Alzheimer's disease patient brain entropy monitoring method and system based on deep learning

Abstract

The invention belongs to the technical field of electroencephalogram signal processing, and relates to an Alzheimer's disease patient brain entropy monitoring method and system based on deep learning, and the method comprises the steps: collecting brain signal data of a target Alzheimer's disease patient, carrying out the noise reduction, extracting a multi-level brain entropy value based on a denoised brain signal sequence, and determining a brain entropy index sequence; acquiring local fluctuation characteristics in a time window according to the brain entropy index sequence, and inputting the local fluctuation characteristics into a convolutional neural network model to judge a potential abnormal mode; extracting a dynamic change vector related to disease progress from the abnormal mode, and inputting the dynamic change vector into a support vector machine classifier to obtain classified disease stage labels; obtaining the difference degree between adjacent stages according to the disease stage labels, evaluating the difference degree, and determining key nodes; and acquiring context data of the brain entropy features corresponding to the key nodes, matching the context data through a sequence alignment algorithm, and outputting a personalized progress early warning signal. The accuracy and individuation level of disease progress monitoring can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroencephalogram signal processing, in particular to an Alzheimer's disease patient brain entropy monitoring method and system based on deep learning. BACKGROUND

[0002] Alzheimer's disease, as a neurodegenerative disease that seriously affects the cognitive function of the elderly, its early diagnosis and disease progression monitoring is of vital importance to improve the quality of life of patients. The research in this field is not only the focus of the medical field, but also the focus of social attention, because timely detection of disease signals can gain valuable time for intervention treatment. However, although the current research and monitoring methods can capture the changes in brain activity to some extent, there are still significant shortcomings, and new technical breakthroughs are needed to cope with complex challenges.

[0003] Existing monitoring methods often fail to fully capture the dynamic changes of brain activity, especially when faced with individual differences and the diversity of disease development, often cannot accurately distinguish the disease characteristics of different stages. These methods rely on fixed indicators or limited analysis dimensions, ignoring the complex patterns and potential laws hidden behind brain activity, resulting in inaccurate judgment of disease progression and missing the key opportunity for early intervention. This limitation makes researchers need to find a more adaptive and deep mining way to reveal the essential characteristics of the disease. In this context, brain entropy, as an important indicator to measure the complexity and chaos of brain activity, has become the core focus of research. However, the processing and analysis of brain entropy data face huge technical difficulties. The first challenge is how to extract the feature patterns that are truly related to disease progression from the massive and high-dimensional brain signals, because these signals often contain a lot of irrelevant noise and are difficult to identify directly. Further, this difficulty in feature extraction brings another problem, that is, how to identify the key change nodes of disease development in data from different patients and different time points, such as abnormal fluctuations in brain entropy patterns at a certain stage may indicate disease deterioration, but existing technologies are difficult to accurately locate such changes. Therefore, how to accurately extract brain entropy features related to Alzheimer's disease progression from complex brain signals and identify key turning points in disease development through these features has become a key problem that needs to be solved. The solution to this problem not only concerns technical breakthroughs, but also directly affects whether timely warnings and personalized intervention plans can be provided for patients. SUMMARY

[0004] The purpose of the present application is to provide an Alzheimer's disease patient brain entropy monitoring method and system based on deep learning, which integrates signal noise reduction, feature extraction, pattern recognition and classification warning, realizes the whole process of automatic analysis from raw signals to accurate warning, and finally improves the accuracy and individualization level of disease progression monitoring.

