Construction method, system and equipment of tremor classification model and medium
By constructing a tremor classification model, using multimodal data and improved convolutional neural networks, the problem of difficulty in diagnosis of PD and ET is solved, and high-precision disease distinction and the formulation of individualized treatment strategies are achieved.
Patent Information
- Application Number
- CN202510296818.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-13
- Publication Date
- 2025-07-04
AI Technical Summary
The prior art is difficult to accurately distinguish Parkinson's disease (PD) tremor from idiopathic tremor (ET), which leads to difficulty in diagnosis and cannot quickly formulate effective treatment strategies.
A tremor classification model is constructed, by obtaining the clinical data of the target group, motor function score data, emotional score data, cognitive function score data and brain image data, using machine learning algorithms, especially improved convolutional neural networks, combined with attention mechanisms, extracting and fusing multimodal data, automatically focusing on key brain regions and clinical features, and distinguishing between PD and ET.
It significantly improves the distinction between PD and ET, can quickly give accurate diagnostic results in complex situations, helping medical staff to formulate individualized treatment strategies and improve diagnostic efficiency.
Smart Images
Figure CN120259734A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical data processing, and particularly relates to a method, system, device and medium for constructing a tremor classification model. Background Art
[0002] Tremor is a rhythmic oscillatory movement generated by the alternating or synchronous contraction of agonist and antagonist muscles. The main diseases causing tremor are Parkinson's disease (PD) and essential tremor (ET). PD is a neurodegenerative disease of the nervous system caused by the progressive loss of dopaminergic neurons in the substantia nigra-striatum. The prevalence of PD in the total population is 0.3%, and the prevalence in patients over 60 years old is 1%. PD tremor is mainly manifested as resting tremor. ET is mainly manifested as intention and action tremors, and its specific pathogenesis remains unclear to date. Epidemiological data show that the prevalence of ET in adults worldwide is about 1%, slightly higher than that of PD. The incidence of ET increases with age, accounting for about 4-5% of the population over 65 years old.
[0003] In clinical practice, the clinical manifestations of ET are similar to those of PD tremor, and sometimes it is difficult to distinguish them only based on clinical manifestations. The main characteristics of PD tremor are asymmetric resting tremor, accompanied by some non-motor symptoms such as constipation and hyposmia. In clinical practice, some tremor patients have atypical symptoms or overlapping of these symptom types, which further increases the difficulty of diagnosis.
[0004] Therefore, there is an urgent need for a method, system, device and medium for constructing a tremor classification model to assist in distinguishing PD tremor and ET tremor through the tremor classification model. Summary of the Invention
[0005] The present invention provides a method, system, device and medium for constructing a tremor classification model to solve the defect that it is difficult to distinguish PD tremor and ET tremor only based on clinical manifestations, resulting in the inability to quickly and accurately formulate treatment decisions for patients.
[0006] A method for constructing a tremor classification model provided by the present invention includes:
[0007] Obtaining clinical data, motor function score data, emotional score data, cognitive function score data, and brain imaging data of a target population, where the target population includes a tremor-type PD patient group and an ET patient group;
[0008] Obtaining brain feature data of the target population according to the brain imaging data of the target population;
[0009] According to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the target population, using machine learning algorithms, the model is made to learn the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data corresponding to the tremor-type PD patient population and the ET patient population, so as to construct a tremor classification model.
[0010] According to a method for constructing a tremor classification model provided by the present invention, the clinical data includes any one of the following or any combination thereof: disease course, surgical age, onset age, levodopa equivalent dose (LEDD), and levodopa drug improvement rate.
[0011] According to a method for constructing a tremor classification model provided by the present invention, obtaining the brain characteristic data of the target population based on the brain imaging data of the target population includes:
[0012] Preprocessing the brain imaging data of the target population;
[0013] Performing data segmentation processing on the preprocessed brain imaging data;
[0014] Performing feature extraction processing on the brain imaging data after data segmentation processing to obtain the brain characteristic data of the target population.
