Neurological disease detection and analysis method and system

Abstract

The invention discloses a nerve disease detection and analysis method and system, and the method comprises the steps: obtaining a bracelet collection signal, a sphygmomanometer collection signal, a movement behavior image and behavior test data, and extracting tremor intensity features, gait symmetry features and autonomic nerve rhythm features through multi-band decomposition of the bracelet collection signal; analyzing the motion behavior image and the standardized motion test to obtain a motion function score; carrying out heart rate variability analysis to identify a neural function abnormality mode; constructing a neural function state map and calculating a feature weight; predicting a disease progress trend in combination with historical monitoring data; and dynamically adjusting a prediction result through subsequent feedback correction information. Through a mode of combining short-time intensive monitoring and long-term intermittent acquisition, long-term trend prediction and dynamic correction based on initial data are realized, and the reliability and practicability of nerve disease risk assessment in a home scene are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical detection technology, and in particular to a neurological disease detection and analysis method and system. Background Art

[0002] Early risk assessment of neurodegenerative diseases is crucial for slowing disease progression. Currently, clinical diagnosis relies primarily on regular hospital checkups. This approach suffers from long monitoring intervals and discrete data, making it difficult to capture early changes in the disease. In recent years, the use of wearable devices has made home monitoring possible, but existing technical solutions still have significant limitations.

[0003] Traditional monitoring methods typically require patients to wear devices for extended periods of time and collect complete data. This presents challenges in practical applications, such as poor compliance and unstable data quality. Furthermore, existing analytical models often rely on static assessments based on a single, complete dataset. This makes it impossible to infer long-term trends from short-term, intensive monitoring, nor does it adapt to subsequent, sparse sampling. More importantly, most systems only provide current status assessments and fail to establish effective dynamic prediction mechanisms to anticipate the trajectory of disease progression.

[0004] Furthermore, existing prediction models often use fixed parameters, making them difficult to adapt to individual differences and the nonlinear nature of disease progression. When new monitoring data becomes available, the system typically requires retraining the model, preventing rapid online updates. This limitation leads to discrepancies between predictions and actual progress, which is particularly pronounced in at-home settings requiring long-term monitoring. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method and system for detecting and analyzing neurological diseases. Through a data processing mechanism that combines short-term intensive acquisition with long-term intermittent monitoring, it realizes long-term trend prediction and dynamic correction based on initial data, solving the problem of being unable to balance monitoring convenience and prediction accuracy.

[0006] To achieve the above objectives, in a first aspect, the present application provides a method for detecting and analyzing neurological diseases, comprising: Obtain the user's wearable device monitoring data and behavioral test data. Wearable device monitoring data includes wristband acquisition signals, blood pressure monitor acquisition signals and motion behavior images. Behavioral test data includes standardized movement test results. The wristband collected signal is decomposed into multiple frequency bands, and the tremor component, gait component and autonomic nerve fluctuation component are separated by wavelet transform; The tremor component is extracted by long short-term memory network to extract tremor intensity features, the gait component is extracted by time series convolution network to extract gait symmetry features, and the autonomic nerve fluctuation component is extracted by Gaussian process regression algorithm to extract autonomic nerve rhythm features; A deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and the motor function score is calculated based on the results of standardized movement tests. Heart rate variability analysis was performed on the signals collected by the blood pressure monitor and the wristband, the circadian rhythm fluctuation characteristics were calculated, and abnormal neurological function patterns were identified using a density clustering algorithm; Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores and neurological dysfunction patterns, a neurological function state map is constructed, and the contribution weight of each feature to disease risk is calculated through the graph attention mechanism; Obtain the user's historical monitoring data, use neural differential equations to learn the continuous time evolution law of the neural function state based on the historical test data in the neural function state map, predict the disease progression trend in the preset future time period, and generate prediction results; Collect at least one of the wristband signal, the blood pressure monitor signal, and the motion behavior image captured by the mobile phone camera within a preset future time period, record it as feedback correction information, and correct the prediction result based on the adaptive Kalman filter according to the feedback correction information; Output a neurological disease risk assessment report based on the prediction results.

[0007] In some embodiments, a neurological function state map is constructed based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores, and neurological dysfunction patterns, and the contribution weight of each feature to disease risk is calculated through a graph attention mechanism, including: Tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores, and neurological dysfunction patterns are used as node features of the graph neural network, and initial edge connection relationships are constructed based on the physiological correlation between each node feature; A graph attention mechanism was used to model the spatial dependencies of node features. The motor function association weights between tremor intensity features and gait symmetry features, the autonomic regulation weights between autonomic rhythm features and neurological dysfunction patterns, and the comprehensive coordination weights between motor function scores and other node features were calculated. Through a multi-layer graph convolutional network, the neighborhood node information is aggregated, the embedded representation of each node feature is iteratively optimized, and a graph embedding vector reflecting the overall state of neural function is output; Based on the graph embedding vector, the attention score calculation module generates the dynamic contribution weight of each node feature to the disease risk. The weights of tremor intensity features and gait symmetry features are used for motor dysfunction assessment, the weights of autonomic rhythm features and abnormal neurological function patterns are used for autonomic nervous function assessment, and the weights of motor function scores are used for comprehensive behavioral function assessment.

[0008] In some embodiments, historical monitoring data of a user is obtained, and a neural differential equation is used in a neural function state map based on the historical monitoring data to learn the continuous time evolution law of the neural function state, predict the disease progression trend in a preset future time period, and generate a prediction result, including: Integrate historical monitoring data into the neural function state map to form multiple historical embedding vectors, and arrange the multiple historical embedding vectors in time series to construct the time evolution sequence of the neural function state; The neural differential equation is used to calculate the time evolution sequence, and the gradual change process of the neural function state is described by a learnable continuous time dynamic function, which is expressed by formula (1). Formula (1) is as follows: ; In formula (1), is the neural differential equation, is a learnable continuous-time dynamics function, for The neural function state vector at the moment, for The weight parameter, is a continuous time variable; A gating mechanism is introduced into the neural differential equation framework to dynamically adjust the state change rate at different time scales to capture the short-term fluctuations and long-term degradation characteristics of neural functions. The neural differential equation model is obtained, which is expressed by formula (2). Formula (2) is as follows: ; In formula (2), is the gating vector corresponding to the gating mechanism, is the residual term, is element-wise multiplication; Based on the current neural function state map and the trained neural differential equation model, multi-step state prediction within a preset future period is performed, and the predicted neural function state trajectory including time continuity is output; Probabilistically process the predicted trajectory of neurological function status, calculate the probability distribution of disease risk at each time point, and identify the key time nodes where the risk exceeds the preset threshold; The predicted trajectory of neurological function status is compared and analyzed with the user's individual baseline data to generate personalized prediction results of disease progression trends. The prediction results include risk level change curves and warning information at key time nodes.

[0009] In some embodiments, at least one of a wristband signal, a blood pressure monitor signal, and a motion behavior image captured by a mobile phone camera is collected within a preset future time period and recorded as feedback correction information. The prediction result is corrected based on the adaptive Kalman filter according to the feedback correction information, including: Inputting the feedback correction information into the feature extraction module, and updating at least one of the tremor intensity feature, gait symmetry feature, autonomic nerve rhythm feature, and motor function score according to the extracted feedback features to generate a neural function state observation vector at the current moment; Construct a state space model, use the predicted trajectory output by the neural differential equation as the state transition prior, and the neural function state observation vector as the measurement value to generate the state estimation result; Adaptive Kalman filter algorithm is used to generate confidence weights of state estimation results by dynamically adjusting process noise covariance matrix and observation noise covariance matrix; And, when the deviation between the observed and predicted values ​​exceeds a dynamic threshold, a recalibration process of the prediction model is triggered; Modify the state estimation results according to the confidence weight; The corrected state estimation results are fed back to the neural differential equation model, the parameters of the dynamic function are adjusted to adapt to the user's individualized neural function change pattern, and the corrected prediction results are output.

[0010] In some embodiments, the method further comprises: Determine whether the parameters in the feedback correction information exceed their corresponding normal fluctuation range. If so, trigger the re-collection mode; Alternatively, the deviation between the revised prediction result and the initial prediction result is calculated, and when the deviation exceeds a preset deviation threshold, a re-collection mode is triggered; The recollection mode is configured as: Re-collect the user's wearable device monitoring data and behavioral test data; The wristband collected signal was decomposed into multiple frequency bands again, and the tremor component, gait component and autonomic nerve fluctuation component were separated using wavelet transform. The tremor component was re-extracted with a long short-term memory network to extract tremor intensity features, the gait component was extracted with a temporal convolutional network to extract gait symmetry features, and the autonomic nerve fluctuation component was extracted with a Gaussian process regression algorithm to extract autonomic nerve rhythm features; A deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and the motor function score is calculated based on the results of standardized movement tests. Re-analyze the heart rate variability of the blood pressure monitor and wristband signals, calculate the circadian rhythm fluctuation characteristics, and identify abnormal neurological function patterns using a density clustering algorithm; Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores and neurological dysfunction patterns, a neurological function state map was constructed, and the contribution weight of each feature to disease risk was calculated through the graph attention mechanism; In the neural function state map, the neural differential equation is re-adopted based on the historical detection data to learn the continuous time evolution law of the neural function state, predict the disease progression trend in the reset preset future time period, and generate updated prediction results.

