Method and system for monitoring brain function associated with parkinson's disease based on multi-modal data

Abstract

Disclosed in the present invention are a method and system for monitoring brain function associated with Parkinson's disease based on multi-modal data. The method comprises: processing original multi-modal Parkinson's disease monitoring data to obtain a standardized multi-modal data set; extracting data features of the standardized multi-modal data set to obtain a dimensionality-reduced feature set; aligning the dimensionality-reduced feature set on the basis of a time series to obtain a time-aligned multi-modal feature set; calculating correlations between features in the time-aligned multi-modal feature set to obtain an optimized feature subset; and inputting the optimized feature subset into a deep neural network model to obtain a brain function state monitoring result associated with Parkinson's disease.

Description

A method and system for monitoring brain function in Parkinson's disease based on multimodal data Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology, and in particular relates to a method and system for monitoring brain function in Parkinson's disease based on multimodal data. Background Technology

[0002] In monitoring brain function in Parkinson's disease, effectively integrating multimodal data is a key technical challenge. First, different modalities possess different characteristics and scales, making standardization and alignment during preprocessing a significant challenge. Second, the volume and quality of data vary considerably across modalities; balancing and utilizing these differences while avoiding data redundancy and noise interference requires carefully designed feature extraction and selection strategies. Third, complex nonlinear relationships exist among multimodal data, making it difficult for traditional machine learning models to fully uncover their intrinsic connections, necessitating the introduction of advanced modeling methods such as deep learning. Finally, designing reasonable evaluation metrics and validation schemes to comprehensively examine the model's performance and generalization ability during model validation is also a pressing issue. These intertwined technical difficulties require researchers to deeply analyze the characteristics of multimodal data and explore innovative integration solutions to ultimately achieve accurate monitoring of brain function in Parkinson's disease. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a method and system for monitoring brain function in Parkinson's disease based on multimodal data, thereby resolving the issues present in the prior art.

[0004] To achieve the above objectives, the present invention provides a method for monitoring brain function in Parkinson's disease based on multimodal data, comprising:

[0005] The original multimodal Parkinson's disease surveillance data were standardized to obtain a standardized multimodal dataset;

[0006] The data features of the standardized multimodal dataset are extracted to obtain the dimensionality-reduced feature set;

[0007] Aligning the dimensionality-reduced feature set based on the time series data yields a time-aligned multimodal feature set.

[0008] The correlation between features in the time-aligned multimodal feature set is calculated to obtain an optimized feature subset;

[0009] The optimized feature subset is input into a preset deep neural network model for identification to obtain the monitoring results of brain function status in Parkinson's disease.

[0010] Optionally, the multimodal dataset includes: electroencephalogram (EEG), magnetic resonance imaging (MRI), and clinical scale data.

[0011] Optionally, the process of standardizing the raw multimodal Parkinson's disease surveillance data includes:

[0012] A standardized EEG was obtained by removing high-frequency noise and baseline drift from the EEG using a combination of wavelet transform and bandpass filtering.

[0013] The artifacts in the magnetic resonance imaging are eliminated by using bias field correction and intensity non-uniformity correction algorithms to obtain standardized magnetic resonance imaging.

[0014] Data cleaning and outlier detection methods were used to remove outliers from the clinical scale data to obtain standardized clinical scale data.

[0015] A standardized multimodal dataset was obtained based on the standardized electroencephalogram, standardized magnetic resonance imaging, and standardized clinical scale data.

[0016] Optionally, principal component analysis can be used to process the standardized multimodal dataset to obtain a dimensionality-reduced feature set.

[0017] Optionally, the process of obtaining a time-aligned multimodal feature set includes:

[0018] The alignment results are obtained by aligning time series data of different modalities using dynamic time warping techniques.

[0019] Based on the alignment results, the dimensionality-reduced feature sets are fused to obtain a time-aligned multimodal fusion feature set.

[0020] Optionally, the process of obtaining the optimized feature subset includes:

[0021] For each feature in the time-aligned multimodal fusion feature set, calculate the mutual information value between the current feature and other features;

[0022] Based on the mutual information values, a correlation matrix between features is obtained;

[0023] The candidate feature subset is obtained by filtering the correlation matrix between the features;

[0024] The optimized feature subset is obtained by performing cluster analysis on the candidate feature subset based on the clustering algorithm.