[0015] According to a method for constructing a tremor classification model provided by the present invention, the preprocessing includes any one of the following or any combination thereof: quality inspection, format unification, intensity normalization, motion artifact correction, and image alignment.
[0016] According to a method for constructing a tremor classification model provided by the present invention, the data segmentation processing includes any one of the following or any combination thereof: skull stripping, white matter segmentation, cortical reconstruction, smoothing, mapping, cortical thickness measurement, and subcortical nucleus volume calculation.
[0017] According to a method for constructing a tremor classification model provided by the present invention, performing the feature extraction processing on the brain imaging data after data segmentation processing to obtain the brain characteristic data of the target population includes:
[0018] Performing one or more of cortical thickness analysis, nucleus volume analysis, and structural covariance analysis on the brain imaging data after data segmentation processing of the target population to obtain regions of interest in the target population's brain related to differentiating PD tremor and ET tremor;
[0019] Obtaining the brain characteristic data of the target population based on the regions of interest in the target population's brain related to differentiating PD tremor and ET tremor.
[0020] According to a method for constructing a tremor classification model provided by the present invention, the brain characteristic data of the target population includes cortical thickness data and / or nucleus volume data.
[0021] According to a method for constructing a tremor classification model provided by the present invention, based on the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the target population, using a machine learning algorithm, enabling the model to learn the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data corresponding to the tremor-type PD patient group and the ET patient group, so as to construct a tremor classification model, including:
[0022] According to the clinical data, motor function score data, emotional score data, cognitive function score data, cortical thickness data, and nucleus volume data of the target population, using an improved convolutional neural network based on the attention mechanism, enabling the model to learn the classification results corresponding to the tremor-type PD patient group and the ET patient group, so as to construct a tremor classification model, wherein the improved convolutional neural network based on the attention mechanism includes an input layer, a convolutional layer, an attention mechanism, a fully connected layer, and an output layer.
[0023] The present invention also provides a tremor classification model construction system, including:
[0024] A data acquisition module, configured to: acquire the clinical data, motor function score data, emotional score data, cognitive function score data, and brain imaging data of the target population, wherein the target population includes the tremor-type PD patient group and the ET patient group;
[0025] A data processing module, configured to: obtain the brain characteristic data of the target population according to the brain imaging data of the target population;
[0026] A model training module, configured to: according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the target population, using a machine learning algorithm, enabling the model to learn the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data corresponding to the tremor-type PD patient group and the ET patient group, so as to construct a tremor classification model.
[0027] The present invention also provides a tremor-assisted classification system, including:
[0028] A data receiving module, configured to: receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the test subject from at least one terminal;
[0029] A classification module, configured to: obtain tremor type data of a person to be tested based on the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested, and a tremor classification model obtained by the construction method of the tremor classification model described in any one of the above;
[0030] A data output module, configured to: send the tremor type data of the person to be tested to at least one terminal.
[0031] It should be noted that a terminal refers to an input / output device connected to a computer system. According to different functions, terminals can be divided into various types: smart terminals or intelligent terminals, dumb terminals, interactive terminals or online terminals. Specifically, a terminal can be various mobile communication devices, such as mobile phones, tablets, etc. The purpose here is to provide users with the functions of inputting data and obtaining data output.
[0032] The present invention also provides an electronic device, including a processor and a memory storing a computer program, where when the processor executes the computer program, the construction method of the tremor classification model described in any one of the above is implemented.
[0033] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the construction method of the tremor classification model described in any one of the above is implemented.
[0034] The present invention also provides a computer program product, where the computer program product includes a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer can execute the construction method of the tremor classification model described in any one of the above.