[0011] In some embodiments, the wristband collected signals include acceleration signals and gyroscope signals. The wristband collected signals are decomposed into multiple frequency bands and wavelet transform is used to separate the tremor component, gait component, and autonomic nerve fluctuation component, including: The acceleration and gyroscope signals are decomposed at multiple scales using wavelet basis functions with time-frequency localization characteristics. The original signals are separated into a first-band signal containing tremor characteristics, a second-band signal containing gait cycle characteristics, and a third-band signal reflecting autonomic nervous system regulation according to preset frequency bands. An adaptive decomposition algorithm based on signal energy distribution is adopted for the first frequency band signal, the second frequency band signal, and the third frequency band signal, and the number of wavelet decomposition layers is dynamically adjusted according to the signal-to-noise ratio of the signal of each frequency band to obtain the decomposed first sub-band signal, the second sub-band signal, and the third sub-band signal, and motion artifact suppression processing is performed on each decomposed sub-band signal; The first sub-band signal after the motion artifact suppression process is merged into a time-domain continuous tremor component through a phase consistency reconstruction algorithm; The second sub-band signal that has been processed for motion artifact suppression is subjected to motion cycle phase calibration using a dynamic time warping algorithm to obtain a gait component. The third sub-band signal, which had undergone motion artifact suppression processing, was subjected to ensemble empirical mode decomposition to remove baseline drift, maximum likelihood estimation to restore physiological rhythm components, and 0.01-0.1 Hz band-pass filtering to extract sympathetic-parasympathetic nerve regulation characteristics, thereby obtaining the autonomic nerve fluctuation component.

[0012] In some embodiments, extracting tremor intensity features from the tremor component using a long short-term memory network includes: Construct a bidirectional long short-term memory network model with attention mechanism; Conduct time series analysis on the tremor component to obtain the amplitude change rate and time series fluctuation characteristics of the tremor component; A tremor intensity feature representing the severity of tremor is generated based on the amplitude change rate and time series fluctuation characteristics; The gait components are extracted through the temporal convolutional network to extract gait symmetry features, including: The gait components are input into a multi-scale dilated temporal convolutional network, and the gait cycle features of different time scales are extracted through multiple dilated convolutional layers set in parallel; The cross-correlation function of the left and right limb motion signals in the time domain and frequency domain is calculated according to the gait cycle characteristics; Hilbert transform is used to obtain the phase difference and output the symmetric feature vector representing the gait asymmetry; The autonomic nerve fluctuation component is extracted through the Gaussian process regression algorithm, including: The autonomic nerve fluctuation component is used as input and its nonlinear change trend is fitted by Gaussian process regression model; and, the squared exponential kernel function was used to model the autocorrelation characteristics of the autonomic nerve fluctuation components; The output includes the low-frequency oscillation power ratio, the high-frequency oscillation power ratio, and the autonomic rhythm characteristics of the ratio of the two.

[0013] In some embodiments, a deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and combined with standardized movement test results to calculate a movement function score, including: The spatiotemporal features of the motion behavior images are encoded through a three-dimensional convolutional neural network, the spatiotemporal sequence features of the eye movements are extracted through the spatiotemporal convolution layer, and the eye movement trajectory feature vector is output; Use graph convolutional networks to build a topological graph model of limb joint motion data, and generate a motion coordination feature matrix based on the spatial connection relationship of the joints; The eye movement trajectory feature vector is fused with the movement coordination feature matrix, and the time-frequency features of the electromyographic signals collected in the standardized movement test are connected. The cross-modal features are weighted through a multi-layer perceptron to output the motor function score.

[0014] In some embodiments, heart rate variability analysis is performed on the signals collected by the blood pressure monitor and the wristband, circadian rhythm fluctuation characteristics are calculated, and abnormal neurological function patterns are identified using a density clustering algorithm, including: Perform dynamic blood pressure trend extraction on the sphygmomanometer collected signal to obtain the systolic / diastolic blood pressure periodic change curve; The RR interval sequence analysis of the wristband collected signal was performed, and the day and night ratio of low-frequency power to high-frequency power was extracted based on frequency domain transformation; The periodic variation curve and the day-night ratio were input into a density clustering model based on DBSCAN. Through adaptive neighborhood radius parameter optimization, abnormal clusters representing sympathetic-parasympathetic imbalance were divided in the feature space of the density clustering model. The data points corresponding to the periodic change curve and the day-night ratio that fall into the abnormal cluster are marked as abnormal neurological function patterns and the abnormal probability is output.

[0015] In the second aspect, the present invention also provides a neurological disease detection and analysis system, which is applicable to the method described in the first aspect, and the system includes: a wearable device, a behavior collection device, a signal processing unit, and a disease analysis unit. The wearable device includes at least one of a bracelet, glasses, a helmet, and headphones; the behavior collection device includes at least one of a mobile phone, a tablet, and a camera; the signal processing unit is communicatively connected to the wearable device and the behavior collection device, and the signal processing unit is used to obtain the user's wearable device monitoring data and behavior test data; the bracelet collected signal is decomposed into multiple frequency bands, and the tremor component, gait component and autonomic nerve fluctuation component are separated by wavelet transform; the tremor intensity feature of the tremor component is extracted by the long short-term memory network, the gait symmetry feature of the gait component is extracted by the time series convolution network, and the autonomic nerve rhythm feature of the autonomic nerve fluctuation component is extracted by the Gaussian process regression algorithm; the deep learning model is used to extract the eye movement trajectory feature and the movement coordination feature of the movement behavior image, and combined with the standardized movement test The motor function score is calculated based on the test results; the heart rate variability analysis is performed on the signals collected by the blood pressure monitor and the bracelet, the circadian rhythm fluctuation characteristics are calculated, and the abnormal neurological function pattern is identified through the density clustering algorithm; the disease analysis unit is used to construct a neurological function state map based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function score and neurological function abnormality pattern, and calculate the contribution weight of each feature to the disease risk through the graph attention mechanism; the user's historical monitoring data is obtained, and the continuous time evolution law of the neurological function state is learned by the neural differential equation based on the historical detection data in the neurological function state map, and the disease progression trend in the preset future time period is predicted to generate a prediction result; in the preset future time period, at least one of the signals collected by the bracelet, the signals collected by the blood pressure monitor and the movement behavior images captured by the mobile phone camera are collected and recorded as feedback correction information, and the prediction result is corrected based on the adaptive Kalman filter according to the feedback correction information; a neurological disease risk assessment report is output according to the prediction result.

[0016] Different from the existing technology, the above technical solution obtains the wristband collection signal, the blood pressure monitor collection signal, the movement behavior image and the behavior test data, and extracts the tremor intensity characteristics, gait symmetry characteristics and autonomic nerve rhythm characteristics of the wristband collection signal through multi-band decomposition; analyzes the movement behavior image and the standardized movement test to obtain the motor function score; performs heart rate variability analysis to identify the abnormal neurological function pattern; constructs the neurological function state map and calculates the feature weight; combines the historical monitoring data to predict the disease progression trend; and dynamically adjusts the prediction result through subsequent feedback correction information. The present invention realizes long-term trend prediction and dynamic correction based on initial data by combining short-term intensive monitoring with long-term intermittent collection, which significantly improves the reliability and practicality of neurological disease risk assessment in home scenarios.

[0017] The above-mentioned records related to the content of the invention are only an overview of the technical solution of this application. In order to enable ordinary technicians in this field to understand the technical solution of this application more clearly, and then implement it according to the text of the specification and the contents recorded in the drawings, and to make the above-mentioned purposes and other purposes, features and advantages of this application easier to understand, the following is an explanation in combination with the specific implementation methods and drawings of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings are only used to illustrate the principles, implementation methods, applications, features and effects of the specific embodiments of the present invention and other related contents, and are not to be considered as limiting the present application.

[0019] In the drawings of the specification: Figure 1 A method step diagram showing steps S101 to S109 of the method described in the specific embodiment; Figure 2 This is a method step diagram for steps S201 to S204 of the method described in the specific implementation manner. DETAILED DESCRIPTION

[0020] In order to explain in detail the possible application scenarios, technical principles, specific solutions that can be implemented, and the purpose and effects of this application, the following is a detailed description of the specific embodiments listed in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of this application and are therefore only examples and are not intended to limit the scope of protection of this application.

[0021] References to "embodiments" herein mean that the specific features, structures, or characteristics described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the word "embodiment" in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or relevance to other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the various technical features mentioned in the embodiments can be combined in any manner to form a corresponding implementable technical solution.

[0022] Unless otherwise defined, the technical terms used herein have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms herein is only for describing specific embodiments and is not intended to limit this application.

[0023] In the description of this application, the term "and / or" is used to describe a logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and both A and B exist. In addition, the character " / " in this document generally indicates that the objects before and after are in a logical "or" relationship.

[0024] In this application, terms such as "first" and "second" are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, priority or sequence relationship between these entities or operations.

[0025] Without further limitations, in this application, the words "include", "comprise", "have" or other similar open-ended expressions used in sentences are intended to cover non-exclusive inclusion. These expressions do not exclude the presence of additional elements in the process, method or product that includes the elements, so that the process, method or product that includes a series of elements may include not only those defined elements, but also other elements that are not explicitly listed, or also include elements inherent to such process, method or product.

[0026] Consistent with the understanding in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceed" are understood to exclude the number itself; expressions such as "above," "below," and "within" are understood to include the number itself. Furthermore, in the description of the embodiments of this application, "multiple" means two or more (including two), and similar expressions related to "multiple," such as "multiple groups" and "multiple times," are also understood in this manner, unless otherwise specifically defined.

[0027] In the description of the embodiments of the present application, the space-related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or position relationship based on the orientation or position relationship shown in the specific embodiments or drawings, and are only for the convenience of describing the specific embodiments of the present application or facilitating the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, it should not be understood as a limitation on the embodiments of the present application.

[0028] The processor described in the embodiments of the present application can be implemented by hardware, firmware, software or a combination thereof, and can use a circuit, a single or multiple application-specific integrated circuits (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or at least one of a microprocessor, and also includes other physical, biological or chemical structures that can achieve similar or equivalent functions to the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of the present application, or any combination of the steps mentioned therein.

