Parkinson's disease classification evaluation model modeling method and device, terminal and medium

Through the deep learning model combining self-attention and capsule layer technical means, the problem of low accuracy in classification assessment of Parkinson's disease is solved, and higher classification accuracy is achieved, especially in the distinction between early and middle- and late PD, which is helpful. for assisted clinical diagnosis.

CN118965082APending Publication Date: 2024-11-15GUANGDONG UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411269881.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-11-15

Smart Images

  • Figure CN118965082A_ABST
    Figure CN118965082A_ABST
Patent Text Reader

Abstract

The invention discloses a Parkinson's disease classification evaluation model modeling method and device, a terminal and a medium, and the method comprises the steps: obtaining PD disease sample data, carrying out the feature extraction of the PD disease sample data, obtaining the PD disease classification features of different sample individuals, carrying out the preprocessing, and carrying out the feature extraction of the PD disease classification features; according to the method, a self-attention module is introduced between a convolution layer and a main capsule layer to serve as training data to train a preset initial deep learning model, the initial model introduces the self-attention module between the convolution layer and the main capsule layer, the output of the self-attention module is sent to the main capsule layer, and feature representation is enhanced by calculating the mutual relation between input features. The dynamic routing mechanism of the capsule network further processes the weighted features, and the attention mechanism can highlight the important features in the input data, so that the subsequent capsule network can more effectively process the PD disease features, thereby improving the overall classification performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of smart wearable devices, and in particular to a method, device, terminal and medium for modeling a Parkinson's disease classification and evaluation model. Background Art

[0002] Parkinson's disease (PD), also often referred to as "tremor paralysis", is a degenerative disease of the nervous system. As a new medical device that can be used to quantify the characteristics of Parkinson's disease, the PKG watch can provide an objective and accurate basis for diagnosis by real-time monitoring of the patient's movement data and physiological indicators, providing valuable data reference for the classification of Parkinson's disease.

[0003] At present, the processing method for PKG monitoring data is mostly to compare the data according to the standards of the unified Parkinson's rating scale based on the monitoring data collected by the PKG watch, and output the quantitative results of the severity of the axial symptoms of Parkinson's patients based on the comparison results. However, in actual applications, this processing method still has technical problems such as low recognition and classification accuracy. Summary of the invention

[0004] The present application provides a modeling method, device, terminal and medium for classifying and evaluating a Parkinson's disease, which are used to solve the technical problems of low recognition and classification accuracy in the existing Parkinson's disease classification and evaluation.

[0005] In order to solve the above technical problems, the first aspect of the present application provides a method for modeling a Parkinson's disease classification and evaluation model, comprising:

[0006] Acquire PD disease sample data, and perform feature extraction from the PD disease sample data to obtain PD disease classification features and PD disease classification identifiers of different sample individuals;

[0007] Performing data preprocessing on the PD disease classification features to obtain preprocessed PD disease classification features;

[0008] The PD disease classification features and the PD disease classification identifier are used as training data to train a preset initial deep learning model. When the training effect of the initial deep learning model meets the convergence condition, a Parkinson's disease classification evaluation model is obtained, wherein the initial deep learning model includes: a convolutional layer, a self-attention module and a capsule layer, wherein the self-attention module is arranged between the convolutional layer and the main capsule layer of the capsule layer.

[0009] Preferably, the hierarchical structure of the initial deep learning model is: input layer, convolution layer, self-attention module, main capsule layer, digital capsule layer, linear layer and output layer.

[0010] Preferably, the PD disease classification characteristics specifically include: demographic characteristics, scale characteristics and tremor characteristics;

[0011] The demographic characteristics include: gender, age, and disease duration of different sample individuals;

[0012] The tremor characteristics include: bradykinesia score, dyskinesia score, movement fluctuations, tremor time proportion, and rest / sleep time proportion.

[0013] Preferably, the data preprocessing specifically includes: data cleaning processing, data encoding processing, data normalization processing and data balancing processing.