[0005] To achieve the above object, the present application provides the following scheme:

[0006] A deep learning-based brain entropy monitoring method for Alzheimer's disease patients, comprising:

[0007] Collecting brain signal data of a target Alzheimer's disease patient and performing noise reduction processing to obtain a denoised brain signal sequence;

[0008] Extracting multi-level brain entropy values based on the denoised brain signal sequence to determine a brain entropy index sequence;

[0009] Obtaining local fluctuation features within a time window according to the brain entropy index sequence, inputting a pre-trained convolutional neural network model for pattern recognition to determine potential abnormal patterns;

[0010] Extracting a dynamic change vector related to disease progression from the abnormal patterns, inputting a pre-trained support vector machine classifier for classification processing to obtain a classified disease stage label;

[0011] Obtaining the difference degree between adjacent stages according to the disease stage label, evaluating the difference degree by threshold comparison method to determine the key node;

[0012] Obtaining context data of brain entropy features corresponding to the key node, matching the context data by sequence alignment algorithm, and outputting an individualized progress warning signal, i.e., the brain entropy monitoring result of the target Alzheimer's disease patient.

[0013] To further achieve the above object, the present application further provides a deep learning-based brain entropy monitoring system for Alzheimer's disease patients, comprising:

[0014] A brain signal acquisition module for acquiring brain signal data of a target Alzheimer's disease patient and performing noise reduction processing to obtain a denoised brain signal sequence;

[0015] A brain entropy index extraction module for extracting multi-level brain entropy values based on the denoised brain signal sequence to determine a brain entropy index sequence;

[0016] An abnormal pattern judgment module for obtaining local fluctuation features within a time window according to the brain entropy index sequence, inputting a pre-trained convolutional neural network model for pattern recognition to determine potential abnormal patterns;

[0017] A disease stage classification module for extracting a dynamic change vector related to disease progression from the abnormal patterns, inputting a pre-trained support vector machine classifier for classification processing to obtain a classified disease stage label;

[0018] A key node determination module is configured to obtain a difference degree between adjacent stages according to the disease stage label, evaluate the difference degree by a threshold comparison method, and determine a key node;

[0019] A progress warning output module is configured to obtain context data of the brain entropy feature corresponding to the key node, match the context data by a sequence comparison algorithm, and output an individualized progress warning signal, i.e., a brain entropy monitoring result of the target Alzheimer's disease patient.

[0020] The present application has the following advantages:

[0021] The present application solves the problem of how to accurately identify the key node of disease progression from complex signals and provide individualized warning by performing multi-level processing and analysis on the brain signal data of Alzheimer's disease patients. The present application first denoises high-dimensional signals by wavelet transform to extract clear signals, and then calculates brain entropy index sequences by multi-scale entropy to capture multi-level features of the signals. Subsequently, the present application identifies abnormal patterns in local fluctuation features within a time window by a convolutional neural network, and extracts dynamic change vectors related to disease progression. Further, the present application classifies vector sets by a support vector machine classifier to determine disease stage labels, evaluates difference degrees between adjacent stages by threshold comparison, and locates key nodes. Finally, the present application performs sequence comparison based on context data of the key nodes to generate individualized warning signals. The present application seamlessly integrates signal denoising, feature extraction, pattern recognition, and classification warning to realize full-process automatic analysis from raw signals to accurate warning, and ultimately improves the accuracy and individualization level of disease progression monitoring to provide important support for clinical decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0023] Figure 1 A flow chart of an Alzheimer's disease patient brain entropy monitoring method based on deep learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] This embodiment provides a deep learning-based method for monitoring brain entropy in Alzheimer's patients, such as... Figure 1 As shown, it includes:

[0027] Collect brain signal data from target Alzheimer's patients and perform noise reduction processing to obtain the denoised brain signal sequence;

[0028] Based on the denoised brain signal sequence, multi-level brain entropy values ​​are extracted to determine the brain entropy index sequence.

[0029] The local fluctuation features within the time window are obtained based on the brain entropy index sequence, and then input into a pre-trained convolutional neural network model for pattern recognition to identify potential abnormal patterns.

[0030] Extract dynamic change vectors related to disease progression from the abnormal patterns, input them into a pre-trained support vector machine classifier for classification, and obtain disease stage labels after classification.