[0035] A construction method, system, device, and medium of a tremor classification model provided by the present invention can at least bring the following beneficial effects:
[0036] Deep extraction of brain features: For the brain imaging data of patients with tremor-type PD and ET, a set of high-precision feature extraction processes is set up, including MRI preprocessing, skull stripping, white matter / gray matter segmentation, quantitative measurement of cortical thickness and subcortical nuclear volume, combined with MCS correction, FDR control, and structural covariance analysis, to deeply mine and screen out key brain regions related to differentiating PD and ET. This process is not limited to simple thickness or volume metrics, but also uses multimodal statistical analysis and significance testing to highlight brain structure features with high discrimination for tremor type differences.
[0037] Fusion of Multimodal Data and Training of Machine Learning Models: In addition to brain images, the present invention incorporates multimodal information such as clinical data (such as disease course, surgical age, LEDD, levodopa drug improvement rate, etc.), motor function scores (MDS-UPDRS), mood scores (HAMA, HAMD), and cognitive function scores (MMSE, MoCA) into the training. By means of an improved convolutional neural network and the fusion of attention mechanisms, it can automatically focus on the brain or clinical features that are most valuable for differentiating tremor types, suppress irrelevant or noisy information, and significantly improve the accuracy and robustness of classification.
[0038] Through the above innovative in-depth exploration of brain structure features and the fusion of multimodal clinical data, the tremor classification model constructed by the present invention can significantly improve the discrimination between PD and ET, with a higher accuracy rate; it still has good adaptability in the face of complex situations such as differences in disease course stages and atypical symptoms; and it can quickly give classification results to help medical staff make more timely diagnostic decisions and develop individualized treatment strategies. This method also provides a feasible idea for other neurological diseases with similar diagnostic difficulties and has broad application prospects in the fields of medical image analysis and artificial intelligence diagnosis. Brief Description of the Drawings
[0039] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0040] Figure 1 It is a schematic flowchart of a method for constructing a tremor classification model provided by the present invention.
[0041] Figure 2 It is a schematic structural diagram of a tremor classification model.
[0042] Figure 3 It is a schematic structural diagram of a tremor-assisted classification system provided by the present invention.
[0043] Figure 4 It is a schematic structural diagram of an electronic device provided by the present invention. Detailed Embodiments
[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them, and they should not be construed as limitations on the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be construed as indicating or implying relative importance.
[0045] Figure 1 It is a schematic flowchart of a method for constructing a tremor classification model provided by the present invention. The execution subject of a method for constructing a tremor classification model provided by the present invention can be any applicable terminal-side device or network-side device, such as a device for constructing a tremor classification model, etc.
[0046] See Figure 1 , a method for constructing a tremor classification model provided by the present invention may include:
[0047] S110. Obtain the clinical data, motor function score data, emotional score data, cognitive function score data, and brain imaging data of the target population, where the target population includes a tremor-type PD patient population and an ET patient population.
[0048] In one embodiment, the clinical data includes any one or any combination of the following: disease course, surgical age, onset age, levodopa equivalent dose (LEDD), and levodopa drug improvement rate. The motor function score data may be the data obtained after the target population is scored by the UPDRS (Unified Parkinson's Disease Rating Scale), the emotional score data may be the data obtained after the target population is scored by the HAMA (Hamilton Anxiety Scale) and the HAMD (Hamilton Depression Scale), the cognitive function score data may be the data obtained after the target population is scored by the MMSE (Mini-Mental State Examination) and the MoCA (Montreal Cognitive Assessment), and the brain imaging data may be the brain MR image data of the target population.
[0049] S120. Obtain the brain feature data of the target population according to the brain imaging data of the target population.
[0050] In one embodiment, S120 may include:
[0051] Preprocess the brain imaging data of the target population, where the preprocessing includes any one or any combination of the following: quality inspection, format unification, intensity normalization, motion artifact correction, and image alignment;
[0052] Perform data segmentation on the preprocessed brain image data;
[0053] The brain image data after data segmentation is subjected to feature extraction processing to obtain brain feature data of the target group.