[0014] Preferably, the data balancing process specifically includes:

[0015] Traverse the PD disease classification features to determine minority class samples;

[0016] According to the minority class samples, a number of new sample data are generated through SMOTE oversampling processing, until the sample ratio meets the preset sample number balance ratio threshold.

[0017] Preferably, the data normalization processing includes: maximum and minimum normalization processing or zero mean normalization processing.

[0018] At the same time, the second aspect of the present application provides a modeling device for Parkinson's disease classification and evaluation model, comprising:

[0019] A feature acquisition unit, used to acquire PD disease sample data, and perform feature extraction from the PD disease sample data to obtain PD disease classification features and PD disease classification identifiers of different sample individuals;

[0020] A feature preprocessing unit, used for performing data preprocessing on the PD disease classification feature to obtain a preprocessed PD disease classification feature;

[0021] A classification evaluation model training unit is used to train a preset initial deep learning model using the PD disease classification features and the PD disease classification identifier as training data. When the training effect of the initial deep learning model meets the convergence condition, a Parkinson's disease classification evaluation model is obtained, wherein the initial deep learning model includes: a convolutional layer, a self-attention module and a capsule layer, wherein the self-attention module is arranged between the convolutional layer and the main capsule layer of the capsule layer.

[0022] Preferably, the hierarchical structure of the initial deep learning model is: input layer, convolution layer, self-attention module, main capsule layer, digital capsule layer, linear layer and output layer;

[0023] The PD disease classification characteristics specifically include: demographic characteristics, scale characteristics and tremor characteristics;

[0024] The demographic characteristics include: gender, age, and disease duration of different sample individuals;

[0025] The tremor characteristics include: bradykinesia score, dyskinesia score, motor fluctuations, tremor time proportion, and rest / sleep time proportion;

[0026] The data preprocessing specifically includes: data cleaning processing, data encoding processing, data normalization processing and data balancing processing.

[0027] A third aspect of the present application provides a Parkinson's disease classification and assessment model modeling terminal, comprising: a memory and a processor;

[0028] The memory is used to store program codes, and the program codes correspond to a modeling method for Parkinson's disease classification and assessment model provided in the first aspect of the present application;

[0029] The processor is used for reading and executing the program code.

[0030] The fourth aspect of the present application provides a computer-readable storage medium, in which program code is stored. When the program code is read and executed by a processor, a Parkinson's disease classification and assessment model modeling method as provided in the first aspect of the present application is implemented.

[0031] It can be seen from the above technical solutions that this application has the following advantages:

[0032] The solution provided in the present application first obtains PD disease sample data and extracts features from the PD disease sample data to obtain PD disease classification features of different sample individuals. After preprocessing, the features are used as training data to train a preset initial deep learning model. The initial model of the present application introduces a self-attention module between the convolutional layer and the main capsule layer. The output of the self-attention module is sent to the main capsule layer, and the feature representation is enhanced by calculating the relationship between the input features. The dynamic routing mechanism of the capsule network further processes these weighted features. The attention mechanism can highlight the important features in the input data, so that the subsequent capsule network can process these PD disease features more effectively, thereby improving the overall classification performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0034] Figure 1 A schematic diagram of a flow chart of an embodiment of a method for building a Parkinson's disease classification and assessment model provided in this application.

[0035] Figure 2 A flow chart of a method for modeling a Parkinson's disease classification and assessment model provided in this application.

[0036] Figure 3 A schematic diagram of the network structure of a Parkinson's disease classification and assessment model provided in this application.

[0037] Figure 4 This is a graph showing the results of a five-fold cross-validation experiment on the Parkinson's disease classification evaluation model based on the present application.

[0038] Figure 5 This is a structural schematic diagram of an embodiment of a modeling device for Parkinson's disease classification and assessment model provided in the present application.