[0025] Optionally, the process of inputting the optimized feature subset into a preset deep neural network model for identification to obtain the monitoring results of Parkinson's disease brain function status includes:

[0026] Preliminary prediction results are obtained by performing nonlinear relationship modeling on the optimized feature subset based on the multilayer perceptron structure in the deep neural network model.

[0027] The preliminary prediction results are compared with a preset standard threshold.

[0028] If the preliminary prediction result does not exceed the preset standard threshold, the patient's current brain function is judged to be normal; if the preliminary prediction result exceeds the preset standard threshold, the patient's current brain function is judged to be abnormal.

[0029] The support vector machine algorithm was used to determine the type and degree of brain function abnormalities in patients with abnormal brain function status, thus obtaining the monitoring results of brain function status in Parkinson's disease.

[0030] This invention also provides a Parkinson's disease brain function monitoring system based on multimodal data, used to implement a method for monitoring Parkinson's disease brain function based on multimodal data, the system comprising:

[0031] The data acquisition and standardization module is used to acquire raw multimodal Parkinson's disease monitoring data and standardize the raw multimodal Parkinson's disease monitoring data to obtain a standardized multimodal dataset.

[0032] The feature extraction and dimensionality reduction module is used to perform feature dimensionality reduction on the standardized multimodal dataset to obtain a dimensionality-reduced feature set;

[0033] The time alignment module is used to align the dimensionality-reduced feature set to obtain a time-aligned multimodal feature set;

[0034] The feature optimization module is used to calculate the correlation between features in the time-aligned multimodal feature set to obtain an optimized feature subset;

[0035] The monitoring module is used to identify the monitoring results of brain function status in Parkinson's disease based on the optimized feature subset.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] This invention eliminates dimensional and numerical range differences between different modalities by standardizing multimodal monitoring data, laying the foundation for subsequent analysis. The feature extraction and dimensionality reduction steps effectively eliminate redundant information, reduce data dimensionality, refine the data, and improve computational efficiency, while retaining key information and helping to highlight core features closely related to brain function in Parkinson's disease.

[0038] This invention, based on time-series feature alignment, accurately matches the correspondence between different modalities of data in the time dimension, ensuring temporal consistency between features. This is crucial for accurately capturing the dynamic changes in brain function over time in Parkinson's disease. Correlation calculation further optimizes the feature subset, eliminating redundant and interfering features, strengthening the complementarity between features, and making the features input into the deep neural network more representative and discriminative.

[0039] Deep neural networks, with their powerful nonlinear modeling capabilities, can accurately identify the complex patterns contained in optimized feature subsets, thereby accurately outputting the monitoring results of brain function status in Parkinson's disease. Attached Figure Description

[0040] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0041] Figure 1 is a flowchart of the Parkinson's disease brain function monitoring method based on multimodal data according to an embodiment of the present invention. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0044] Example 1

[0045] As shown in Figure 1, this embodiment provides a method for monitoring brain function in Parkinson's disease based on multimodal data, including the following steps:

[0046] The original multimodal Parkinson's disease monitoring data is standardized to obtain a standardized multimodal dataset; data features of the standardized multimodal dataset are extracted to obtain a dimensionality-reduced feature set; the dimensionality-reduced feature set is aligned based on time series to obtain a time-aligned multimodal feature set; the correlation between features in the time-aligned multimodal feature set is calculated to obtain an optimized feature subset; the optimized feature subset is input into a preset deep neural network model for recognition to obtain the monitoring results of Parkinson's disease brain function status.

[0047] Step 1: Obtain raw multimodal Parkinson's disease monitoring data, including electroencephalogram (EEG), magnetic resonance imaging (MRI), and clinical scale data. Based on the characteristics of each Parkinson's disease monitoring modality, different standardization methods are used to process the data to obtain a standardized multimodal dataset.