[0029] The computer program involved in the embodiment can be stored in a computer device readable storage medium, which includes but is not limited to a disk, a tape, a magnetic card, a floppy disk, a flash memory, an optical disc, an optical card, a read-only memory (ROM), a random access memory (RAM), an erasable programmable ROM (EPROM) and an electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical or chemical structures that can achieve similar or equivalent functions to the storage media listed above, such as DNA, RNA, protein and other units with information storage capabilities. In a specific embodiment, the storage medium involved can be one of the above-mentioned media types or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiment can be stored in a single medium in a centralized manner or in a distributed manner on multiple media. The memory containing the computer device readable storage medium can be a non-volatile memory or a random access memory. These computer device readable storage media can be built into the device or connected to the device involved in the embodiment as an external device or part of an external device. In some embodiments, a memory having a computer-readable storage medium is deployed locally; in other embodiments, a solution of deploying the memory away from the processor may also be adopted, such as a network-attached memory accessed via an RF circuit or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), etc., or a suitable combination thereof, as long as the computer device can access the memory. In addition, the computer program involved in the embodiment can be stored in plaintext / ciphertext form, or can be designed as training data, which can be integrated and reorganized through model training and implicitly stored in the parameter state of a deep neural network or other machine learning model.

[0030] See also Figure 1 In a first aspect, this embodiment provides a method for detecting and analyzing a neurological disease, comprising: S101. Obtaining wearable device monitoring data and behavioral test data of the user, where the wearable device monitoring data includes signals collected by a wristband, signals collected by a blood pressure monitor, and motion behavior images, and the behavioral test data includes standardized movement test results; S102, performing multi-band decomposition on the wristband collected signal, and using wavelet transform to separate the tremor component, gait component, and autonomic nerve fluctuation component; S103: Extracting tremor intensity features from the tremor component using a long short-term memory network, extracting gait symmetry features from the gait component using a temporal convolutional network, and extracting autonomic rhythm features from the autonomic nerve fluctuation component using a Gaussian process regression algorithm; S104. Using a deep learning model to extract eye movement trajectory features and movement coordination features from the movement behavior images, and calculating the movement function score based on the standardized movement test results; S105, performing heart rate variability analysis on the signals collected by the blood pressure monitor and the wristband, calculating circadian rhythm fluctuation characteristics, and identifying abnormal neurological function patterns using a density clustering algorithm; S106. Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores, and neurological dysfunction patterns, a neurological function state map is constructed, and the contribution weight of each feature to disease risk is calculated using a graph attention mechanism. S107. Obtain the user's historical monitoring data, use a neural differential equation to learn the continuous time evolution law of the neural function state in the neural function state map based on the historical detection data, predict the disease progression trend in a preset future time period, and generate a prediction result; S108. Collect at least one of the wristband signal, the blood pressure monitor signal, and the motion behavior image captured by the mobile phone camera within a preset future time period, record it as feedback correction information, and correct the prediction result based on the adaptive Kalman filter according to the feedback correction information; S109. Output a neurological disease risk assessment report based on the prediction results.

[0031] In step S101, wearable device monitoring data is collected in real time through a wearable device. Preferably, the wristband collected signals include acceleration signals and gyroscope signals, which are used to capture limb movement characteristics. The blood pressure monitor collected signals can be obtained through a photoelectric sensor to reflect the cardiovascular regulation function. The movement behavior image is recorded by the mobile phone camera and contains the temporal and spatial information of facial expressions and limb movements. The standardized movement test results are obtained through a preset clinical assessment procedure, such as a finger tapping test or a balance test, which is used to quantify the degree of motor dysfunction. These multi-source data together constitute the basic data set for neurological function assessment.

[0032] In step S102, the wristband's collected signal is decomposed into multiple frequency bands using wavelet transform. The tremor component corresponds to a specific frequency band, characterizing involuntary movements; the gait component corresponds to another specific frequency band, reflecting the characteristics of the walking cycle; and the autonomic nerve fluctuation component, located in a specific low-frequency band, reflects the regulatory functions of the sympathetic and parasympathetic nerves. During the signal decomposition process, the decomposition parameters are dynamically adjusted based on the actual signal quality, ensuring that each component retains valid physiological information while suppressing motion artifacts.

[0033] In step S103, a long-short-term memory network is used to extract tremor intensity features. Tremor intensity features include amplitude variation parameters and duration parameters. The amplitude variation parameter characterizes the intensity of tremor signal fluctuations and reflects the degree of motor control abnormality; the duration parameter records the duration of the tremor attack and is used to assess symptom severity. A temporal convolutional network extracts gait symmetry features by analyzing the correlation between left and right limb signals. Autonomic rhythm features are extracted using a Gaussian process regression algorithm and include quantitative indicators such as the ratio of low-frequency power to high-frequency power. Tremor intensity features reflect abnormal patterns of motor control function, gait symmetry features reflect abnormal patterns of coordination function, and autonomic rhythm features reflect abnormal patterns of autonomic regulation function.

[0034] In step S104, the motion behavior image data is processed using a deep learning model. Eye movement trajectory features include parameters for saccade velocity and gaze stability, and movement coordination features are derived by analyzing the motion relationships between limb joints. Standardized motion test results and image features are then combined through feature fusion to produce a motor function score that comprehensively assesses the functional status of the motor system.

[0035] In step S105, heart rate variability analysis preferably processes the RR interval series data to calculate circadian rhythm fluctuation characteristics. A density clustering algorithm is used to identify abnormal neurological function patterns, and its parameters are adaptively adjusted based on the actual data distribution. This step focuses on capturing abnormal circadian regulation characteristics of autonomic nervous system function.

[0036] In step S106, the graph attention network adopts a three-layer structure. The first layer processes motion-related features, the second layer analyzes autonomic nervous regulation features, and the third layer fuses the two types of features through a cross-domain attention mechanism to quantify the contribution of different feature dimensions to disease risk. For example, in the exemplary implementation, the weight of the tremor intensity feature is generally higher than that of the gait symmetry feature.

[0037] In step S107, the neural differential equation modeling preferably adopts the continuous time series analysis method, takes the historical monitoring data as the initial state, learns the evolution law of the neural function state through the reversible neural network, and predicts the functional change trend within the same time range in the future based on the monitoring data within the past period of time. The prediction results include the change curves of each characteristic dimension and the comprehensive risk assessment value.

[0038] In step S108, the feedback correction mechanism is implemented using an adaptive Kalman filter. The correction process first calculates the deviation between the predicted result and the newly observed data, and then dynamically adjusts the state estimation parameters. Preferably, incremental learning of the neural function state map is automatically triggered when the correction amplitude exceeds a preset threshold three times in a row.

[0039] The neurological disease risk assessment report output in step S109 contains clinically required information such as current status assessment, trend prediction, and intervention recommendations.

[0040] This embodiment constructs a dynamically evolving neurological function assessment system by integrating wearable device monitoring data with behavioral test data. Tremor intensity features, gait symmetry features, and autonomic rhythm features are obtained from wristband collected signals through multi-band decomposition and feature extraction technology. At the same time, the eye movement trajectory features and movement coordination features obtained from motor behavior image analysis are combined to form a multi-dimensional motor function score. Further, abnormal neurological function patterns are identified through heart rate variability analysis and density clustering, and a graph attention mechanism is used to construct a neurological function state map reflecting the weights of each feature. Based on historical monitoring data, neural differential equation modeling is used to predict disease progression trends, and adaptive Kalman filtering is used to dynamically correct the prediction results. This embodiment achieves an organic combination of short-term comprehensive data collection and long-term local monitoring, which not only ensures the accuracy of the initial assessment, but also maintains long-term prediction accuracy through continuous feedback correction. The final output of the neurological disease risk assessment report provides a reliable basis for clinical decision-making, while relying on common equipment to achieve a convenient detection method that can be used at home.

[0041] See also Figure 2 In some embodiments, a neurological function state map is constructed based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores, and neurological dysfunction patterns. The contribution weight of each feature to disease risk is calculated through a graph attention mechanism, including: S201. Using tremor intensity features, gait symmetry features, autonomic nerve rhythm features, motor function scores, and neurological dysfunction patterns as node features of a graph neural network, and constructing initial edge connection relationships based on the physiological correlations between the node features; S202. Use the graph attention mechanism to model the spatial dependency of node features, calculate the motor function association weight between tremor intensity features and gait symmetry features, the autonomic nerve regulation weight between autonomic nerve rhythm features and neurological dysfunction patterns, and the comprehensive coordination weight between motor function scores and other node features; S203, aggregating neighborhood node information through a multi-layer graph convolutional network, iteratively optimizing the embedding representation of each node feature, and outputting a graph embedding vector reflecting the overall state of neural function; S204. Based on the graph embedding vector, the attention score calculation module generates the dynamic contribution weight of each node feature to the disease risk. The weights of the tremor intensity feature and the gait symmetry feature are used for motor dysfunction assessment. The weights of the autonomic rhythm feature and the abnormal neurological function pattern are used for autonomic nervous function assessment. The weight of the motor function score is used for comprehensive behavioral function assessment.

[0042] In step S201, node features are constructed based on the actual relationships within physiological systems. Tremor intensity features are connected to gait symmetry features via the motor control neural network, autonomic rhythm features are associated with patterns of neurological dysfunction via brainstem regulatory pathways, and motor function scores, as a comprehensive indicator, are bidirectionally connected to each node feature. Initial edge connection weights are assigned using a semi-supervised approach based on medical prior knowledge. For example, the initial weights for connections between motor-related features are set higher than those for cross-system connections to ensure that the graph structure is clinically interpretable from the initial stages of training.

[0043] In step S202, spatial dependency modeling is implemented using a multi-head graph attention mechanism, with each attention head focusing on a specific type of feature interaction. A temporal alignment module is incorporated into the calculation of motor function association weights to ensure that the interaction between tremor intensity and gait symmetry within the same physiological cycle is accurately captured. Autonomic neural control weights are enhanced through frequency-domain coherence analysis, focusing on extracting phase synchronization features of low-frequency signals. A cross-modal attention mechanism is employed for the comprehensive coordination weights, enabling the motor function score to dynamically adjust its influence on different node features.