[0031] The difference between adjacent stages is obtained based on the disease stage labels, and the difference is evaluated by a threshold comparison method to determine key nodes;

[0032] Contextual data of brain entropy features corresponding to the key nodes are obtained, and the contextual data is matched by a sequence alignment algorithm to output personalized progression warning signals, namely the brain entropy monitoring results of the target Alzheimer's patient.

[0033] Specifically, this embodiment addresses the challenge of accurately identifying key disease progression points and providing personalized early warnings from complex signals by performing multi-level processing and analysis of brain signal data from Alzheimer's patients. First, wavelet transform is used to denoise high-dimensional signals and extract clear signals. Then, multi-scale entropy calculation is used to obtain brain entropy index sequences, capturing the multi-level features of the signals. Next, a convolutional neural network is used to identify abnormal patterns in local fluctuation features within a time window and extract dynamic change vectors related to disease progression. Furthermore, a support vector machine classifier is used to classify the vector set, determine disease stage labels, and evaluate the differences between adjacent stages through threshold comparison to locate key nodes. Finally, sequence alignment is performed based on the context data of the key nodes to generate personalized early warning signals. This embodiment seamlessly integrates signal denoising, feature extraction, pattern recognition, and classification-based early warning, achieving fully automated analysis from raw signals to accurate early warnings. Ultimately, this improves the accuracy and personalization of disease progression monitoring, providing crucial support for clinical decision-making.

[0034] Further, the brain signal data of the target Alzheimer's disease patient is collected and denoising processing is performed to obtain a denoised brain signal sequence, comprising:

[0035] The original brain signal data of the target Alzheimer's disease patient is obtained from the head of the target Alzheimer's disease patient by a signal acquisition device and stored as an initial signal set;

[0036] The high-dimensional data in the initial signal set is decomposed by using a wavelet transform method to separate different frequency components and obtain a decomposed signal component;

[0037] If there is a frequency component with an energy value exceeding a preset threshold in the decomposed signal component, the frequency component is marked as a noise component, and the marked noise component is filtered out to reconstruct the remaining signal component to obtain a denoised brain signal sequence.

[0038] Specifically, the high-precision electroencephalogram device is used to collect the brain electrical signal data of the patient in the resting state for 10 minutes, the sampling frequency is 500 Hz, a high-dimensional signal sequence containing 300,000 sampling points is obtained, the data is stored in the form of a time sequence matrix, the matrix dimension is 300,000x1, the signal value range is between-50 microvolts and 50 microvolts, and contains noise interference. Then, the denoising processing link is entered, the wavelet transform method is used for denoising, the Daubechies wavelet basis (db4) is specifically selected as the decomposition tool, the signal is decomposed into 5 layers of wavelet coefficients, the high-frequency coefficients are processed by the soft threshold method, the threshold is set to 2 times the signal standard deviation, that is, assuming that the standard deviation is 5.2 microvolts, the threshold is 10.4 microvolts, the high-frequency coefficients higher than the threshold are retained, and the low-frequency coefficients below the threshold are set to 0 to filter out the noise. Then, the wavelet coefficients after processing are inversely transformed to reconstruct the denoised signal sequence, the signal value range is reduced to-40 microvolts to 40 microvolts, and the noise component is significantly reduced.

[0039] Further, based on the denoised brain signal sequence, multi-level brain entropy values are extracted to determine a brain entropy index sequence, comprising:

[0040] Based on the denoised brain signal sequence, multi-scale entropy calculation method is used to extract brain entropy value data at different levels;

[0041] The brain entropy value data is used to construct a multi-level brain entropy value distribution structure, and if there is an abnormal fluctuation in the brain entropy value distribution structure, the local signal reconstruction is performed on the abnormal part to obtain an adjusted brain entropy value distribution;

[0042] According to the adjusted brain entropy value distribution, an entropy value aggregation method is used to determine the final brain entropy index sequence.