[0048] Furthermore, for the EEG data, a combination of wavelet transform and bandpass filtering was used to remove high-frequency noise and baseline drift, retaining effective EEG signal components. Wavelet transform decomposes the signal in the EEG into different frequency bands, effectively removing high-frequency noise. Bandpass filtering retains effective signals within specific frequency ranges, such as alpha waves (8-13Hz) and beta waves (13-30Hz). This combined method filters out interference while preserving key EEG activity information. For the MRI data, bias field correction and intensity non-uniformity correction algorithms were used to eliminate artifacts introduced by field inhomogeneity, improving image contrast and signal-to-noise ratio. For the clinical scale data, data cleaning and outlier detection methods were used to remove errors and outliers, ensuring data accuracy and consistency. Feature extraction was performed on the preprocessed EEG, MRI, and clinical scale data to obtain feature vector representations for each modality. For electroencephalography (EEG), features such as power spectral density and phase synchronization can be extracted; for magnetic resonance imaging (MRI), features such as volume, shape, and texture can be extracted; and for clinical scales, features such as total score and scores for each dimension can be extracted. The minimax normalization method is used to normalize the feature vectors of each modality, scaling the feature value range to the [0, 1] interval to eliminate the influence of differences in feature value ranges. The normalized EEG, MRI, and clinical scale feature vectors are merged to construct a standardized multimodal dataset, resulting in a more comprehensive patient representation and providing high-quality data support for subsequent intelligent diagnosis and disease prediction. For the diagnosis of Alzheimer's disease, EEG can reflect abnormal neuronal activity, MRI can show the degree of brain atrophy, and clinical scales can quantify the decline in cognitive function. These pieces of information complement each other, providing a multi-dimensional view of the disease.

[0049] Step 2: For the standardized multimodal dataset, principal component analysis is used to extract the main features of each mode to obtain a dimensionality-reduced feature set, thereby reducing data redundancy and noise interference.

[0050] Furthermore, a standardized multimodal dataset is obtained, and principal component analysis (PCA) is used to extract features for each modality. PCA identifies the principal features for each modality and sorts them by feature value, selecting the top N features as the principal features for that modality. The extracted principal features for each modality are combined into a dimensionality-reduced feature set, resulting in a compressed multimodal data representation. For example, assuming the original EEG data contains signals from 128 leads, PCA might reveal that the first 10 principal components can explain 90% of the data variance. These 10 principal components can then be selected as the principal features of the EEG data, significantly reducing the data dimensionality. For magnetic resonance imaging (MRI) data, PCA can help extract key anatomical structural features. For instance, in brain images, changes in the volume of regions such as the hippocampus and amygdala may be important features for distinguishing certain neurodegenerative diseases. By preserving these principal features, data compression can be achieved without significant information loss. While clinical scale data has lower dimensionality, redundancy may still exist. Principal component analysis may reveal high correlations between certain scale items, allowing a smaller number of principal components to represent most of the information in the original data.

[0051] For the dimensionality-reduced feature set, similarity calculation methods, such as cosine similarity, are used to calculate the similarity between data from different modalities. Based on the calculated similarity matrix, clustering algorithms such as K-means are used to divide the data from different modalities into several clusters. For the data within each cluster, multimodal fusion methods, such as multi-kernel learning or deep learning, are used to learn the correlation between data from different modalities, obtaining a unified cross-modal representation. The fused multimodal data representation is then used for subsequent tasks, such as classification and retrieval, reducing data redundancy and noise interference, and improving task performance.

[0052] Step 3: Input the dimensionality-reduced feature set into the preset alignment algorithm, and use dynamic time warping technology to align the time series data of different modalities to obtain a time-aligned multimodal feature set.

[0053] Furthermore, based on a pre-defined feature extraction algorithm, features are extracted from the multimodal data to obtain an initial feature set. Principal component analysis is then used to reduce the dimensionality of the initial feature set, resulting in a dimensionality-reduced feature subset. This dimensionality-reduced feature subset is input into a pre-defined alignment algorithm, and dynamic time warping is employed to align the time series data of different modalities. During the alignment process, a dynamic programming algorithm is used to find the optimal matching path between different modal data, minimizing the matching distance. Data segments with matching distances exceeding a pre-defined threshold are considered to have failed to align and are discarded. Based on the alignment results, the features of different modalities are fused to obtain a time-aligned multimodal fusion feature set. This time-aligned multimodal fusion feature set is then used as input for subsequent machine learning algorithms for pattern recognition and classification prediction tasks.