[0039] Figure 6 A schematic diagram of the structure of a modeling terminal embodiment of a Parkinson's disease classification and assessment model provided in this application. DETAILED DESCRIPTION

[0040] The embodiments of the present application provide a Parkinson's disease classification and evaluation model modeling method, device, terminal and medium, which are used to solve the technical problems of low recognition and classification accuracy in the existing Parkinson's disease classification and evaluation.

[0041] In order to make the purpose, features, and advantages of the invention of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described below are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0042] First, a detailed description of an embodiment of a Parkinson's disease classification assessment model modeling method provided by the present application is as follows:

[0043] See also Figure 1 and Figure 2 This embodiment provides a Parkinson's disease classification and assessment model modeling method, including:

[0044] Step 101, obtaining PD disease sample data, and extracting features from the PD disease sample data to obtain PD disease classification features and PD disease classification identifiers of different sample individuals;

[0045] Step 102, performing data preprocessing on the PD disease classification features to obtain preprocessed PD disease classification features;

[0046] Step 103: using the PD disease classification features and the PD disease classification identifier as training data to train a preset initial deep learning model, and when the training effect of the initial deep learning model meets the convergence condition, a Parkinson's disease classification evaluation model is obtained;

[0047] Among them, the initial deep learning model includes: a convolutional layer, a self-attention module and a capsule layer, wherein the self-attention module is arranged between the convolutional layer and the main capsule layer of the capsule layer. More specifically, the hierarchical structure of the initial deep learning model is: input layer, convolutional layer, self-attention module, main capsule layer, digital capsule layer, linear layer and output layer.

[0048] It should be noted that, first, in step 101, by acquiring PD disease sample data and performing feature extraction from the PD disease sample data, PD disease classification features and PD disease classification identifiers of different sample individuals are obtained respectively;

[0049] Among them, the PD disease classification characteristics specifically include: 3 demographic characteristics, 9 scale characteristics and 12 tremor characteristics, as shown in Table 1.

[0050]

[0051] More specifically, the tremor characteristics in the samples of this embodiment can be collected through the PKG system, which can be used to objectively evaluate the symptoms of tremor, bradykinesia and dyskinesia in PD patients. The PKG system consists of two parts: a PKG watch and a reporting system. The appearance of the PKG watch is similar to that of an ordinary watch, which is more easily accepted by patients, and it can record continuously for one week when fully charged, so one week is generally used as a wearing cycle. The device has low requirements for patients and only needs to be worn normally. The PKG watch will continue to record the wearer's tremor, movement fluctuations and other parameters every two minutes and upload the data to the reporting system. After the wearing period, the bradykinesia score (BKS), movement disorder score (DKS), movement fluctuation (FDS), tremor time proportion (PTT), static / sleepiness time proportion (PTI) and other data are obtained according to a specific algorithm, and the report is given in detail in the form of graphics and numbers.

[0052] The scale characteristics in the sample of this embodiment mainly include: UPDRS, MMSE, MoCA, etc. Among them, UPDRS is a widely used PD assessment scale internationally, which consists of four parts, covering the assessment of PD motor symptoms and non-motor symptoms. UPDRS-Ⅰ is used to evaluate the non-motor symptoms of PD; UPDRS-Ⅱ is used to evaluate the impact of PD's motor symptoms on daily life; UPDRS-Ⅲ is used to evaluate the motor function of PD patients. In addition, since PD patients also have non-motor symptoms such as hyposmia and sleep disorders, this embodiment also collects data from multiple other scales of the tester, such as: Mini Mental State Evaluation Scale (MMSE), Montreal Cognitive Assessment (MoCA) Scale, Parkinson's Non-Motor Symptom Assessment Scale (NMSS), End-of-Dose Phenomenon Questionnaire (WOQ-9), RBD Screening Scale (RBDSQ) and Smell Stick-Odor Identification Ability Test (SS-12) Scale, etc. MMSE and MoCA are used to assess the degree of cognitive dysfunction in the elderly; NMSS is used to assess the severity of overall non-motor symptoms; WOQ-9 is used to identify wearing-off phenomenon; RBDSQ is used to assess sleep disorders; and the Smell Stick-Odor Identification Test (SS-12) is used to assess olfactory impairment.