[0044] In step S203, each layer of the multi-layer graph convolutional network undergoes two phases: feature space projection and neighborhood information aggregation. Feature space projection is achieved through a learnable affine transformation, mapping node features of varying dimensions into a unified embedding space. Neighborhood information aggregation utilizes a gating mechanism to dynamically control the intensity of information propagation based on edge weights. During the iterative optimization process, the node embedding representation gradually integrates local details and global topological features, ultimately generating a graph embedding vector that encompasses a multi-scale representation, from microscopic feature fluctuations to macroscopic functional states.

[0045] In step S204, preferably, a differentiable weight constraint mechanism is introduced into the dynamic contribution weight calculation to ensure that the weight distribution of the three dimensions of motor dysfunction assessment, autonomic nervous function assessment, and comprehensive behavioral function assessment conforms to the cognitive logic of clinical medicine. In specific implementation, motor dysfunction assessment focuses on the abnormal degree of feature amplitude, autonomic nervous function assessment focuses on the duration of rhythm disorder, and comprehensive behavioral function assessment integrates the coordination index of spatiotemporal features. Furthermore, the output weight can be calibrated by clinical expert knowledge to ensure its consistency with existing medical evaluation standards.

[0046] This embodiment constructs a neurological function state graph, using multiple physiological indicators such as tremor intensity characteristics and gait symmetry characteristics as graph nodes. Initial edge connections are established based on physiological correlations, and a graph attention mechanism is used to calculate feature interactions such as motor function association weights and autonomic nervous system regulation weights. Ultimately, dynamic contribution weights are output for multidimensional assessment. This embodiment combines medical prior knowledge with graph neural networks, capturing nonlinear correlations between features through spatial dependency modeling. This allows short-term multivariate data to reflect the overall state of neurological function, enabling long-term disease risk prediction based on fragmented monitoring data in a home setting. It also supports subsequent dynamic corrections through iteratively optimized graph embedding vectors, ensuring clinical interpretability while improving the comprehensiveness and timeliness of the assessment.

[0047] In some embodiments, historical monitoring data of a user is obtained, and a neural differential equation is used in a neural function state map based on the historical monitoring data to learn the continuous time evolution law of the neural function state, predict the disease progression trend in a preset future time period, and generate a prediction result, including: Integrate historical monitoring data into the neural function state map to form multiple historical embedding vectors, and arrange the multiple historical embedding vectors in time series to construct the time evolution sequence of the neural function state; The neural differential equation is used to calculate the time evolution sequence, and the gradual change process of the neural function state is described by a learnable continuous time dynamic function, which is expressed by formula (1). Formula (1) is as follows: ; In formula (1), is the neural differential equation, is a learnable continuous-time dynamics function, for The neural function state vector at the moment, for The weight parameter, As a continuous-time variable, a learnable continuous-time dynamics function is defined in the neural differential equation framework and expressed by formula (3), which is as follows: ; in, is the sigmoid activation function, is the weight parameter The first weight matrix of is the weight parameter The second weight matrix, is the weight parameter The third weight matrix, is the nonlinear activation function in the neural differential equation, is the first bias parameter, is the second bias parameter, is the third bias parameter, is the hyperbolic tangent function; A gating mechanism is introduced into the neural differential equation framework to dynamically adjust the state change rate at different time scales to capture the short-term fluctuations and long-term degradation characteristics of neural functions. The neural differential equation model is obtained, which is expressed by formula (2). Formula (2) is as follows: ; In formula (2), is the gating vector corresponding to the gating mechanism, with a value range of [0,1], generated by the sigmoid function, which controls the state update ratio. is the residual term, capturing the long-term degradation trend, is element-wise multiplication; Based on the current neural function state map and the trained neural differential equation model, multi-step state prediction within a preset future period is performed, and the predicted neural function state trajectory including time continuity is output; Probabilistically process the predicted trajectory of neurological function status, calculate the probability distribution of disease risk at each time point, and identify the key time nodes where the risk exceeds the preset threshold; The predicted trajectory of neurological function status is compared and analyzed with the user's individual baseline data to generate personalized prediction results of disease progression trends. The prediction results include risk level change curves and warning information at key time nodes.

[0048] In this embodiment, historical monitoring data is integrated into the neural function state map through the time alignment module. The formed historical embedding vector retains the spatial topological relationship of the original features, ensuring that the time evolution sequence contains both temporal information and the map structure characteristics. The neural differential equation framework uses an adaptive step-size numerical solution method (such as the dopri5 algorithm) to handle continuous-time modeling, in which the learnable continuous-time dynamic function is constructed through a multi-layer perceptron, and its weight parameters are During the training process, gradient optimization is performed through the adjoint sensitivity method, enabling the system to automatically learn multi-scale dynamic laws ranging from minute-level tremor fluctuations to monthly-level functional degradation.

[0049] The gating mechanism is implemented using physiologically meaningful time scales, preferably short-term fluctuation gating Focus on state changes within 0-6 hours, extract transient features through LSTM network; long-term degradation gating The correlation is a gradual decline in function in weeks, and the residual term A linear transformation with an attenuation factor is used to simulate irreversible damage to neural function. This dual-time-scale modeling approach is particularly important in predicting degenerative diseases such as Parkinson's disease, as it can distinguish temporary symptom fluctuations from substantial functional deterioration.

[0050] Preferably, the gate vector The generation of is adaptively calibrated by the time-frequency analysis results of historical monitoring data. The division of short-term fluctuation time scale (0-6 hours) is determined according to the power spectrum density peak of the tremor signal during the user's daily activity cycle; the residual term The attenuation factor is dynamically adjusted according to the slope of the historical degeneration trend to ensure that the long-term degeneration modeling matches the rate of decline in neurological function scores recorded during clinical follow-up.

[0051] Preferably, the probabilistic processing stage uses Monte Carlo dropout simulation to predict uncertainty. The calculation of the disease risk probability distribution integrates three key factors: the Euclidean distance between the predicted neurological function trajectory and a preset threshold, the confidence interval of the historical prediction error, and the dynamic offset of the individualized baseline data. Critical time nodes can be identified using a change point detection algorithm. When the first-order derivative of the disease risk probability exceeds a preset threshold, an alert is triggered. Preferably, the preset threshold is set as a safety margin of no more than 5% risk increase per week based on the consensus of specialists.

[0052] A transfer learning strategy is employed to generate personalized predictions, using individual user baseline data as source domain features. The distribution matching of the predicted neurological status trajectory is adjusted using a maximum mean difference loss function. The risk level change curve utilizes dynamic color coding, mapping the predicted values ​​of motor dysfunction assessment, autonomic nervous system function assessment, and comprehensive behavioral function assessment into a unified risk index. Warning information for critical time nodes is integrated with multimodal prediction results using a Bayesian network, ultimately producing a structured report that meets the requirements of a clinical decision support system.

[0053] This embodiment models the continuous-time evolution of neural functional states through neural differential equations. The time evolution sequence constructed from historical monitoring data is input into a dynamic system with a gating mechanism. A learnable continuous-time dynamic function is used to simultaneously capture short-term fluctuations and long-term degradation characteristics, ultimately outputting a time-continuous prediction trajectory. This embodiment incorporates the time-scale division of physiological constraints into the design of the gating mechanism. Through the synergistic effect of short-term fluctuation gating and long-term degradation gating, prediction results based on fragmented monitoring data can both reflect immediate symptom changes and track disease progression trends. Combining probabilistic processing with individual baseline comparisons, this achieves cross-scale disease risk assessment from short-term observations to long-term predictions. Furthermore, a dynamically updated neural functional state map supports continuous revision of prediction results, significantly improving the timeliness and personalization of predictions while ensuring clinical applicability.

[0054] In some embodiments, at least one of a wristband signal, a blood pressure monitor signal, and a motion behavior image captured by a mobile phone camera is collected within a preset future time period and recorded as feedback correction information. The prediction result is corrected based on the adaptive Kalman filter according to the feedback correction information, including: Inputting the feedback correction information into the feature extraction module, and updating at least one of the tremor intensity feature, gait symmetry feature, autonomic nerve rhythm feature, and motor function score according to the extracted feedback features to generate a neural function state observation vector at the current moment; Construct a state space model, take the predicted trajectory output by the neural differential equation as the state transfer prior, take the neural function state observation vector as the measurement value, generate the state estimation result, and the constructed state space model is expressed by formula (4), which is as follows: ; Defining Observations The observation matrix The predicted value is mapped and expressed as follows: , when the observation matrix When is the unit matrix, the observation equation is simplified to , in formula (4), For the The neural function state observation vector at the moment, that is, the measurement value, The predicted trajectory output by the neural differential equation is The neural function state vector at the moment, that is, the predicted value, is the state transfer matrix, obtained by discretizing the continuous-time differential equation, is the observation noise, is the process noise; Adaptive Kalman filter algorithm is used to generate confidence weights of state estimation results by dynamically adjusting process noise covariance matrix and observation noise covariance matrix; And, when the deviation between the observed value and the predicted value exceeds the dynamic threshold, the recalibration process of the prediction model is triggered, based on the predicted value With measured value The noise parameters are dynamically adjusted by the difference between , process noise reflects model uncertainty, observation noise reflects data reliability, and the process noise covariance matrix is ​​expressed by formula (5): ; in, is the process noise covariance matrix, is the model uncertainty coefficient, is the baseline noise level, is the weight parameter of the continuous time dynamics function in formula (3) The Jacobian matrix of is the identity matrix; The observation noise covariance matrix is ​​expressed by formula (6): ; in, is the observation noise covariance matrix, is the scaling factor of the dynamic threshold, is the observed value, is the standard deviation of the normal fluctuation range, is a diagonal matrix; Modify the state estimation results according to the confidence weight; The corrected state estimation results are fed back to the neural differential equation model, the parameters of the dynamic function are adjusted to adapt to the user's individualized neural function change pattern, and the corrected prediction results are output.