[0054] Step four: For the time-aligned multimodal feature set, calculate the correlation between features using the mutual information method, select features with a correlation higher than a preset threshold, and obtain the optimized feature subset.

[0055] Further, a time-aligned multimodal feature set is obtained. For each feature in the feature set, its mutual information value with other features is calculated to obtain a feature correlation matrix. Based on a preset threshold, the feature correlation matrix is ​​filtered, retaining feature pairs with correlations higher than the threshold to obtain a candidate feature subset. For the candidate feature subset, a clustering algorithm is used to group highly correlated features into the same class, and a representative feature is selected from each class to form an optimized feature subset. If the dimensionality of the feature subset is still high, principal component analysis is used to reduce the dimensionality of the feature subset, resulting in a dimensionality-reduced feature subset. The optimized feature subset is used as input, and a support vector machine algorithm is used to classify the multimodal data to obtain the classification result. Based on the classification result, it is determined whether the current feature subset can effectively distinguish different categories of data. If the classification accuracy is lower than a preset threshold, step 1 is returned, and a new feature subset is selected. The final optimized feature subset is determined and applied to the classification and recognition task of multimodal data to improve the system's performance and efficiency.

[0056] The optimized feature subset is used as input, and the multimodal data is classified using the support vector machine algorithm to obtain the classification result.

[0057] Acquire multimodal data, preprocess the data, and extract an initial feature set; use a feature selection algorithm to optimize and filter the initial features to obtain an optimized feature subset; divide the optimized feature subset into a training set and a test set; use the training set data to train the support vector machine model to obtain a trained classification model; input the test set data into the trained support vector machine model to perform classification prediction on the multimodal data; evaluate the classification performance of the support vector machine model based on the classification prediction results; if the classification performance does not reach a preset threshold, return to continue optimizing the feature subset until the classification performance meets the requirements, and finally output the classification result.

[0058] Step 5: Input the optimized feature subset into the preset deep neural network model, and use a multilayer perceptron structure to model the nonlinear relationship between multimodal data to obtain preliminary prediction results.

[0059] S1: Based on the requirements for monitoring brain function in Parkinson's disease, determine the type and source of multimodal data, and acquire raw multimodal Parkinson's disease brain function monitoring data using data acquisition techniques. S2: Preprocess the acquired raw multimodal Parkinson's disease brain function monitoring data, including data cleaning, normalization, and feature extraction, to obtain a standardized multimodal dataset. S3: Use feature selection algorithms to screen features in the multimodal dataset, and optimize to obtain the optimal feature subset through correlation analysis and importance assessment. S4: Construct a deep neural network model using a multilayer perceptron structure, designing the input layer, hidden layers, and output layer, and determining the number of neurons and activation functions in each layer. S5: Input the optimized feature subset into the constructed deep neural network model, and train the model parameters using forward and backward propagation algorithms to model the nonlinear relationships between multimodal data. S6: During training, use cross-validation to evaluate the model, and continuously optimize model performance by adjusting hyperparameters and network structure until the preset convergence condition is met. S7: Use a trained deep neural network model to predict new multimodal data. Input the data into the model, perform forward propagation calculations, and obtain preliminary prediction results, which serve as a reference for subsequent decision-making.

[0060] Step six: Apply the model that meets the performance requirements to the new multimodal dataset, and use the prediction results of the model to determine the brain function status of Parkinson's disease, and obtain the final monitoring results.

[0061] Furthermore, based on pre-established performance evaluation criteria, candidate models are evaluated to select the optimal model that meets the performance requirements. A dataset of Parkinson's disease patients containing multimodal data is acquired, and the data is preprocessed to extract key features. The preprocessed multimodal data is input into the optimal model, and the model predicts the patient's brain function status. The prediction results are compared with a preset threshold; if the prediction results exceed the threshold, the patient's brain function status is determined to be abnormal. For patients determined to have abnormal brain function, a support vector machine algorithm is used to further determine the degree and type of abnormality. A decision tree algorithm, combined with the patient's clinical manifestation data, is used to determine the specific stage of the patient's condition. By combining the model prediction results, the degree of abnormality determination, and the disease staging results, the final monitoring results of the patient's brain function are obtained to guide subsequent treatment.