[0043] Specifically, based on the denoised brain signal sequence, the multi-scale entropy calculation method is used to extract brain entropy value data at different levels. First, assuming that a denoised brain signal sequence is obtained, the length is 1000 sampling points, the sampling frequency is 100 Hz, and the signal value range is between -50 and 50 microvolts. Then, the multi-scale entropy calculation algorithm is implemented, specifically the sample entropy method is combined with multi-scale analysis, and the signal sequence is coarse-grained according to different scale factors (for example, scale 1 to 5). Taking scale factor 2 as an example, the average value of every 2 points of the original sequence is taken to form a new sequence with a length of 500, and the sample entropy value is calculated. The embedding dimension m is set to 2, and the tolerance r is 0.2 times the standard deviation of the signal. The sample entropy value is 1.25. Similarly, scales 1 to 5 are calculated respectively to obtain a series of entropy values such as 1.30, 1.25, 1.18, 1.10, and 1.05, reflecting the complexity changes of the signal at different scales. Then, these entropy values are arranged in order of scale to form a brain entropy index sequence [1.30, 1.25, 1.18, 1.10, 1.05], which is used for subsequent feature analysis.

[0044] Further, according to the brain entropy index sequence, the local fluctuation features in the time window are obtained, and the pre-trained convolutional neural network model is input for pattern recognition to determine the potential abnormal pattern, including:

[0045] The brain entropy index sequence is segmented to obtain local fluctuation data from each time window, and a fluctuation signal set is obtained.

[0046] The fluctuation signal set is subjected to noise interference elimination by a standardization method to obtain a processed fluctuation signal set.

[0047] Based on the processed fluctuation signal set, local fluctuation features are extracted to construct a feature vector group, and the distribution range of the feature vector is determined. If the distribution range of the feature vector exceeds the preset range threshold, the abnormal point is marked, and the marked feature vector group is obtained.

[0048] The marked feature vector group is classified by the convolutional neural network model to determine whether there is a potential abnormal pattern. If there is, the frequency and time distribution of the abnormal pattern are analyzed to determine the time correlation of the abnormal pattern. If the time correlation of the abnormal pattern meets the preset continuity condition, the brain entropy index sequence in the related time window is subjected to in-depth analysis to obtain the final abnormal pattern determination result.

[0049] Specifically, the present embodiment aims at the processing and analysis of brain entropy index sequence, and designs a systematic method to extract local fluctuation features within a time window, and judges potential abnormalities through convolutional neural network pattern recognition. First, assuming that a brain entropy index sequence data is obtained, the data sampling frequency is 1 Hz, the time length is 600 seconds, there are 600 data points, and the numerical range is between 0.5 and 2.5, reflecting the complexity change of brain activity. Automatically divide the data into multiple time windows, and set the length of each time window to 30 seconds, that is, 30 data points, and overlap between time windows by 15 seconds to capture continuity features. In each time window, calculate the local fluctuation features, such as standard deviation and frequency energy, and the specific algorithm is: calculate the standard deviation of 30 data points in each time window to obtain the fluctuation amplitude, such as a time window standard deviation of 0.3; At the same time, extract the main frequency component through fast Fourier transform (FFT), and assume that the main frequency energy is concentrated at 2Hz, and the energy value is 1.8, which is used to represent periodic fluctuations. These feature values are stored as a two-dimensional feature matrix, with each row representing a time window and each column representing a different feature (such as standard deviation and main frequency energy). Then, input the feature matrix into the convolutional neural network (CNN) model, the model structure includes two convolutional layers (filter size 3x3, step 1) and a fully connected layer, the activation function is ReLU, the training data set is 1000 labeled samples (500 normal, 500 abnormal), the loss function is cross entropy, the optimizer is Adam, and the learning rate is 0.001. After 50 rounds of training, the model accuracy reaches 85%. In the inference stage, each time window feature is predicted, and if the abnormal probability of a certain time window exceeds 0.7, it is marked as a potential abnormal pattern, for example, a time window prediction probability of 0.75, automatically records and generates an abnormal report, and at the same time, correlates historical data analysis to confirm whether it forms a trend with other abnormal time windows within the past 30 minutes. If the abnormal probability of three consecutive time windows exceeds 0.7, an advanced warning is triggered.