[0055] In this embodiment, the feature extraction module for feedback correction information uses the same processing flow as historical monitoring data to ensure consistency in the feature space. Tremor intensity feature updates are implemented through a long short-term memory network, preserving the temporal dependency of the original signal. Gait symmetry feature updates introduce a phase calibration module to eliminate temporal misalignment caused by intermittent acquisition. Autonomic rhythm feature updates maintain the Gaussian process regression framework and add sliding window smoothing to accommodate dynamic adjustment needs. Motor function score updates use a weighted fusion strategy, where the weights of new and historical data are automatically adjusted based on acquisition duration to ensure that data at different monitoring frequencies can be effectively integrated.

[0056] In the construction of the state space model, the observation matrix The dimension is determined by the characteristic dimension of the neural function state vector. It is initially set to the unit matrix to reflect the directly observable characteristics of the features of each dimension. When some data are missing, it is automatically converted to a mask matrix. Preferably, the state transfer matrix is ​​obtained by discretization of the neural differential equation, and the fixed-step Runge-Kutta method is used to realize the conversion from the continuous time domain to the discrete time domain. The time step is adaptively determined based on the average acquisition interval of the user's historical monitoring data. The covariance matrix of the observation noise and the process noise is dynamically adjusted through an adaptive mechanism, wherein the process noise covariance matrix is ​​constructed through the weight parameter The flattening process ensures that the Jacobian matrix matches the identity matrix dimension, and the exponential term of the observation noise covariance matrix is ​​designed to operate element-by-element and perform element-wise multiplication with the diagonal matrix.

[0057] Preferably, the value of the dynamic threshold is obtained by calculating the rolling standard deviation of the neural function state vector in the user's historical data. When the prediction deviation exceeds three times the standard deviation, the recalibration process is triggered; during the calibration process, the transfer learning strategy is preferably used to adjust the dynamic function parameters and weight parameters. The gradient information of is obtained in real time through automatic differentiation technology, ensuring strict synchronization with the noise estimation of the adaptive filtering process. The corrected state estimate is rebalanced by the graph attention network to maintain compatibility with the overall topological structure of the neural function state map. Its corresponding continuous-time variable Establish clear mapping relationships.

[0058] Model uncertainty coefficients in the process noise covariance matrix The baseline noise level is calculated by inversely proportional calculation of the root mean square error of the neural differential equation training set. Take the median of the user's historical state vector fluctuation amplitude; observation noise scaling factor Linked with the dynamic threshold, it automatically increases when the number of consecutive corrections exceeds the preset value to reduce the weight of abnormal observations.

[0059] This embodiment realizes the dynamic optimization of neurological function degeneration prediction through the real-time feedback correction mechanism of multi-source physiological signals. The feature extraction module maintains the consistency of the feature space of historical data and real-time signals. The state space model uses adaptive Kalman filtering to dynamically balance the credibility of the predicted trajectory and the observed data. The intelligent switching mechanism of the observation matrix effectively handles the data missing situation, and the adaptive adjustment of the noise covariance matrix accurately distinguishes the model error from the measurement interference. When the prediction deviation exceeds the threshold based on historical fluctuation statistics, the parameter recalibration process is triggered, and individualized adaptation is achieved through transfer learning of neural differential equation parameters. This embodiment significantly improves the stability of long-term prediction. While ensuring the convenience of home use, it enables short-term high-precision acquisition data to continuously correct the prediction results under long-term sparse monitoring, effectively overcoming the evaluation bias caused by the discontinuity of home environment data.

[0060] In some embodiments, the method further comprises: Determine whether the parameters in the feedback correction information exceed their corresponding normal fluctuation range. If so, trigger the re-collection mode; Alternatively, the deviation between the revised prediction result and the initial prediction result is calculated, and when the deviation exceeds a preset deviation threshold, a re-collection mode is triggered; The recollection mode is configured as: Re-collect the user's wearable device monitoring data and behavioral test data; The wristband collected signal was decomposed into multiple frequency bands again, and the tremor component, gait component and autonomic nerve fluctuation component were separated using wavelet transform. The tremor component was re-extracted with a long short-term memory network to extract tremor intensity features, the gait component was extracted with a temporal convolutional network to extract gait symmetry features, and the autonomic nerve fluctuation component was extracted with a Gaussian process regression algorithm to extract autonomic nerve rhythm features; A deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and the motor function score is calculated based on the results of standardized movement tests. Re-analyze the heart rate variability of the blood pressure monitor and wristband signals, calculate the circadian rhythm fluctuation characteristics, and identify abnormal neurological function patterns using a density clustering algorithm; Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores and neurological dysfunction patterns, a neurological function state map was constructed, and the contribution weight of each feature to disease risk was calculated through the graph attention mechanism; In the neural function state map, the neural differential equation is re-adopted based on the historical detection data to learn the continuous time evolution law of the neural function state, predict the disease progression trend in the reset preset future time period, and generate updated prediction results.

[0061] In this embodiment, the triggering mechanism of the re-collection mode is verified in multiple dimensions to ensure reliability. The parameters in the feedback correction information include not only the time domain features such as the amplitude of the tremor component and the coefficient of variation of the gait cycle collected by the bracelet, but also the spatial features such as the smoothness index of the eye movement trajectory obtained by the analysis of the motion behavior image. The normal fluctuation range of these parameters is determined by the dynamic percentile method, that is, the distribution interval of the characteristic value is calculated based on the user's recent valid monitoring results, and an appropriate safety margin is retained. When it is detected that the parameter value exceeds the range for multiple consecutive times, it indicates that there may be a device abnormality or a sudden change in the user's status. At this time, the triggered re-collection mode will give priority to collecting the original signal corresponding to the disputed parameter.

[0062] The default calculation window length for the distribution interval is set to the user's most recent 30 valid monitoring cycles. Safety margins are set differently based on feature type: 1.5 times the standard deviation within the window for tremor, 1.2 times for gait, and the 95th percentile interval of a nonparametric method for eye movement. The window length and safety margin parameters can be manually adjusted based on the stage of the disease through the clinician interface.

[0063] While maintaining the standard processing flow, the re-acquisition mode implements a focused monitoring strategy for abnormal parameters. For abnormal tremor components, the acquisition time is extended and the sampling frequency is increased. When gait characteristics deviate, additional test scenarios are added under different walking conditions. This targeted acquisition strategy optimizes acquisition efficiency while maintaining data quality by adjusting the combination of acquisition time, sampling frequency, and test scenarios.

[0064] During feature extraction, model parameters are adaptively adjusted based on the characteristics of the re-acquired data. For example, if signal quality deteriorates, the wavelet transform's denoising power is automatically increased; if feature distribution shifts are detected, the neural network model's input normalization parameters are adjusted accordingly. This dynamic adaptation mechanism effectively improves the robustness of feature extraction in abnormal situations.

[0065] Optimally, a feature credibility assessment mechanism is introduced during the graph construction phase. The initial weights of each feature in the graph attention network are dynamically adjusted based on the quality of the re-collected data. For parameters with suboptimal collection conditions, their contribution weights are appropriately reduced, while the decision weights of other reliable features are increased. This adaptive feature fusion strategy ensures the stability of the final prediction results.

[0066] Predictive model updates preferably employ a gradual adjustment strategy, controlling the magnitude of model parameter updates by comparing the difference between old and new predictions. When the difference is small, fine-tuning is used to preserve existing knowledge, while significant differences trigger a full model retraining process. This hierarchical model update mechanism ensures the timeliness of predictions while avoiding model oscillations caused by single anomalous data.

[0067] This embodiment realizes the self-correction capability of the prediction system by establishing a dynamic feedback correction mechanism, which automatically triggers the re-acquisition mode when a parameter anomaly or a deviation in the prediction result is detected. The mechanism first determines the reliability of the data based on the normal fluctuation range or deviation threshold set by personalization, and then performs a complete re-acquisition and processing process, including multi-source data acquisition, signal decomposition, feature extraction, map construction and trend prediction. By prioritizing the acquisition of controversial parameters, adaptively adjusting the processing strategy, and progressive model updating, the system can effectively respond to data anomalies while ensuring the continuity of the prediction, so that it can continuously provide accurate neurological function status assessment and disease progression prediction, significantly improving the reliability and practicality of long-term monitoring.

[0068] In some embodiments, the wristband collected signals include acceleration signals and gyroscope signals. The wristband collected signals are decomposed into multiple frequency bands and wavelet transform is used to separate the tremor component, gait component, and autonomic nerve fluctuation component, including: The acceleration and gyroscope signals are decomposed at multiple scales using wavelet basis functions with time-frequency localization characteristics. The original signals are separated into a first-band signal containing tremor characteristics, a second-band signal containing gait cycle characteristics, and a third-band signal reflecting autonomic nervous system regulation according to preset frequency bands. An adaptive decomposition algorithm based on signal energy distribution is adopted for the first frequency band signal, the second frequency band signal, and the third frequency band signal, and the number of wavelet decomposition layers is dynamically adjusted according to the signal-to-noise ratio of the signal of each frequency band to obtain the decomposed first sub-band signal, the second sub-band signal, and the third sub-band signal, and motion artifact suppression processing is performed on each decomposed sub-band signal; The first sub-band signal after the motion artifact suppression process is merged into a time-domain continuous tremor component through a phase consistency reconstruction algorithm; The second sub-band signal that has been processed for motion artifact suppression is subjected to motion cycle phase calibration using a dynamic time warping algorithm to obtain a gait component. The third sub-band signal, which had undergone motion artifact suppression processing, was subjected to ensemble empirical mode decomposition to remove baseline drift, maximum likelihood estimation to restore physiological rhythm components, and 0.01-0.1 Hz band-pass filtering to extract sympathetic-parasympathetic nerve regulation characteristics, thereby obtaining the autonomic nerve fluctuation component.

[0069] In this embodiment, the db4 wavelet is preferably used as the wavelet basis function with time-frequency localization. Its tight support properties are suitable for capturing transient characteristics of limb movement. Acceleration and gyroscope signals are synchronously acquired to ensure spatiotemporal alignment. The predefined frequency bands are determined based on the typical physiological frequency bands of clinical tremor, gait, and autonomic nervous system fluctuations.