[0062] As a specific implementation of this embodiment, establishing performance evaluation criteria is crucial in monitoring brain function in Parkinson's disease patients. These criteria include indicators such as model accuracy, sensitivity, and specificity. For example, the model is required to achieve an accuracy of over 90% on the test set, with sensitivity and specificity both not less than 85%. These criteria allow for the selection of the optimal model, ensuring its reliability in practical applications. Acquiring multimodal data is fundamental to monitoring. This includes patient motion data (such as gait analysis), speech data, and electroencephalogram (EEG) data. Gait analysis may record the patient's walking speed, stride length, and balance. Speech data may include the patient's pitch, volume, and clarity. EEG data may focus on electrical activity patterns in specific brain regions. Data preprocessing is a key step in improving model performance. For gait data, outliers, such as extreme values ​​due to equipment malfunction, may need to be removed. Speech data may require noise reduction to remove background noise. EEG data may require filtering to remove electromyographic interference. Inputting the preprocessed data into the optimal model for prediction is the core of monitoring. Suppose a deep learning model is used, it might output a value between 0 and 1, representing the probability of a patient having abnormal brain function. A threshold of 0.7 might be set; values ​​above this are considered abnormal. For patients diagnosed with abnormalities, further analysis using a Support Vector Machine (SVM) can provide more detailed information. The SVM might classify the patient as having mild, moderate, or severe abnormalities, which helps doctors develop more precise treatment plans. The application of decision tree algorithms can help better understand the specific stage of a patient's disease. For example, combining a patient's clinical manifestations, such as the degree of tremor and muscle rigidity, a decision tree might classify the patient as having early, middle, or late-stage Parkinson's disease. This classification is invaluable for developing long-term treatment strategies. Finally, the model's predictions, the SVM's abnormality assessment, and the decision tree's disease staging results are combined to form a comprehensive monitoring report. This report might include the patient's current brain function status, the specific manifestations of the abnormalities, and the stage of disease progression. This information can help doctors develop personalized treatment plans, such as adjusting medication dosages or recommending specific physical therapies. This multimodal, multi-algorithm integrated analysis method can more comprehensively and accurately monitor the brain function status of Parkinson's disease patients, providing them with better medical services.

[0063] This embodiment first acquires multimodal data including electroencephalography (EEG), magnetic resonance imaging (MRI), and clinical scales. Through standardization and principal component analysis (PCA) dimensionality reduction, the main features of each modality are extracted. Then, dynamic time warping is used to align the time-series data of different modalities, and highly correlated features are selected using mutual information. Finally, the optimized feature subset is input into a pre-defined deep neural network model, and a multilayer perceptron structure is used to model the nonlinear relationships between the multimodal data, achieving accurate prediction and monitoring of brain function status in Parkinson's disease. This invention, through multimodal data fusion and deep learning technology, effectively improves the accuracy and reliability of monitoring brain function status in Parkinson's disease, providing important reference for clinical diagnosis and treatment.

[0064] Example 2

[0065] The present invention also provides a Parkinson's disease brain function monitoring system based on multimodal data, the system comprising:

[0066] The data acquisition and standardization module is used to acquire raw multimodal Parkinson's disease monitoring data and standardize the raw multimodal Parkinson's disease monitoring data to obtain a standardized multimodal dataset.

[0067] The feature extraction and dimensionality reduction module is used to perform feature dimensionality reduction on the standardized multimodal dataset to obtain a dimensionality-reduced feature set;

[0068] The time alignment module is used to align the dimensionality-reduced feature set to obtain a time-aligned multimodal feature set;

[0069] The feature optimization module is used to calculate the correlation between features in the time-aligned multimodal feature set to obtain an optimized feature subset;

[0070] The monitoring module is used to identify the monitoring results of brain function status in Parkinson's disease based on the optimized feature subset.