[0054] In one embodiment, the data segmentation process includes any one of the following or any combination thereof: skull stripping, white matter segmentation, cortical reconstruction, smoothing, mapping, cortical thickness measurement, and subcortical nucleus volume calculation, wherein,
[0055] Skull stripping: FreeSurfer uses a built-in algorithm to remove the skull cap and generate a brain tissue mask to remove unnecessary extracranial tissue. If the stripping effect is not ideal, you can manually modify the parameters or perform a quality check.
[0056] White matter segmentation: FreeSurfer automatically segments white matter regions based on image intensity and topological information. The generated white matter segmentation maps usually need to be quality checked to ensure that the boundaries between white matter and gray matter and cerebrospinal fluid (CSF) are accurate. FreeSurfer will generate a "white matter-cortex" interface based on these white matter segmentation maps.
[0057] Cortical reconstruction: FreeSurfer generates the "white matter-cortex" interface and the "cortex-cerebrospinal fluid" interface based on the white matter segmentation map to form the structure of the cortex. This process includes topology correction and mesh optimization, making the cortical reconstruction more accurate and consistent with the anatomical characteristics of the brain.
[0058] Smoothing: FreeSurfer was used to smooth the white matter and cortical surfaces through an iterative smoothing algorithm, which can remove some irregular details while maintaining the main characteristic morphology of the cortex, thereby improving the accuracy of subsequent cortical thickness measurement and functional mapping.
[0059] Mapping: FreeSurfer uses spherical coordinates to map each subject's cortical structure onto a standard spatial template. This method facilitates comparison of cortical morphology between individuals, such as analysis of cortical thickness, surface area, and other features.
[0060] Cortical thickness measurement: FreeSurfer was used to estimate cortical thickness by calculating the distance between the white matter surface and the cortical surface. The thickness of each cortical vertex was measured and recorded in this way, generating a cortical thickness map. This thickness map can be used for group comparison analysis and to explore relationships with pathological or cognitive variables.
[0061] Calculation of Subcortical Structures: The subcortical structures are segmented using probabilistic atlases and shape models by FreeSurfer. The generated segmentation results can be used to calculate the volumes of individual subcortical nuclei. The volume information of these structures is output by FreeSurfer, and tools for quality inspection are provided to ensure the segmentation accuracy.
[0062] In one embodiment, the processed brain image data can be segmented according to the data of the target population, and one or more of cortical thickness analysis, nucleus volume analysis, and structural covariance analysis can be performed to obtain the regions of interest in the brain of the target population related to differentiating PD tremors and ET tremors, and then the brain feature data of the target population can be obtained. The brain feature data of the target population can include cortical thickness data and / or nucleus volume data.
[0063] Among them, the specific methods of cortical thickness analysis, nucleus volume analysis, and structural covariance analysis are as follows.
[0064] Cortical Thickness Analysis: The t1-weighted images of the target population are preprocessed using MRIcroGL software, and all imaging data are processed using FreeSurfer software. The FreeSurfer analysis process includes skull stripping, white matter segmentation, cortical reconstruction, smoothing, mapping, cortical thickness measurement, and calculation of subcortical nuclei. Monte Carlo Simulation (MCS) is used for cluster correction, and the threshold is set to P < 0.05. The significantly different brain regions obtained in this part are defined as regions of interest.
[0065] Nucleus Volume Analysis: FreeSurfer is used to extract the volumes of subcortical structures. In this analysis, gender and age are included as covariates to control their potential confounding effects. The significance level (P value) is set to 0.05, and the false discovery rate (FDR) is appropriately adjusted. The differences obtained in the nuclei are defined as regions of interest.
[0066] Structural Covariance Analysis: This analysis is performed in the SurfStat package of Matlab software. The structural covariance analysis uses Gaussian Random Field (GRF) correction, and the threshold is set to P < 0.05. In the regression analysis, gender and age are included as covariates to control their potential confounding effects. The brain regions with significant differences mentioned above are used for structural covariance analysis, and the significant results of the structural covariance analysis are also defined as regions of interest.