[0053] The verification example provided in this embodiment is implemented based on a data set consisting of 100 PD patients aged 33-84 years, and each record corresponds to an individual. The data collection and evaluation were completed by two movement disorder experts with rich experience in PD diagnosis. 35 patients with HY stage 1-2.5 were defined as early PD patients, and the remaining 65 patients formed the mid-to-late PD group. As shown in Table 2, the HY stage was determined about one hour after the patient took the medicine. In addition, the gender, age and disease course information of the tester must also be recorded.

[0054]

[0055] In Table 2, μ and σ represent the mean and variance, respectively. The difference in gender between the early PD group and the mid-late PD group was estimated based on the chi-square test, and the differences in other characteristics were estimated using the independent sample t test. p < 0.05 represents a statistically significant difference, and the table has been used Annotation.

[0056] Next, in step 102, the data obtained in step 101 is preprocessed to improve the data quality so as to improve the effect of subsequent model training. Specifically, the preprocessing methods of the data set in this embodiment include: data cleaning, data encoding, data normalization and data balancing. If the tremor and scale data set has 8 columns of features with a small number of missing values, 7 of which are continuous numerical features such as scales, and the mean filling method is adopted, and there is 1 column for "the position of the wristwatch wearing hand", and the mode filling method can be adopted. The entire data cleaning process in the method of this embodiment can be completed by python.

[0057] Data encoding mainly focuses on features or labels that are not numerical data, so data encoding needs to be done before data modeling. For the label column of the tremor and scale dataset, "1" is used to represent "mid-late PD" and "0" is used to represent "early PD". For the "position of the hand where the watch is worn" feature column, "1" is used to represent "right hand" and "0" is used to represent "left hand". This study uses the minimax normalization method to compress the features of both datasets to between [0, 1].

[0058] Data normalization refers to scaling the original data to eliminate the impact of the large distribution range of different features. There are mainly the following ways to normalize the data:

[0059] (1) Minimax normalization:

[0060] The core idea of ​​this method is to make linear changes to the original data and compress the data to between [0, 1] or [-1, 1]. The formula for compressing data to between [0, 1] is:

[0061]

[0062] In the formula, and Represent the original data and the processed data respectively. and Respectively represent the minimum and maximum values ​​of the original data.

[0063] The formula for compressing data to between [-1, 1] is:

[0064]

[0065] In the formula, Represents the average value of the original data.

[0066] (2) Zero mean normalization:

[0067] Zero mean normalization is also called z-score normalization, which can make the data mean 0 and variance 1. The specific formula is as follows:

[0068]

[0069] In the formula, and They represent the mean and variance of the original data respectively.

[0070] Data balancing is used to perform random interpolation between minority class samples according to the imbalance ratio of samples, increase samples, and balance the ratio of different samples. It can be implemented based on the SMOTE oversampling algorithm. The working principle and implementation process of SMOTE are as follows:

[0071] 1) Calculate each minority class sample The Euclidean distance with all other minority class samples is obtained The closest samples ( is a hyperparameter, usually =5);

[0072] 2) For each minority class sample ,from Randomly select a sample from the nearest neighbor samples ;

[0073] 3) In and A point is randomly selected between the lines of to obtain a new synthetic sample, that is, the feature vector of the new synthetic sample is and The weighted average of , where the weights are random values;

[0074] 4) Repeat steps 1) to 3) to generate a specified number of new samples to achieve data balance.