[0050] Further, a dynamic change vector related to disease progression is extracted from the abnormal pattern, input into a pre-trained support vector machine classifier for classification processing, and a classified disease stage label is obtained, including:

[0051] The abnormal pattern is preliminarily screened through pattern detection technology, and the abnormal pattern related to disease progression is obtained and further analyzed to obtain change trend information related to disease progression;

[0052] The vector extraction method is used to convert the dynamic change in the change trend information into a change vector, and the pattern analysis technology is used to perform secondary verification on the distribution and trend of the change vector to judge the matching degree of the change vector with the progression tracking;

[0053] If the matching degree meets a preset matching threshold, the change vector is continuously monitored by a progress tracking mechanism to obtain dynamic update data of disease progress.

[0054] According to the dynamic update data, a support vector machine classifier is used to finally classify the relevance of the change vector to the disease progress, determine the influence range of the abnormal pattern on the disease progress, and output a disease stage label.

[0055] Specifically, the embodiment extracts a dynamic change vector related to disease progress from the abnormal pattern, calculates the slope of heart rate change, obtains a rising stage slope of 1.5 times / minute from 75 to 120, and a falling stage slope of -1.33 times / minute, and constructs a multidimensional vector combining blood pressure data (assuming synchronous rise of 10 mmHg) to form a dynamic change vector (1.5, -1.33, 10) and put it into a dynamic change vector set.

[0056] The support vector machine algorithm is used for classification, the radial basis function kernel function is selected, the kernel parameter γ is set to 0.1, the regularization parameter C is set to 1.0, the data set is divided into 80% training set (800 samples) and 20% test set (200 samples) by cross-validation method to optimize model parameters and prevent overfitting. During the training process, the algorithm maps the data to a high-dimensional space by finding the optimal hyperplane, calculates the distance from each sample to the hyperplane, maximizes the interval between classes, for example, the disease stage label is divided into three categories: early, medium and late. The model will calculate the support vector based on the distribution of the feature vector to obtain the classification boundary and output the disease stage label of each patient.

[0057] Further, according to the disease stage label, the difference degree between adjacent stages is obtained, the difference degree is evaluated by threshold comparison method, and the key node is determined, including:

[0058] The disease stage data corresponding to the disease stage label is extracted, the stage sequence data set is constructed, and the stage division information is obtained.

[0059] According to the stage division information, the data feature difference between adjacent stages is identified, and the difference degree quantization value is obtained by using the standardization processing method.

[0060] The difference degree quantization value is judged by the threshold comparison method, the key node is determined, and the key node is analyzed by stage, the importance ranking of the key node is obtained by combining the context relationship of the key node and adjacent stages.

[0061] According to the importance ranking, the key node is screened, and the logical regression model is used to verify the positioning of the screened key node to determine the final key node set.

[0062] Specifically, the embodiment first assumes that there is a data set with five disease stage labels, stages 1 to 5, and each stage has corresponding feature values such as severity scores, 10, 15, 25, 40, and 50. For the difference degree calculation between adjacent stages, a simple numerical difference algorithm is used, that is, the score of the next stage is subtracted from the score of the previous stage, and the difference degree sequence is 5, 10, 15, and 10. By analyzing the difference degree sequence, it can be found that the difference degree from stage 2 to stage 3 is 10, and the difference degree from stage 3 to stage 4 is 15, showing a significant jump change. To evaluate the importance of the difference degree, a threshold is set, for example, the threshold is 12, and the threshold comparison method is used to compare the difference degree with the threshold one by one, and it is found that the difference degree from stage 3 to stage 4 is 15, which is greater than the threshold 12, indicating that this is a significant change point, which may be a key node. To further confirm the position of the key node, auxiliary business data such as patient treatment response rate is introduced, and it is assumed that the response rate from stage 3 to stage 4 decreases from 80% to 50%, with a decrease of 30%, and the decrease of other adjacent stages is less than 10%. By calculating the correlation between the response rate decrease and the difference degree, the Pearson correlation coefficient is calculated, and the correlation coefficient is 0.85, indicating that the difference degree is highly correlated with the response rate change, thereby confirming that stage 3 to stage 4 is a key node.