[0070] The adaptive decomposition algorithm calculates the short-term energy entropy of signals in each frequency band using a sliding window. When the signal-to-noise ratio falls below a preset threshold, the number of decomposition layers is automatically increased to enhance feature separation. Otherwise, the number of layers is reduced to avoid over-decomposition. Preferably, the preset signal-to-noise ratio threshold is determined by the energy distribution statistics of the user's historical monitoring data. When the signal quality assessment index falls below this threshold, the adjustment of the number of decomposition layers is triggered. Motion artifact suppression processing uses independent component analysis combined with motion sensor reference signals, and blind source separation to eliminate interference components introduced by non-physiological limb movement.

[0071] The phase consistency reconstruction algorithm calculates the instantaneous phase gradient of the first subband signal and selects frequency band components with phase synchronization exceeding a preset value for superposition, ensuring the temporal continuity of the tremor component. Preferably, the preset phase synchronization value is dynamically determined by analyzing the phase stability characteristics of the user's historical tremor signals, prioritizing the retention of frequency band components that match typical pathological tremor patterns. The dynamic time warping algorithm performs nonlinear alignment on the second subband signal using the user's historical gait cycle as a template, eliminating phase shifts caused by changes in walking speed and preserving the cyclic stability of the gait components. In processing the third subband signal, ensemble empirical mode decomposition (EMD) uses noise-assisted decomposition to eliminate low-frequency drift caused by breathing or device movement. Maximum likelihood estimation, based on the Markov characteristics of physiological rhythms, recovers rhythmic components obscured by noise. A 0.01-0.1Hz bandpass filter ultimately extracts the frequency band reflecting sympathetic-parasympathetic balance, with a cutoff frequency that strictly corresponds to the very low and low frequency bands analyzed for heart rate variability.

[0072] This embodiment uses multi-band decomposition of wavelet transforms and adaptive signal processing to accurately separate mixed acceleration / gyroscope signals into tremor, gait, and autonomic nervous system fluctuation components. Dynamic adjustment of the decomposition level ensures robust feature extraction. Phase consistency reconstruction and dynamic time warping address tremor signal fragmentation and gait phase asynchrony, respectively. Combined with physiological rhythm restoration and bandpass filtering, reliable extraction of sympathetic-parasympathetic nervous system features is achieved, providing high-fidelity multimodal physiological feature input for predicting long-term disease risk based on short-term data.

[0073] In some embodiments, extracting tremor intensity features from the tremor component using a long short-term memory network includes: Construct a bidirectional long short-term memory network model with attention mechanism; Conduct time series analysis on the tremor component to obtain the amplitude change rate and time series fluctuation characteristics of the tremor component; A tremor intensity feature representing the severity of tremor is generated based on the amplitude change rate and time series fluctuation characteristics; The gait components are extracted through the temporal convolutional network to extract gait symmetry features, including: The gait components are input into a multi-scale dilated temporal convolutional network, and the gait cycle features of different time scales are extracted through multiple dilated convolutional layers set in parallel; The cross-correlation function of the left and right limb motion signals in the time domain and frequency domain is calculated according to the gait cycle characteristics; Hilbert transform is used to obtain the phase difference and output the symmetric feature vector representing the gait asymmetry; The autonomic nerve fluctuation component is extracted through the Gaussian process regression algorithm, including: The autonomic nerve fluctuation component is used as input and its nonlinear change trend is fitted by Gaussian process regression model; and, the squared exponential kernel function was used to model the autocorrelation characteristics of the autonomic nerve fluctuation components; The output includes the low-frequency oscillation power ratio, the high-frequency oscillation power ratio, and the autonomic rhythm characteristics of the ratio of the two.

[0074] In this embodiment, a bidirectional long-short-term memory network model with an attention mechanism captures long-range dependencies of tremor components through gated units. The attention mechanism dynamically weights amplitude mutation information at key time nodes. Its bidirectional structure simultaneously analyzes historical and future tremor trends, thereby enhancing the sensitivity of tremor intensity characteristics to disease progression. Preferably, the amplitude change rate quantifies the rate of tremor energy fluctuation through a differential operation, while the temporal fluctuation characteristic uses the Hurst exponent to characterize the nonstationarity of the tremor signal. The fusion of the two generates a tremor intensity characteristic that can distinguish the deterioration patterns of physiological and pathological tremors.

[0075] In the multi-scale dilated temporal convolutional network, parallel dilated convolutional layers utilize differentiated dilation coefficients to cover multi-granularity analysis, from single steps to complete gait cycles. Their dilated structure expands the receptive field while maintaining temporal resolution. The cross-correlation function is calculated using a sliding window covariance method combined with instantaneous phase difference detection using the Hilbert transform. This allows symmetric eigenvectors to simultaneously reflect the amplitude difference and time delay of left and right limb movements, accurately quantifying gait abnormalities caused by neurological diseases such as Parkinson's disease.

[0076] In the Gaussian process regression algorithm, the squared exponential kernel function adaptively matches the quasi-periodicity of the autonomic nervous system fluctuation component through a length scale parameter. The nonlinear trend of the fitted function preserves the physiological delay characteristics of sympathetic-parasympathetic nervous system regulation. The extraction of the power ratio of low-frequency to high-frequency oscillations strictly follows the frequency band division criteria of heart rate variability analysis. The low-frequency band (0.04-0.15Hz) reflects sympathetic nervous system tone, while the high-frequency band (0.15-0.4Hz) represents parasympathetic nervous system activity. The ratio of the two, as an indicator of autonomic nervous system balance, directly corresponds to clinical assessment criteria.

[0077] This implementation uses a bidirectional long-short-term memory network to capture the temporal evolution of tremor components, incorporating an attention mechanism to enhance the extraction of key pathological features, enabling tremor intensity characteristics to sensitively reflect disease progression. A multi-scale dilated temporal convolutional network analyzes gait cycles at different time granularities, combining cross-correlation analysis and phase difference detection to accurately quantify gait asymmetry. Gaussian process regression accurately extracts autonomic rhythm features through nonlinear modeling. The synergistic effect of these three feature extraction methods provides a highly accurate, multi-dimensional quantitative basis for predicting neurological disease progression based on short-term monitoring data.

[0078] In some embodiments, a deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and combined with standardized movement test results to calculate a movement function score, including: The spatiotemporal features of the motion behavior images are encoded through a three-dimensional convolutional neural network, the spatiotemporal sequence features of the eye movements are extracted through the spatiotemporal convolution layer, and the eye movement trajectory feature vector is output; Use graph convolutional networks to build a topological graph model of limb joint motion data, and generate a motion coordination feature matrix based on the spatial connection relationship of the joints; The eye movement trajectory feature vector is fused with the movement coordination feature matrix, and the time-frequency features of the electromyographic signals collected in the standardized movement test are connected. The cross-modal features are weighted through a multi-layer perceptron to output the motor function score.

[0079] In this embodiment, the spatiotemporal convolutional layer of the 3D convolutional neural network utilizes a separable 3D convolutional architecture. The spatial convolution kernel extracts eye position features within a single frame, while the temporal convolution kernel captures the changing patterns of eye movement trajectories between consecutive frames. These two kernels work together to achieve spatiotemporal encoding of eye movement behavior. The eye movement trajectory feature vector includes quantitative metrics such as saccade velocity and fixation duration. Its dimensions are aligned with clinical parameters from a standardized eye movement database to ensure feature interpretability.

[0080] The topological graph model constructed by the graph convolutional network uses human anatomical connections as initial edge weights and dynamically adjusts the functional connectivity strength between joints through adaptive graph learning. A movement coordination feature matrix is ​​generated using a spatiotemporal graph attention mechanism. This matrix preserves the movement dominance of large joints such as the shoulder, elbow, and wrist, while capturing the fine coordination patterns of finger micro-movements, thereby comprehensively reflecting the detailed characteristics of movement dysfunction.

[0081] In the feature fusion stage, gated attention units are used to achieve cross-modal interaction. Preferably, the time-frequency features of the electromyographic signals are used to extract the activation timing and intensity information of each muscle group through short-time Fourier transform, which complements the visual behavioral features. The weighting coefficients of the multi-layer perceptron are determined through comparative pre-training, so that the motor function score values ​​can simultaneously meet the discrimination criteria of the clinical scale and the sensitivity requirements of home monitoring, providing standardized input for the transformation of short-term assessment results into long-term prognosis prediction.

[0082] This implementation uses a three-dimensional convolutional neural network to accurately extract the spatiotemporal features of eye movement trajectories, combined with a graph convolutional network to model the topological relationships of limb joint movements, enabling multidimensional analysis of motor behavior. By integrating eye movement features, movement coordination features, and standardized electromyographic features, a cross-modal weighting mechanism is employed to generate a comprehensive motor function score. This solution transforms complex clinical motor function assessments into quantifiable, objective indicators, providing a highly accurate assessment basis for predicting the progression of neurodegenerative diseases based on short-term behavioral data, while also facilitating implementation in the home.

[0083] In some embodiments, heart rate variability analysis is performed on the signals collected by the blood pressure monitor and the wristband, circadian rhythm fluctuation characteristics are calculated, and abnormal neurological function patterns are identified using a density clustering algorithm, including: Perform dynamic blood pressure trend extraction on the sphygmomanometer collected signal to obtain the systolic / diastolic blood pressure periodic change curve; The RR interval sequence analysis of the wristband collected signal was performed, and the day and night ratio of low-frequency power to high-frequency power was extracted based on frequency domain transformation; The periodic variation curve and the day-night ratio were input into a density clustering model based on DBSCAN. Through adaptive neighborhood radius parameter optimization, abnormal clusters representing sympathetic-parasympathetic imbalance were divided in the feature space of the density clustering model. The data points corresponding to the periodic change curve and the day-night ratio that fall into the abnormal cluster are marked as abnormal neurological function patterns and the abnormal probability is output.

