feature x _i in the category y The number of , and the ratio of the number of the training set, the category y contains the disease degree data, the feature x _i contains the clinical data or the brain MRI data; p ( x _i ): is represented as feature x The probability of _i appearing in the training set; p ( y ): represents the probability of belonging to the category y in the training set; y : represents the category of the training set. The Alzheimer's disease assessment system of claim 1, wherein each of the test samples includes at least one of a clinical data and a brain MRI data, the brain MRI data includes a disease degree data, the disease degree The data has several severity levels. The Alzheimer's disease assessment system of claim 1, wherein the data preprocessing module has a data cleaning program, and the data cleaning program is used to perform a data cleaning operation on the plurality of test samples to confirm the plurality of test samples The data missing status of the respective clinical data and brain MRI data of the test samples. If the data missing ratio of one of the test samples is higher than a threshold, the test sample will be deleted from the several test samples to generate the clean data. data. The Alzheimer's disease assessment system of claim 1, wherein the data preprocessing module has a consistency program, and the consistency program is used to unify the data units of the respective clinical data of the plurality of test samples, and to unify the data units. The data unit of the respective brain MRI data of the plurality of test samples to generate the clean data. The Alzheimer's disease assessment system of claim 1, wherein the data preprocessing module has a covariate correction program, and the covariate correction program is used to check age data and education data in the clinical data of each of the test samples, If the age data of one of the test samples exceeds an age range value, or the education data of the test sample is lower than an education level, the test sample is deleted from the plurality of test samples to generate the clean data. The Alzheimer's disease assessment system of claim 1, wherein the data preprocessing module has a normalization program, and the normalization program is used to perform a data normalization operation on the plurality of test samples, so as to obtain the data The respective clinical data and brain MRI data of each test sample are scaled to the interval [0,1] to generate the clean data.

Images