[0075] Next, in step 103, the evaluation model is trained using the data preprocessed in step 102. The specific implementation example is as follows:

[0076] The data of 100 PD patients collected initially were processed by data preprocessing operations such as data balancing to obtain data in the form of (130, 24). The data set was first input into the initial model for batch training. The batch size is a hyperparameter. After multiple experiments, the optimal batch size was selected as 30. In addition, techniques such as learning rate decay and early stopping were used to improve the performance and generalization ability of the model. This experiment used stratified five-fold cross-validation to achieve data set division and performance evaluation. In the stratified five-fold cross-validation, the data set was divided into 5 parts, and the training was carried out in 5 times. Each time, 80% of the data was used as the training set, and the remaining 20% ​​was used as the test set. In each split, the proportion of samples of the two different categories in the data set was roughly equal or similar.

[0077] This experiment builds an initial model based on the capsule network and implements it using the Pytorch deep learning framework. The layers from the input layer to the output layer are: convolutional layer, self-attention module, main capsule layer, digital capsule layer, and linear layer.

[0078] Among them, a convolution layer is applied to the input data to extract features, and the output of the convolution layer is passed to the self-attention module. The attention module calculates the degree of correlation between features and highlights important features. The features processed by self-attention are passed to the main capsule layer. The output of the main capsule layer is passed to the digital capsule layer. The digital capsule linear layer is prepared, and the linear layer maps the output of the capsule layer to the final classification result.

[0079] like Figure 3 As shown in the figure, the input data comes from the 24 features of the tester. Since the input is a one-dimensional sequence, the convolution of the first dimension is set, the convolution kernel size is set to 3, and the step size is set to 1. According to the following formula, the output after the first convolution layer is (64, 22).

[0080]

[0081] The attention mechanism enhances feature representation by calculating the relationship between input features. Its core steps are as follows:

[0082] 1): Generate query, key and value matrices: Perform linear transformation on the input features to generate query matrix (Query), key matrix (Key) and value matrix (Value). q , W k and W v is a learnable weight matrix.

[0083]

[0084] 2): Calculate the attention weight, calculate the dot product of the query and the key, get the attention weight, and then normalize it through the Softmax function. Among them, dk is the dimension of the key vector, which plays a scaling role.

[0085]

[0086] 3): Weighted summation: Use the attention weights to perform weighted summation on the value matrix to obtain the final attention output.

[0087]

[0088] Step 4: Residual connection: Add the inputs to form a residual connection to keep part of the original features.

[0089]

[0090] The shape of the input to the attention layer is: [BATCH_SIZE, NUM_CHANNELS, SEQ_LENGTH] where BATCH_SIZE is the batch size, NUM_CHANNELS is the number of output channels of the convolutional layer, and SEQ_LENGTH-2 is the length of the sequence after convolution.

[0091] Output shape of the attention layer: The output shape of the self-attention mechanism is the same as the input shape, which is [BATCH_SIZE, NUM_CHANNELS, SEQ_LENGTH], and the output shape is [30, 64, 98].

[0092] The main capsule layer exists mainly to prepare for the input of the digital capsule layer. The main capsule layer consists of 8 ordinary one-dimensional convolutional layers, with the convolution kernel size set to 3 and the stride length set to 2, and the output is 320 vectors of length 8.

[0093] Since this experimental model is a binary classification model, the output of the digital capsule layer is set to two vectors, which can represent the similarity between the input and two different types. The two vectors are combined and flattened and then input into the linear layer. The number of output neurons is set to 1, and finally the classification results are normalized by the sigmoid function.

[0094] In the binary classification experiment of early PD and mid-late PD, the classification results higher than or equal to 0.5 are regarded as mid-late PD, and the rest are early PD.

[0095] When the training degree of the model meets the preset convergence condition or iteration number condition, the current model is output as the final Parkinson's disease classification and evaluation model, and the model construction is completed.