[0063] Further, the context data corresponding to the brain entropy feature of the key node is obtained, and the context data is matched by a sequence alignment algorithm to output an individualized progress warning signal, including:

[0064] The brain entropy feature data related to the position of the key node is separated by data extraction technology, and the context data associated with the brain entropy feature data is obtained. The context data is sorted by a hierarchical screening method to determine the context segment highly correlated with the brain entropy feature data;

[0065] The context segment is matched layer by layer by a sequence alignment algorithm, and the risk segment is marked according to the matching result;

[0066] The individualized signal features in the risk segment are extracted, and the signal features are deeply mined by feature analysis technology to obtain an individualized signal pattern;

[0067] The individualized signal pattern is dynamically compared in time sequence by a progress warning mechanism, and if the comparison result shows continuous abnormality, a warning identifier is generated;

[0068] According to the warning identifier, the position information and the brain entropy feature of the key node are integrated, and the warning identifier is converted into a progress warning signal by a preset mapping rule to determine the final output result.

[0069] Specifically, the embodiment first extracts the brain entropy features of the key node positions, selects a time window of 5 seconds and a sampling rate of 256 Hz, calculates the brain entropy values in each time window, obtains a series of numerical values such as 0.72, 0.68, 0.75, and forms a brain entropy sequence with a length of 1000 as the basic data. Then, for the context data acquisition of these key nodes, the sliding window method is used to extract the EEG signal data of 10 seconds before and after, calculate the spectral features of the context, for example, the average value of the power spectral density in the alpha frequency band (8-12 Hz) is 3.5 microvolts per square / Hz, and combine it with the brain entropy sequence to form a multi-dimensional feature vector. Subsequently, in the sequence comparison stage, the dynamic time warping algorithm (DTW) is used to match the context data and the brain entropy sequence in the historical database, calculate the Euclidean distance between the two sequences, set the threshold value to 0.1, and if the distance is less than 0.1, it is considered to be a successful match. The matching result shows that the distance between the current sequence and a certain historical abnormal mode sequence is 0.08, indicating a potential risk. Finally, based on the matching result, a personalized progress warning signal is generated, combined with a machine learning model such as support vector machine (SVM), the matching distance 0.08 and the context feature value are input, and the output warning probability is 85%. According to the probability value classification, if it is greater than 80%, it is marked as high risk, and the warning notification mechanism is automatically triggered.

[0070] To further optimize the technical scheme, the embodiment also provides an Alzheimer's disease patient brain entropy monitoring system based on deep learning, comprising:

[0071] A brain signal acquisition module is configured to acquire brain signal data of a target Alzheimer's disease patient and perform noise reduction processing to obtain a denoised brain signal sequence.

[0072] A brain entropy index extraction module is configured to extract multi-level brain entropy values based on the denoised brain signal sequence and determine a brain entropy index sequence.

[0073] An abnormal mode judgment module is configured to acquire local fluctuation features within a time window based on the brain entropy index sequence, input a pre-trained convolutional neural network model for pattern recognition, and judge potential abnormal modes.

[0074] A disease stage classification module is configured to extract dynamic change vectors related to disease progression from the abnormal modes, input a pre-trained support vector machine classifier for classification processing, and obtain classified disease stage labels.

[0075] A key node determination module is configured to acquire the difference degree between adjacent stages based on the disease stage labels, evaluate the difference degree through a threshold comparison method, and determine key nodes.