[0067] S130. According to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the target group, using a machine learning algorithm, the model is made to learn the corresponding clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the tremor-type PD patient group and the ET patient group, so as to construct a tremor classification model.
[0068] In one embodiment, S130 can, according to the clinical data, motor function score data, emotional score data, cognitive function score data, cortical thickness data, and nuclear mass volume data of the target group, use an improved convolutional neural network based on the attention mechanism to make the model learn the corresponding classification results of the tremor-type PD patient group and the ET patient group, so as to construct a tremor classification model. Among them, the improved convolutional neural network based on the attention mechanism includes an input layer, a convolutional layer, an attention mechanism, a fully connected layer, and an output layer.
[0069] See Figure 2 , regarding the improved convolutional neural network based on the attention mechanism.
[0070] 1) Convolutional layer (Conv layer 1, Conv layer 2, Conv layer 3, etc.): Figure 2 Shows the first convolutional layer (Conv layer 1) passed by the model and subsequent multiple convolutional structures (Conv layer 2, 3, etc.). These convolutional layers extract high-dimensional expressions of image features through convolutional kernels of different sizes (Kernel) and gradually compress the feature dimensions in combination with operations such as the pooling layer (Pooling) to retain the main information. The detailed information is as follows:
[0071] Conv layer 1:
[0072] Convolution operation: Use 1D convolution (kernel size = 3, padding = 1) to map the single-channel input to 32 channels.
[0073] Subsequent processing: After ReLU activation, the feature dimensions are compressed to half of the original through max pooling (kernel size = 2, stride = 2).
[0074] Conv layer 2:
[0075] Convolution operation: Convolve the output of Conv layer 1, map 32 channels to 64 channels, and also use kernel size = 3 and padding = 1.
[0076] Subsequent processing: Similarly, through ReLU activation and max pooling operations, the feature dimension is further compressed.
[0077] Conv layer 3:
[0078] Convolution operation: Map 64 channels to 128 channels through convolution, with the same parameter settings as described above.
[0079] Subsequent processing: After activation and pooling, the output sequence length is reduced to 1 / 8 of the original input.
[0080] 2) Attention mechanism (SE-Net): In Figure 2 "SE-Net" module can be regarded as a type of attention mechanism (Squeeze-and-Excitation Network), which assigns weights in the channel or feature dimension, giving greater weights to channels with higher information content, thereby "emphasizing" the key features for distinguishing Parkinson's tremor and essential tremor. Improvements represented by SE-Net or other self-attention modules can automatically highlight the most critical brain region features or clinical indicators for diagnosis, suppressing irrelevant or noisy information. This is particularly important for diseases such as Parkinson's disease tremor and essential tremor where there may be partial overlap in symptoms, as it can capture the core differences in atypical manifestations and improve accuracy. Embedding an SE module after each convolutional module, its main steps include:
[0081] Squeeze: Using global average pooling, compress the spatial information of each channel into a single statistic. Excitation: Calculate the weight of each channel through a two-layer fully connected network (with ReLU activation in the middle and Sigmoid activation at the end), and the weight value reflects the information content contained in this channel.
[0082] Rescaling: Multiply the generated channel attention weights by the original features channel by channel, thereby automatically "emphasizing" the features crucial for distinguishing PD from ET and suppressing irrelevant or noisy information.
[0083] 3) Fully connected layers (FC layers): The feature tensors output by convolution and attention are "flattened" into high-dimensional vectors, and then input into several layers of fully connected networks for further non-linear mapping and feature fusion. If both clinical data and imaging data are used, multi-modal fusion will be performed at this stage or at the attention module to integrate clinical indicators and imaging features and obtain a more comprehensive basis for tremor classification and discrimination. The feature tensors output by convolution and the SE module are transformed into high-dimensional vectors through a flatten operation, with a length of 128×(input_size / 8). Next, this vector passes through multiple layers of fully connected networks in sequence for non-linear mapping and feature fusion:
[0084] FC1: Map the flattened vector to 128 dimensions, activate it through ReLU, and use Dropout (dropout rate 0.6) to prevent overfitting.