[0096] The following is an example of validating the trained Parkinson's disease classification evaluation model based on a preset data set. The specific results are as follows:

[0097] The hardware environment configuration and details of the experiment are as follows: the PD-CapsNet model provided in this example is implemented using the Pytorch deep learning framework, SVM, Xgboost and RF are implemented using the Scikit-learn machine learning library, the operating system is Windows 10, and the programming language is Python 3.7. The cross entropy loss function is used as the objective function, and the initial learning rate is set to After every 10 rounds of training, the learning rate decays to 1 / 5 of the original. The batch size is set to 30, the RMSprop optimization algorithm is selected as the optimization strategy, and the early stopping technique is used to prevent the model from overfitting. The performance comparison results of the algorithm model constructed in this embodiment and other algorithms are shown in Table 3:

[0098]

[0099] This embodiment proposes a capsule network model fused with a self-attention mechanism. The model integrates the self-attention mechanism in the convolutional layer and the main capsule layer to innovate and achieve the binary classification of Parkinson's patients into early and middle and late stages. Its precision and recall rates have been significantly improved. From the ROC curve in the stratified five-fold cross-validation process, the average AUC score is 0.85 and the average accuracy is 86.15%. The model has certain significance for doctors' auxiliary diagnosis. To further observe the stability and reliability of the proposed PD-CapsNet model, as shown in Figure 4 The ROC curve of the stratified five-fold cross-validation process of the experiment was shown. The classification prediction model based on the scale and tremor parameters can be used as an effective method for diagnosing Parkinson's disease and help assist clinicians in making objective diagnoses.

[0100] The above is a detailed description of an embodiment of a Parkinson's disease classification and evaluation model building method provided by the present application. The following is a detailed description of an embodiment of a Parkinson's disease classification and evaluation model building device provided by the present application.

[0101] See also Figure 5 , a modeling device for Parkinson's disease classification and evaluation model provided in this embodiment includes:

[0102] The feature acquisition unit 201 is used to acquire PD disease sample data and perform feature extraction from the PD disease sample data to obtain PD disease classification features and PD disease classification identifiers of different sample individuals;

[0103] A feature preprocessing unit 202 is used to perform data preprocessing on the PD disease classification feature to obtain a preprocessed PD disease classification feature;

[0104] The classification evaluation model training unit 203 is used to train a preset initial deep learning model using the PD disease classification features and the PD disease classification identifier as training data. When the training effect of the initial deep learning model meets the convergence condition, a Parkinson's disease classification evaluation model is obtained, wherein the initial deep learning model includes: a convolutional layer, a self-attention module and a capsule layer, wherein the self-attention module is arranged between the convolutional layer and the main capsule layer of the capsule layer.

[0105] Furthermore, the hierarchical structure of the initial deep learning model is: input layer, convolutional layer, self-attention module, primary capsule layer, digital capsule layer, linear layer and output layer;

[0106] The specific characteristics of PD disease classification include: demographic characteristics, scale characteristics, and tremor characteristics;

[0107] Demographic characteristics included: sex, age, and disease duration of different sample individuals;

[0108] Tremor characteristics included: bradykinesia score, dyskinesia score, motor fluctuations, proportion of tremor time, and proportion of resting / sleepy time;

[0109] Data preprocessing specifically includes: data cleaning processing, data encoding processing, data normalization processing and data balancing processing.

[0110] In addition, if Figure 6 As shown, the present application also provides a modeling terminal for Parkinson's disease classification assessment model, the terminal types include but are not limited to: personal computers, servers and embedded intelligent devices, the main components of the terminal include: a memory 33 and a processor 31, the memory 33 and the processor 31 can be connected via a communication bus 34;

[0111] The memory is used to store program codes, and the program codes correspond to a modeling method for Parkinson's disease classification and assessment model as described in the above embodiment;

[0112] The processor is used to read and execute program code.

[0113] The present application also provides a computer-readable storage medium, which corresponds to the memory 33 in the previous embodiment. The computer-readable storage medium stores program code. When the program code is read and executed by a processor, a Parkinson's disease classification and evaluation model modeling method is implemented as provided in the above embodiment.