[0071] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for monitoring brain function in Parkinson's disease based on multimodal data, characterized in that, Includes the following steps: The original multimodal Parkinson's disease surveillance data were standardized to obtain a standardized multimodal dataset; The data features of the standardized multimodal dataset are extracted to obtain the dimensionality-reduced feature set; Aligning the dimensionality-reduced feature set based on the time series data yields a time-aligned multimodal feature set. The correlation between features in the time-aligned multimodal feature set is calculated to obtain an optimized feature subset; The optimized feature subset is input into a preset deep neural network model for identification to obtain the monitoring results of brain function status in Parkinson's disease.

2. The method for monitoring Parkinson's disease brain function based on multimodal data according to claim 1, characterized in that, The multimodal dataset includes: electroencephalogram (EEG), magnetic resonance imaging (MRI), and clinical scale data.

3. The method for monitoring Parkinson's disease brain function based on multimodal data according to claim 2, characterized in that, The process of standardizing raw multimodal Parkinson's disease surveillance data includes: A standardized EEG was obtained by removing high-frequency noise and baseline drift from the EEG using a combination of wavelet transform and bandpass filtering. The artifacts in the magnetic resonance imaging are eliminated by using bias field correction and intensity non-uniformity correction algorithms to obtain standardized magnetic resonance imaging. Data cleaning and outlier detection methods were used to remove outliers from the clinical scale data to obtain standardized clinical scale data. A standardized multimodal dataset was obtained based on the standardized electroencephalogram, standardized magnetic resonance imaging, and standardized clinical scale data.

4. The method for monitoring Parkinson's disease brain function based on multimodal data according to claim 3, characterized in that, Principal component analysis was used to process the standardized multimodal dataset to obtain a dimensionality-reduced feature set.

5. The method for monitoring Parkinson's disease brain function based on multimodal data according to claim 4, characterized in that, The process of obtaining time-aligned multimodal feature sets includes: The alignment results are obtained by aligning time series data of different modalities using dynamic time warping techniques. Based on the alignment results, the dimensionality-reduced feature sets are fused to obtain a time-aligned multimodal fusion feature set.

6. The method for monitoring Parkinson's disease brain function based on multimodal data according to claim 5, characterized in that, The process of obtaining the optimized feature subset includes: For each feature in the time-aligned multimodal fusion feature set, calculate the mutual information value between the current feature and other features; Based on the mutual information values, a correlation matrix between features is obtained; The candidate feature subset is obtained by filtering the correlation matrix between the features; The optimized feature subset is obtained by performing cluster analysis on the candidate feature subset based on the clustering algorithm.

7. The method for monitoring Parkinson's disease brain function based on multimodal data according to claim 6, characterized in that, The process of inputting the optimized feature subset into a preset deep neural network model for identification to obtain the monitoring results of Parkinson's disease brain function status includes: Preliminary prediction results are obtained by performing nonlinear relationship modeling on the optimized feature subset based on the multilayer perceptron structure in the deep neural network model. The preliminary prediction results are compared with a preset standard threshold. If the preliminary prediction result does not exceed the preset standard threshold, the patient's current brain function is judged to be normal; if the preliminary prediction result exceeds the preset standard threshold, the patient's current brain function is judged to be abnormal. The support vector machine algorithm was used to determine the type and degree of brain function abnormalities in patients with abnormal brain function status, thus obtaining the monitoring results of brain function status in Parkinson's disease.

8. A Parkinson's disease brain function monitoring system based on multimodal data, characterized in that, For implementing the Parkinson's disease brain function monitoring method based on multimodal data as described in claim 1, the system comprises: The data acquisition and standardization module is used to acquire raw multimodal Parkinson's disease monitoring data and standardize the raw multimodal Parkinson's disease monitoring data to obtain a standardized multimodal dataset. The feature extraction and dimensionality reduction module is used to perform feature dimensionality reduction on the standardized multimodal dataset to obtain a dimensionality-reduced feature set; The time alignment module is used to align the dimensionality-reduced feature set to obtain a time-aligned multimodal feature set; The feature optimization module is used to calculate the correlation between features in the time-aligned multimodal feature set to obtain an optimized feature subset; The monitoring module is used to identify the monitoring results of brain function status in Parkinson's disease based on the optimized feature subset.

Images