[0085] FC2: Map the 128 dimensions to 64 dimensions, also through activation and Dropout.
[0086] FC3 (mapping before the output layer): Map the 64 dimensions to the final output dimension.
[0087] 4) Output layer (Output): As Figure 2 shown on the far right, the final output of the model can be essential tremor and Parkinson's disease tremor. The 64-dimensional features output by the fully connected layer are mapped through the last layer to obtain 2-dimensional logits. In this embodiment, the confidence of each type of tremor is calculated through activation functions such as Softmax or Sigmoid, realizing the accurate determination of whether the subject has PD tremor or ET.
[0088] The following will provide specific examples of constructing a tremor classification model.
[0089] 1. Data acquisition
[0090] In this embodiment, the brain MRI data of 140 subjects (including 71 ET patients and 69 PD patients) from Beijing Tiantan Hospital Affiliated to Capital Medical University from September 2018 to June 2024 are selected, including 100 patients diagnosed with Parkinson's disease (PD) of the tremor type and 100 patients diagnosed with essential tremor (ET). At the same time, to fully verify the multi-modal data fusion ability of this embodiment, the following non-imaging data of all subjects are also collected: including disease duration (years), surgical age (years), onset age (years), levodopa equivalent dose (LEDD), levodopa drug improvement rate; motor function scoring data: UPDRS (motor symptom score), action test score, etc.; emotion scoring data: such as anxiety and depression scale scores; cognitive function scoring data: such as MoCA and MMSE scale scores, etc.
[0091] 2. Data preprocessing and feature extraction
[0092] Imaging data preprocessing: Perform quality inspection, format unification, intensity normalization, motion artifact correction, and image alignment on the T1-weighted MRI images of the subjects. Subsequently, use FreeSurfer or similar software packages for data segmentation to obtain quantitative brain structure indicators such as cortical thickness and nucleus volume. Perform Monte Carlo simulation cluster correction (p<0.05) or FDR correction on the cortical thickness and nucleus volume, and select the regions of significant difference (ROI) as the subsequent input features.
[0093] Non-imaging data preprocessing: Normalize (or standardize) various scores and clinical indicators. If there are missing values, use the mean imputation method or multiple imputation method to fill them in.
[0094] 3. Model training and tuning
[0095] 1) Model structure and implementation
[0096] According to this embodiment, the model is an "improved convolutional neural network based on the attention mechanism", and its core structure includes ( Figure 2 ):
[0097] Convolutional layers (Conv Layers): Deeply extract convolutional features from brain imaging features (such as cortical thickness matrix, nucleus volume vector) and clinical data;
[0098] Attention mechanism (SE-Net or self-attention module): Allocate weights in the feature channel or spatial dimension to highlight the most critical brain regions or features for distinguishing PD tremors and ET;
[0099] Fully connected layers (FC Layers): Fuse the high-dimensional features output by the convolution and attention modules with the vectors corresponding to non-imaging data, and map them multi-layer to the final classification output;
[0100] Classification layer (Classification Layer): Use the Softmax function to output the predicted probability distributions of the two categories of PD tremors and ET.
[0101] 2) Division of training set and validation / test set
[0102] Divide all 200 subjects into a training set, a validation set, and a test set according to the ratio of 8:1:1. Inside the training set, 10-fold cross-validation can also be used to further evaluate the robustness of the model and select appropriate hyperparameters. After inputting the training data, the model will first obtain multi-level representations of imaging features through the convolutional layer, and then enter the attention module to allocate weights to important features; the vectors after feature fusion will continuously learn non-linear mappings in the fully connected layer and output the prediction results through the Softmax classification layer.