[0076] The progress early warning output module is configured to acquire context data corresponding to the brain entropy feature of the key node, match the context data through a sequence alignment algorithm, and output an individualized progress early warning signal, i.e., a brain entropy monitoring result of the target Alzheimer's disease patient.

[0077] The above-described embodiments are merely used to describe the preferred modes of the present application, and are not used to limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope of the present application as defined by the claims.

Claims

1. A method for monitoring brain entropy in Alzheimer's patients based on deep learning, characterized in that, include: Collect brain signal data from target Alzheimer's patients and perform noise reduction processing to obtain the denoised brain signal sequence; Based on the denoised brain signal sequence, multi-level brain entropy values ​​are extracted to determine the brain entropy index sequence. The local fluctuation features within the time window are obtained based on the brain entropy index sequence, and then input into a pre-trained convolutional neural network model for pattern recognition to identify potential abnormal patterns. Extract dynamic change vectors related to disease progression from the abnormal patterns, input them into a pre-trained support vector machine classifier for classification, and obtain disease stage labels after classification. The difference between adjacent stages is obtained based on the disease stage labels, and the difference is evaluated by a threshold comparison method to determine key nodes; Contextual data of brain entropy features corresponding to the key nodes are obtained, and the contextual data is matched by a sequence alignment algorithm to output personalized progression warning signals, namely the brain entropy monitoring results of the target Alzheimer's patient.

2. The method for monitoring brain entropy in Alzheimer's patients based on deep learning according to claim 1, characterized in that, Brain signal data from target Alzheimer's patients was collected and denoised to obtain denoised brain signal sequences, including: Raw brain signal data is acquired from the head of a target Alzheimer's patient using a signal acquisition device and stored as an initial signal set. The high-dimensional data in the initial signal set is decomposed using wavelet transform to separate different frequency components and obtain the decomposed signal components. If there are frequency components in the decomposed signal components whose energy values ​​exceed a preset threshold, the frequency components are marked as noise components, and the marked noise components are filtered out to reconstruct the remaining signal components and obtain the denoised brain signal sequence.

3. The method for monitoring brain entropy in Alzheimer's patients based on deep learning according to claim 1, characterized in that, Based on the denoised brain signal sequence, multi-level brain entropy values ​​are extracted to determine the brain entropy index sequence, including: Based on the denoised brain signal sequence, a multi-scale entropy calculation method is used to extract brain entropy data at different levels. A multi-level brain entropy distribution structure is constructed using the brain entropy data. If there are abnormal fluctuations in the brain entropy distribution structure, the abnormal part is reconstructed locally to obtain the adjusted brain entropy distribution. Based on the adjusted brain entropy value distribution, the final brain entropy index sequence is determined using entropy value aggregation.

4. The method for monitoring brain entropy in Alzheimer's patients based on deep learning according to claim 1, characterized in that, Local fluctuation features within a time window are obtained based on the brain entropy index sequence, and input into a pre-trained convolutional neural network model for pattern recognition to identify potential abnormal patterns, including: The brain entropy index sequence is segmented to obtain local fluctuation data from each time window, thus obtaining a set of fluctuation signals. The set of wave signals is subjected to noise interference elimination using a standardized method to obtain a processed set of wave signals. Local fluctuation features are extracted from the processed set of fluctuation signals, a feature vector group is constructed, the distribution range of the feature vectors is determined, and if the distribution range of the feature vectors exceeds the preset range threshold, the outlier is marked and the marked feature vector group is obtained. The labeled feature vector groups are classified using a convolutional neural network model to determine whether there are potential abnormal patterns. If so, the frequency and temporal distribution of the abnormal patterns are analyzed to determine their temporal correlation. If the temporal correlation of the abnormal patterns meets the preset continuity conditions, the brain entropy index sequence within the relevant time window is analyzed in depth to obtain the final abnormal pattern determination result.