[0084] In this embodiment, dynamic blood pressure trend extraction uses a sliding window smoothing algorithm to process the original blood pressure signal, where the cyclical variation curve of systolic and diastolic pressure is obtained by 24-hour segmented fitting, and the least squares fitting coefficient of each period is used to characterize the circadian rhythm characteristics of blood pressure. Preferably, RR interval series analysis uses the Lomb-Scargle periodogram method to process non-uniform sampling data, and the day and night ratio of low-frequency power (0.04-0.15Hz) to high-frequency power (0.15-0.4Hz) is calculated using the segmented integration method to ensure the reliability of the analysis in the discontinuous monitoring scenario at home.

[0085] The DBSCAN-based density clustering model optimizes the neighborhood radius parameter through a dynamic core point determination strategy. Specifically, the k-distance graph method is used to determine the optimal neighborhood range, enabling the feature space to simultaneously accommodate the multidimensional characteristics of blood pressure variability and heart rate variability. The criteria for determining abnormal clusters are combined with threshold boundaries set by clinical guidelines. When the cluster center deviates from the normal physiological reference range, abnormal neurological function is automatically flagged. The output abnormal probability is calculated using kernel density estimation to calculate the membership of the data point to the abnormal cluster center. The numerical range corresponds to the grading criteria of existing neurological function assessment scales, facilitating the interpretation of clinical results.

[0086] This example uses dynamic blood pressure trend extraction and RR interval analysis to obtain blood pressure cycle curves and heart rate variability diurnal ratios, and automatically identifies patterns of sympathetic-parasympathetic imbalance using a density clustering model. This method converts discrete monitoring data into continuous neurological function assessment indicators. Through intelligent clustering of abnormalities, it enables objective identification of early neurological dysfunction, providing a quantitative basis for predicting autonomic nervous system deterioration based on short-term physiological data, while also meeting the needs of convenient monitoring in home settings.

[0087] In the second aspect, this embodiment also provides a neurological disease detection and analysis system, which is applicable to the method described in the first aspect, and the system includes: a wearable device, a behavior collection device, a signal processing unit, and a disease analysis unit. The wearable device includes at least one of a bracelet, glasses, a helmet, and headphones; the behavior collection device includes at least one of a mobile phone, a tablet, and a camera; the signal processing unit is communicatively connected with the wearable device and the behavior collection device, and the signal processing unit is used to obtain the user's wearable device monitoring data and behavior test data; the bracelet collected signal is decomposed into multiple frequency bands, and the tremor component, gait component and autonomic nerve fluctuation component are separated by wavelet transform; the tremor intensity feature of the tremor component is extracted by the long short-term memory network, the gait symmetry feature of the gait component is extracted by the time series convolution network, and the autonomic nerve rhythm feature of the autonomic nerve fluctuation component is extracted by the Gaussian process regression algorithm; the deep learning model is used to extract the eye movement trajectory feature and the movement coordination feature of the movement behavior image, and combined with the standardized movement test The motor function score is calculated based on the test results; the heart rate variability analysis is performed on the signals collected by the blood pressure monitor and the bracelet, the circadian rhythm fluctuation characteristics are calculated, and the abnormal neurological function pattern is identified through the density clustering algorithm; the disease analysis unit is used to construct a neurological function state map based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function score and neurological function abnormality pattern, and calculate the contribution weight of each feature to the disease risk through the graph attention mechanism; the user's historical monitoring data is obtained, and the continuous time evolution law of the neurological function state is learned by the neural differential equation based on the historical detection data in the neurological function state map, and the disease progression trend in the preset future time period is predicted to generate a prediction result; in the preset future time period, at least one of the signals collected by the bracelet, the signals collected by the blood pressure monitor and the movement behavior images captured by the mobile phone camera are collected and recorded as feedback correction information, and the prediction result is corrected based on the adaptive Kalman filter according to the feedback correction information; a neurological disease risk assessment report is output according to the prediction result.

[0088] In this embodiment, the wearable device, a wristband, collects physiological signals such as heart rate and blood pressure; glasses and helmets track eye movements; and headphones capture voice tremor. The behavioral data collection devices, such as mobile phones and tablets, use built-in sensors to acquire gait data, while cameras record motion images. The signal processing unit includes a multi-channel signal synchronization module to ensure time alignment of data from different devices.

[0089] The neural function state map in the disease analysis unit is constructed using a graph neural network, with nodes representing various features and edge weights dynamically adjusted via a graph attention mechanism. The neural differential equation system uses a neural ordinary differential equation architecture to model the continuous dynamics of disease progression through latent state differentiation. The adaptive Kalman filter achieves incremental correction of prediction results through online updates of the noise covariance matrix.

[0090] This system acquires multimodal data from wearable devices and behavioral data collection equipment. After the signal processing unit extracts key features such as tremor intensity, gait symmetry, and autonomic rhythms, the disease analysis unit constructs a dynamic neurological function state map. Combining historical data with real-time feedback, it uses neural differential equations to predict disease progression. This system enables comprehensive assessment and long-term prediction of neurological function in a home setting, providing an objective basis for early intervention.

[0091] By adopting the above technical solutions, the present invention is different from the existing technology and has the following beneficial effects: short-term comprehensive monitoring data is obtained through wearable devices and behavioral collection devices, multi-band signal decomposition technology is used to separate tremor intensity characteristics, gait symmetry characteristics, and autonomic nerve rhythm characteristics, and deep learning models are combined to extract eye movement trajectories and movement coordination characteristics to construct a multi-dimensional neural function state map. Based on the continuous evolution law of neural function modeling using neural differential equations, it is possible to predict long-term disease progression trends from short-term data. The system obtains local monitoring data through a real-time feedback mechanism and uses an adaptive filtering algorithm to dynamically correct the prediction results. When data anomalies or prediction deviations are detected, the complete data re-collection and processing process is automatically triggered to ensure the accuracy of the prediction. This technical solution that combines short-term comprehensive collection with long-term local monitoring effectively solves the problem of data discontinuity in long-term prediction of neurological diseases in a home environment. It not only ensures the comprehensiveness of the initial assessment, but also maintains the reliability of long-term predictions through dynamic correction, providing an objective basis for early intervention in neurological diseases.

[0092] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of this application, this does not limit the scope of patent protection of this application. All technical solutions generated by replacing or modifying equivalent structures or equivalent processes based on the essential concepts of this application using the contents recorded in the specification and drawings of this application, as well as directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are included in the scope of patent protection of this application.

Claims

1. A method for detecting and analyzing neurological diseases, characterized in that: include: Obtain the user's wearable device monitoring data and behavioral test data. Wearable device monitoring data includes wristband acquisition signals, blood pressure monitor acquisition signals and motion behavior images. Behavioral test data includes standardized movement test results. The wristband collected signal is decomposed into multiple frequency bands, and the tremor component, gait component and autonomic nerve fluctuation component are separated by wavelet transform; The tremor component is extracted by long short-term memory network to extract tremor intensity features, the gait component is extracted by time series convolution network to extract gait symmetry features, and the autonomic nerve fluctuation component is extracted by Gaussian process regression algorithm to extract autonomic nerve rhythm features; A deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and the motor function score is calculated based on the results of standardized movement tests. Heart rate variability analysis was performed on the signals collected by the blood pressure monitor and the wristband, the circadian rhythm fluctuation characteristics were calculated, and abnormal neurological function patterns were identified using a density clustering algorithm; Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores and neurological dysfunction patterns, a neurological function state map is constructed, and the contribution weight of each feature to disease risk is calculated through the graph attention mechanism; Obtain the user's historical monitoring data, use neural differential equations to learn the continuous time evolution law of the neural function state based on the historical test data in the neural function state map, predict the disease progression trend in the preset future time period, and generate prediction results; Collect at least one of the wristband signal, the blood pressure monitor signal, and the motion behavior image captured by the mobile phone camera within a preset future time period, record it as feedback correction information, and correct the prediction result based on the adaptive Kalman filter according to the feedback correction information; Output a neurological disease risk assessment report based on the prediction results.

2. The neurological disease detection and analysis method according to claim 1, characterized in that: Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores, and neurological dysfunction patterns, a neurological function state map is constructed. The contribution weight of each feature to disease risk is calculated through the graph attention mechanism, including: Tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores, and neurological dysfunction patterns are used as node features of the graph neural network, and initial edge connection relationships are constructed based on the physiological correlation between each node feature; A graph attention mechanism was used to model the spatial dependencies of node features. The motor function association weights between tremor intensity features and gait symmetry features, the autonomic regulation weights between autonomic rhythm features and neurological dysfunction patterns, and the comprehensive coordination weights between motor function scores and other node features were calculated. Through a multi-layer graph convolutional network, the neighborhood node information is aggregated, the embedded representation of each node feature is iteratively optimized, and a graph embedding vector reflecting the overall state of neural function is output; Based on the graph embedding vector, the attention score calculation module generates the dynamic contribution weight of each node feature to the disease risk. The weights of tremor intensity features and gait symmetry features are used for motor dysfunction assessment, the weights of autonomic rhythm features and abnormal neurological function patterns are used for autonomic nervous function assessment, and the weights of motor function scores are used for comprehensive behavioral function assessment.

3. The neurological disease detection and analysis method according to claim 1, characterized in that: Obtain the user's historical monitoring data, use neural differential equations to learn the continuous time evolution law of neural function state based on historical test data in the neural function state map, predict the disease progression trend in the preset future time period, and generate prediction results, including: Integrate historical monitoring data into the neural function state map to form multiple historical embedding vectors, and arrange the multiple historical embedding vectors in time series to construct the time evolution sequence of the neural function state; The time evolution sequence is calculated using a neural differential equation, and the gradual change process of the neural function state is described by a learnable continuous-time dynamic function, which is expressed by formula (1). The formula (1) is as follows: ； In formula (1), is the neural differential equation, is a learnable continuous-time dynamics function, for The neural function state vector at the moment, for The weight parameter, is a continuous time variable; A gating mechanism is introduced into the neural differential equation framework to dynamically adjust the state change rate at different time scales to capture the short-term fluctuations and long-term degradation characteristics of neural functions. The neural differential equation model is obtained and expressed by formula (2). Formula (2) is as follows: ； In formula (2), is the gating vector corresponding to the gating mechanism, is the residual term, is element-wise multiplication; Based on the current neural function state map and the trained neural differential equation model, multi-step state prediction within a preset future period is performed, and the predicted neural function state trajectory including time continuity is output; Probabilistically process the predicted trajectory of neurological function status, calculate the probability distribution of disease risk at each time point, and identify the key time nodes where the risk exceeds the preset threshold; The predicted trajectory of neurological function status is compared and analyzed with the user's individual baseline data to generate personalized prediction results of disease progression trends, which include risk level change curves and warning information at key time nodes.