[0103] 3) Model performance evaluation
[0104] Monitor the changes in Loss and accuracy (Accuracy) of the model at each Epoch stage through the validation set. If the performance of the validation set does not improve significantly within several consecutive Epochs, training can be stopped early (Early Stopping); finally evaluate the optimal model on the test set. After evaluation, it can be obtained that the classification performance of the tremor classification model constructed in this embodiment is better.
[0105] The tremor classification model constructed in this embodiment can be used for tremor assisted classification. For example, the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested are input into the tremor classification model, and the tremor type data of the person to be tested is obtained through the tremor classification model to assist medical staff in diagnosing patients more accurately and efficiently.
[0106] A method, system, device, and medium for constructing a tremor classification model provided by the present invention use a machine learning model to learn the clinical data, motor function score data, emotional score data, cognitive function score data, and brain imaging data corresponding to the tremor type PD patient group and the ET patient group, so as to train a high-precision tremor classification model, which can be used to quickly and accurately distinguish the tremor type of the person to be tested, help medical staff formulate reasonable and individualized treatment strategies for patients, and greatly reduce the mortality rate.
[0107] The following describes a system for constructing a tremor classification model provided by the present invention. The system for constructing a tremor classification model described below can be correspondingly referred to the method for constructing a tremor classification model described above.
[0108] A system for constructing a tremor classification model provided by the present invention may include:
[0109] A data acquisition module, configured to: acquire the clinical data, motor function score data, emotional score data, cognitive function score data, and brain imaging data of the target group, where the target group includes the tremor type PD patient group and the ET patient group;
[0110] A data processing module, configured to: obtain the brain feature data of the target group according to the brain imaging data of the target group;
[0111] A model training module, configured to: according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the target group, use a machine learning algorithm to make the model learn the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data corresponding to the tremor type PD patient group and the ET patient group, so as to construct a tremor classification model.
[0112] See Figure 3 The present invention also provides a tremor assisted classification system, including:
[0113] A data receiving module, configured to: receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested from at least one terminal;
[0114] A classification module, configured to: obtain tremor type data of a person to be tested according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested, and the tremor classification model obtained by the tremor classification model construction method described in any one of the above;
[0115] A data output module, configured to: send the tremor type data of the person to be tested to at least one terminal.
[0116] In specific applications, medical staff can input the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested into the tremor-assisted classification system through a terminal. The data receiving module receives the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested. Then, the classification module uses the tremor classification model to obtain the tremor type data of the person to be tested. Then, the data output module sends the tremor type data of the person to be tested to the terminal for display, which can assist medical staff in accurately judging the tremor type of the person to be tested and improving the diagnosis and treatment efficiency.
[0117] Figure 4 An example of a schematic physical structure diagram of an electronic device is shown in Figure 4 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the tremor classification model construction method described in any one of the above and / or the following steps:
[0118] Receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested from at least one terminal;
[0119] According to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the person to be tested, obtain the tremor type data of the person to be tested through the tremor classification model obtained by the tremor classification model construction method described in any one of the above;
[0120] Send the tremor type data of the person to be tested to at least one terminal.
[0121] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a 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 and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.
[0122] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the construction method of any one of the above-mentioned tremor classification models and / or the following steps:
[0123] Receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the subject to be tested from at least one terminal;
[0124] Obtain the tremor type data of the subject to be tested according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the subject to be tested and the tremor classification model obtained by the construction method of any one of the above-mentioned tremor classification models;
[0125] Send the tremor type data of the subject to be tested to at least one terminal.
[0126] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the construction method of any one of the above-mentioned tremor classification models and / or the following steps:
[0127] Receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the subject to be tested from at least one terminal;
[0128] Obtain the tremor type data of the subject to be tested according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain characteristic data of the subject to be tested and the tremor classification model obtained by the construction method of any one of the above-mentioned tremor classification models;
[0129] Send the tremor type data of the person to be tested to at least one terminal.