5. The method for monitoring brain entropy in Alzheimer's patients based on deep learning according to claim 1, characterized in that, Dynamic change vectors related to disease progression are extracted from the abnormal patterns and input into a pre-trained support vector machine classifier for classification to obtain disease stage labels, including: The abnormal patterns are initially screened by pattern detection technology to obtain abnormal patterns related to disease progression and conduct in-depth analysis to obtain information on the changing trends associated with disease progression. The dynamic changes in the trend information are transformed into change vectors using vector extraction methods, and the distribution and trend of the change vectors are verified a second time using pattern analysis techniques to determine the degree of matching between the change vectors and the progress tracking. If the matching degree meets the preset matching threshold, the change vector is continuously monitored through the progress tracking mechanism to obtain dynamic update data on disease progression; Based on the dynamically updated data, a support vector machine classifier is used to perform a final classification on the correlation between the change vector and disease progression, determine the scope of the impact of abnormal patterns on disease progression, and output disease stage labels.

6. The method for monitoring brain entropy in Alzheimer's patients based on deep learning according to claim 1, characterized in that, The differences between adjacent stages are obtained based on the disease stage labels. These differences are then evaluated using a threshold comparison method to identify key nodes, including: Extract the disease stage data corresponding to the disease stage labels, construct a stage sequence dataset, and obtain stage division information; Based on the stage division information, the differences in data characteristics between adjacent stages are identified, and a standardized processing method is used to obtain a quantitative value of the degree of difference. The difference level is quantified by a threshold comparison method to determine key nodes, and a phase analysis is performed on the key nodes. The importance ranking of the key nodes is obtained by combining the contextual relationship between the key nodes and adjacent phases. Key nodes are selected based on their importance ranking, and the location of the selected key nodes is verified using a logistic regression model to determine the final set of key nodes.

7. The method for monitoring brain entropy in Alzheimer's patients based on deep learning according to claim 1, characterized in that, Contextual data of brain entropy features corresponding to the key nodes are obtained, and the contextual data is matched using a sequence alignment algorithm to output personalized progression warning signals, including: Brain entropy feature data related to key node locations is separated by data extraction technology, and context data associated with the brain entropy feature data is obtained. The context data is organized by a hierarchical screening method to determine the context fragments that are highly related to the brain entropy feature data. The context segments are matched layer by layer using a sequence alignment algorithm, and risk segments are marked based on the matching results; Personalized signal features are extracted from the risk segments, and the signal features are deeply mined through feature analysis technology to obtain personalized signal patterns; The progress warning mechanism is used to dynamically compare personalized signal patterns over time. If the comparison results show continuous abnormalities, a warning sign is generated. Based on the warning identifier, the location information and brain entropy features of key nodes are integrated, and the warning identifier is converted into a progress warning signal using a preset mapping rule to determine the final output result.

8. A brain entropy monitoring system for Alzheimer's patients based on deep learning, characterized in that, include: The brain signal acquisition module is used to collect brain signal data from target Alzheimer's patients and perform noise reduction processing to obtain the denoised brain signal sequence. The brain entropy index extraction module is used to extract multi-level brain entropy values ​​based on the denoised brain signal sequence and determine the brain entropy index sequence. The abnormal pattern judgment module is used to obtain local fluctuation features within a time window based on the brain entropy index sequence, input them into a pre-trained convolutional neural network model for pattern recognition, and judge potential abnormal patterns. The disease stage classification module is used to extract dynamic change vectors related to disease progression from the abnormal patterns, input them into a pre-trained support vector machine classifier for classification processing, and obtain the classified disease stage labels. The key node determination module is used to obtain the degree of difference between adjacent stages based on the disease stage label, evaluate the degree of difference through a threshold comparison method, and determine the key nodes. The progression warning output module is used to acquire contextual data of the brain entropy features corresponding to the key nodes, match the contextual data through a sequence alignment algorithm, and output a personalized progression warning signal, namely the brain entropy monitoring result of the target Alzheimer's patient.

Images