4. The neurological disease detection and analysis method according to claim 1, characterized in that: At least one of the wristband signal, the blood pressure monitor signal, and the motion behavior image captured by the mobile phone camera is collected within a preset future period, and recorded as feedback correction information. The prediction result is corrected based on the adaptive Kalman filter according to the feedback correction information, including: Inputting the feedback correction information into the feature extraction module, and updating at least one of the tremor intensity feature, gait symmetry feature, autonomic nerve rhythm feature, and motor function score according to the extracted feedback features to generate a neural function state observation vector at the current moment; Constructing a state space model, using the predicted trajectory output by the neural differential equation as a state transition prior, and using the neural function state observation vector as a measurement value to generate a state estimation result; Adaptive Kalman filter algorithm is used to generate confidence weights of state estimation results by dynamically adjusting process noise covariance matrix and observation noise covariance matrix; And, when the deviation between the observed and predicted values ​​exceeds a dynamic threshold, a recalibration process of the prediction model is triggered; Correcting the state estimation result according to the confidence weight; The corrected state estimation results are fed back to the neural differential equation model, the parameters of the dynamic function are adjusted to adapt to the user's individualized neural function change pattern, and the corrected prediction results are output.

5. The neurological disease detection and analysis method according to claim 1, characterized in that: The method further comprises: Determining whether the parameters in the feedback correction information exceed their corresponding normal fluctuation ranges, and if so, triggering a re-collection mode; Alternatively, the deviation between the revised prediction result and the initial prediction result is calculated, and when the deviation exceeds a preset deviation threshold, a re-collection mode is triggered; The reacquisition mode is configured as: Re-collect the user's wearable device monitoring data and behavioral test data; The wristband collected signal was decomposed into multiple frequency bands again, and the tremor component, gait component and autonomic nerve fluctuation component were separated using wavelet transform. The tremor component was re-extracted with a long short-term memory network to extract tremor intensity features, the gait component was extracted with a temporal convolutional network to extract gait symmetry features, and the autonomic nerve fluctuation component was extracted with a Gaussian process regression algorithm to extract autonomic nerve rhythm features; A deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and the motor function score is calculated based on the results of standardized movement tests. Re-analyze the heart rate variability of the blood pressure monitor and wristband signals, calculate the circadian rhythm fluctuation characteristics, and identify abnormal neurological function patterns using a density clustering algorithm; Based on tremor intensity characteristics, gait symmetry characteristics, autonomic nerve rhythm characteristics, motor function scores and neurological dysfunction patterns, a neurological function state map was constructed, and the contribution weight of each feature to disease risk was calculated through the graph attention mechanism; In the neural function state map, the neural differential equation is re-adopted based on the historical detection data to learn the continuous time evolution law of the neural function state, predict the disease progression trend in the reset preset future time period, and generate updated prediction results.

6. The neurological disease detection and analysis method according to claim 1, characterized in that: The wristband collected signals include acceleration signals and gyroscope signals. The wristband collected signals are decomposed into multiple frequency bands, and wavelet transform is used to separate the tremor component, gait component and autonomic nerve fluctuation component, including: The acceleration and gyroscope signals are decomposed at multiple scales using wavelet basis functions with time-frequency localization characteristics. The original signals are separated into a first-band signal containing tremor characteristics, a second-band signal containing gait cycle characteristics, and a third-band signal reflecting autonomic nervous system regulation according to preset frequency bands. Adopting an adaptive decomposition algorithm based on signal energy distribution for the first frequency band signal, the second frequency band signal, and the third frequency band signal, dynamically adjusting the number of wavelet decomposition layers according to the signal-to-noise ratio of each frequency band signal, to obtain decomposed first sub-band signal, second sub-band signal, and third sub-band signal, and performing motion artifact suppression processing on each of the decomposed sub-band signals; The first sub-band signal after the motion artifact suppression process is merged into a time-domain continuous tremor component through a phase consistency reconstruction algorithm; The second sub-band signal that has been processed for motion artifact suppression is subjected to motion cycle phase calibration using a dynamic time warping algorithm to obtain a gait component. The third sub-band signal, which had undergone motion artifact suppression processing, was subjected to ensemble empirical mode decomposition to remove baseline drift, maximum likelihood estimation to restore physiological rhythm components, and 0.01-0.1 Hz band-pass filtering to extract sympathetic-parasympathetic nerve regulation characteristics, thereby obtaining the autonomic nerve fluctuation component.

7. The neurological disease detection and analysis method according to claim 1, characterized in that: The tremor intensity features are extracted from the tremor component through the long short-term memory network, including: Construct a bidirectional long short-term memory network model with attention mechanism; Conduct time series analysis on the tremor component to obtain the amplitude change rate and time series fluctuation characteristics of the tremor component; generating a tremor intensity feature representing the severity of the tremor based on the amplitude change rate and the time series fluctuation feature; The gait components are extracted through the temporal convolutional network to extract gait symmetry features, including: The gait components are input into a multi-scale dilated temporal convolutional network, and the gait cycle features of different time scales are extracted through multiple dilated convolutional layers set in parallel; Calculating the cross-correlation function of the left and right limb motion signals in the time domain and the frequency domain according to the gait cycle characteristics; Hilbert transform is used to obtain the phase difference and output the symmetric feature vector representing the gait asymmetry; The autonomic nerve fluctuation component is extracted through the Gaussian process regression algorithm, including: The autonomic nerve fluctuation component is used as input and its nonlinear change trend is fitted by Gaussian process regression model; and, the squared exponential kernel function was used to model the autocorrelation characteristics of the autonomic nerve fluctuation components; The output includes the low-frequency oscillation power ratio, the high-frequency oscillation power ratio, and the autonomic rhythm characteristics of the ratio of the two.

8. The neurological disease detection and analysis method according to claim 1, characterized in that: A deep learning model is used to extract eye movement trajectory features and movement coordination features from movement behavior images, and combined with standardized movement test results to calculate movement function scores, including: The motion behavior image is encoded with spatiotemporal features by a three-dimensional convolutional neural network, the spatiotemporal sequence features of the eye movement are extracted through a spatiotemporal convolution layer, and an eye movement trajectory feature vector is output; Use graph convolutional networks to build a topological graph model of limb joint motion data, and generate a motion coordination feature matrix based on the spatial connection relationship of the joints; The eye movement trajectory feature vector is fused with the movement coordination feature matrix, and the time-frequency features of the electromyographic signals collected in the standardized movement test are simultaneously connected. The cross-modal feature weighting is performed through a multi-layer perceptron to output the motor function score value.

9. The neurological disease detection and analysis method according to claim 1, characterized in that: Heart rate variability analysis is performed on the signals collected by the blood pressure monitor and the wristband, the circadian rhythm fluctuation characteristics are calculated, and the abnormal neurological function patterns are identified through the density clustering algorithm, including: Performing dynamic blood pressure trend extraction on the blood pressure meter collected signal to obtain a periodic variation curve of systolic / diastolic pressure; Performing RR interval sequence analysis on the wristband collected signal, and extracting the day and night ratio of low-frequency power to high-frequency power based on frequency domain transformation; Inputting the periodic variation curve and the day-night ratio into a DBSCAN-based density clustering model, and dividing abnormal clusters representing sympathetic-parasympathetic imbalance in the feature space of the density clustering model through adaptive neighborhood radius parameter optimization; The data points corresponding to the periodic variation curve and the day-night ratio that fall into the abnormal cluster are marked as abnormal neurological function patterns and the abnormal probability is output.

10. A neurological disease detection and analysis system, characterized in that: The method according to any one of claims 1 to 9, wherein the system comprises: Wearable devices, including at least one of a wristband, glasses, helmet, or earphones; Behavior collection devices, including at least one of a mobile phone, tablet, or camera; A signal processing unit is communicatively connected to the wearable device and the behavior collection device, and is used to obtain the user's wearable device monitoring data and behavior test data; the wristband collected signal is decomposed into multiple frequency bands, and the tremor component, gait component and autonomic nerve fluctuation component are separated by wavelet transform; the tremor intensity feature of the tremor component is extracted by a long short-term memory network, the gait symmetry feature of the gait component is extracted by a temporal convolutional network, and the autonomic nerve rhythm feature of the autonomic nerve fluctuation component is extracted by a Gaussian process regression algorithm; the eye movement trajectory feature and movement coordination feature are extracted from the movement behavior image by a deep learning model, and the movement function score is calculated in combination with the standardized movement test results; the heart rate variability analysis is performed on the blood pressure monitor collected signal and the wristband collected signal, the circadian rhythm fluctuation feature is calculated, and the abnormal neurological function pattern is identified by a density clustering algorithm; The disease analysis unit is used to construct a neural function state map based on tremor intensity characteristics, gait symmetry characteristics, autonomic nervous rhythm characteristics, motor function scores and neural function abnormality patterns, and calculate the contribution weight of each feature to the disease risk through the graph attention mechanism; obtain the user's historical monitoring data, use neural differential equations to learn the continuous time evolution law of the neural function state in the neural function state map according to the historical detection data, predict the disease progression trend in a preset future time period, and generate a prediction result; collect at least one of the bracelet collection signal, blood pressure monitor collection signal and movement behavior image captured by the mobile phone camera in the preset future time period, record it as feedback correction information, and correct the prediction result based on the feedback correction information based on the adaptive Kalman filter; output a neurological disease risk assessment report according to the prediction result.

Images