[0130] The device embodiments described above are merely illustrative. 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 to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.
[0131] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for constructing a tremor classification model, characterized in that Including: Obtaining clinical data, motor function score data, emotional score data, cognitive function score data, and brain imaging data of the target population, where the target population includes a tremor-type PD patient population and an ET patient population; Obtaining brain feature data of the target population based on the brain imaging data of the target population; According to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the target population, using a machine learning algorithm, enabling the model to learn the corresponding clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the tremor-type PD patient population and the ET patient population, so as to construct a tremor classification model.
2. The method for constructing a tremor classification model according to claim 1, wherein, The clinical data includes any one or any combination of the following: disease course, surgical age, onset age, levodopa equivalent dose, and levodopa drug improvement rate.
3. The method for constructing a tremor classification model according to claim 1, wherein The obtaining brain feature data of the target population based on the brain imaging data of the target population includes: Performing preprocessing on the brain imaging data of the target population; Performing data segmentation processing on the preprocessed brain imaging data; Performing feature extraction processing on the brain imaging data after data segmentation processing to obtain brain feature data of the target population.
4. The method for constructing a tremor classification model according to claim 3, characterized in that The preprocessing includes any one or any combination of the following: quality inspection, format unification, intensity normalization, motion artifact correction, image alignment, and / or The data segmentation processing includes any one or any combination of the following: skull stripping, white matter segmentation, cortical reconstruction, smoothing, mapping, cortical thickness measurement, and subcortical nucleus volume calculation.
5. The method for constructing a tremor classification model according to claim 3, wherein The performing feature extraction processing on the brain imaging data after data segmentation processing to obtain brain feature data of the target population includes: Performing one or more of cortical thickness analysis, nucleus volume analysis, and structural covariance analysis on the brain imaging data after data segmentation processing of the target population to obtain regions of interest in the target population's brain related to differentiating PD tremor and ET tremor; Obtaining brain feature data of the target population based on the regions of interest in the target population's brain related to differentiating PD tremor and ET tremor.
6. The method for constructing a tremor classification model according to claim 1, wherein The brain feature data of the target population includes cortical thickness data and / or nucleus volume data.
7. A method for constructing a tremor classification model according to any one of claims 1-6, characterized in that, The according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the target population, using a machine learning algorithm, enabling the model to learn the corresponding clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the tremor-type PD patient population and the ET patient population, so as to construct a tremor classification model includes: According to the clinical data, motor function score data, emotional score data, cognitive function score data, cortical thickness data, and nucleus volume data of the target population, using an improved convolutional neural network based on the attention mechanism, enabling the model to learn the corresponding classification results of the tremor-type PD patient population and the ET patient population, so as to construct a tremor classification model, where the improved convolutional neural network based on the attention mechanism includes an input layer, a convolutional layer, an attention mechanism, a fully connected layer, and an output layer.
8. A tremor assistance classification system, characterized in that, Including: A data receiving module, configured to: receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the subject to be tested from at least one terminal; A classification module, configured to: obtain the tremor type data of the subject to be tested according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the subject to be tested, and the tremor classification model obtained by the tremor classification model construction method described in any one of claims 1-7; A data output module, configured to: send the tremor type data of the subject to be tested to at least one terminal.
9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, the following steps are implemented: Receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the subject to be tested from at least one terminal; Obtain the tremor type data of the subject to be tested according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the subject to be tested, and the tremor classification model obtained by the tremor classification model construction method described in any one of claims 1-7; Send the tremor type data of the subject to be tested to at least one terminal.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the following steps are implemented: Receive the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the subject to be tested from at least one terminal; Obtain the tremor type data of the subject to be tested according to the clinical data, motor function score data, emotional score data, cognitive function score data, and brain feature data of the subject to be tested, and the tremor classification model obtained by the tremor classification model construction method described in any one of claims 1-7; Send the tremor type data of the subject to be tested to at least one terminal.