[0114] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the terminals, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0115] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0116] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0117] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0118] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0119] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0120] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0121] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A Parkinson's disease classification and assessment model modeling method, characterized in that: include: Acquire PD disease sample data, and perform feature extraction from the PD disease sample data to obtain PD disease classification features and PD disease classification identifiers of different sample individuals; Performing data preprocessing on the PD disease classification features to obtain preprocessed PD disease classification features; The PD disease classification features and the PD disease classification identifier are used as training data to train a preset initial deep learning model. When the training effect of the initial deep learning model meets the convergence condition, a Parkinson's disease classification evaluation model is obtained, wherein the initial deep learning model includes: a convolutional layer, a self-attention module and a capsule layer, wherein the self-attention module is arranged between the convolutional layer and the main capsule layer of the capsule layer.

2. A Parkinson's disease classification and evaluation model modeling method according to claim 1, characterized in that: The hierarchical structure of the initial deep learning model is: input layer, convolution layer, self-attention module, main capsule layer, digital capsule layer, linear layer and output layer.

3. A Parkinson's disease classification and evaluation model building method according to claim 1, characterized in that: The PD disease classification characteristics specifically include: demographic characteristics, scale characteristics and tremor characteristics; The demographic characteristics include: gender, age, and disease duration of different sample individuals; The tremor characteristics include: bradykinesia score, dyskinesia score, motor fluctuations, tremor time proportion, and rest / sleep time proportion.

4. A Parkinson's disease classification and evaluation model building method according to claim 1, characterized in that: The data preprocessing specifically includes: data cleaning processing, data encoding processing, data normalization processing and data balancing processing.

5. A Parkinson's disease classification and evaluation model building method according to claim 4, characterized in that: The data balancing process specifically includes: Traverse the PD disease classification features to determine minority class samples; According to the minority class samples, a number of new sample data are generated through SMOTE oversampling processing, until the sample ratio meets the preset sample number balance ratio threshold.

6. A Parkinson's disease classification and evaluation model building method according to claim 4, characterized in that: The data normalization processing includes: maximum and minimum normalization processing or zero mean normalization processing.

7. A modeling device for Parkinson's disease classification and evaluation model, characterized in that: include: A feature acquisition unit, used to acquire PD disease sample data, and perform feature extraction from the PD disease sample data to obtain PD disease classification features and PD disease classification identifiers of different sample individuals; A feature preprocessing unit, used for performing data preprocessing on the PD disease classification feature to obtain a preprocessed PD disease classification feature; A classification evaluation model training unit is used to train a preset initial deep learning model using the PD disease classification features and the PD disease classification identifier as training data. When the training effect of the initial deep learning model meets the convergence condition, a Parkinson's disease classification evaluation model is obtained, wherein the initial deep learning model includes: a convolutional layer, a self-attention module and a capsule layer, wherein the self-attention module is arranged between the convolutional layer and the main capsule layer of the capsule layer.

8. A Parkinson's disease classification and evaluation model building device according to claim 7, characterized in that: The hierarchical structure of the initial deep learning model is: input layer, convolution layer, self-attention module, main capsule layer, digital capsule layer, linear layer and output layer; The PD disease classification characteristics specifically include: demographic characteristics, scale characteristics and tremor characteristics; The demographic characteristics include: gender, age, and disease duration of different sample individuals; The tremor characteristics include: bradykinesia score, dyskinesia score, motor fluctuations, tremor time proportion, and rest / sleep time proportion; The data preprocessing specifically includes: data cleaning processing, data encoding processing, data normalization processing and data balancing processing.

9. A modeling device for Parkinson's disease classification and evaluation model, characterized in that: include: Memory and processor; The memory is used to store program codes, and the program codes correspond to a Parkinson's disease classification and evaluation model modeling method according to any one of claims 1 to 6; The processor is used for reading and executing the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, and when the program code is read and executed by a processor, a Parkinson's disease classification and assessment model building method as described in any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Method for quantitatively evaluating symptoms of tremor of patient with Parkinson's disease according to approximate entropy and cross approximate entropy

    CN104398263A

  • Log anomaly detection method and device

    CN116610525A