Medical data processing method and system for reversible adhesion of hydrogel brain electrode

Abstract

The invention discloses a hydrogel brain electrode reversible adhesion medical data processing method and system, and relates to the technical field of medical data processing.The method comprises the specific steps that firstly, a reversible adhesion electrode is attached to the brain of a patient, and multi-modal data are circularly collected to form a dynamic data set; preprocessing and fusing the data to generate a correlation graph; based on a transfer learning training model, a diagnosis result and a prescription are output; then, collaborative intervention is executed, the state is detected, parameters are determined, and drugs are delivered; and finally, acquiring intervened data according to nodes, optimizing and updating by using an algorithm, and forming a circulating optimization link. According to the method, the reversible adhesion electrode and a multi-mode acquisition technology are integrated, a brain function-molecular mechanism association map is constructed, information limitation is broken, and the disease recognition accuracy is improved; a personalized electroencephalogram characteristic spectrum is constructed through transfer learning and the like, collaborative intervention and closed-loop feedback optimization are realized by relying on multiple algorithms, the traditional intervention problem is solved, and accurate and dynamic development of nerve regulation and control is promoted.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of medical data processing, in particular to a medical data processing method and system for reversible adhesion of hydrogel brain electrodes. BACKGROUND

[0002] With the rapid development of brain science research and neural regulation technology, the diagnosis and treatment of brain diseases gradually move towards precision and individualization. As a core tool for obtaining brain electrophysiological signals, semi-dry brain electrodes play a key role in the clinical research and treatment of brain diseases such as epilepsy and cognitive impairment, and in the field of brain-computer interface technology. At the same time, a single brain electrical signal has been difficult to meet the diagnosis needs of complex brain diseases, and in clinical practice, it is gradually necessary to combine physiological state, molecular expression and behavioral characteristics and other multi-dimensional information to fully reveal the pathological mechanism of the disease. In addition, the adhesion performance of the brain electrode directly affects the stability and continuity of signal acquisition. The electrical adhesion property can realize the rapid and stable fitting of the electrode and the scalp, reduce the pressure damage of mechanical fixation to the patient's brain skin, support repeated acquisition and intervention operation. Under this background, how to realize the efficient acquisition and accurate fusion of multi-modal data, combined with the electrical adhesion type semi-dry brain electrode to construct a complete link from data processing to individualized intervention, has become an important direction to promote the clinical transformation of neural regulation technology.

[0003] Traditional brain electrode technology has obvious limitations: although wet brain electrodes can reduce contact impedance and obtain high-quality brain electrical signals, they need to be coated with conductive paste, which is cumbersome to use and inconvenient to clean, and can easily cause skin discomfort or allergies; dry brain electrodes are easy to operate, but the contact with the scalp is unstable, the signal noise is large, and the signal acquisition quality is poor. The semi-dry brain electrode combines the advantages of both, but the existing semi-dry brain electrode mostly adopts mechanical fixation, which can easily cause pressure on the scalp and affect comfort, and the fixation effect is greatly affected by factors such as movement, and lacks an efficient adhesion scheme based on electrical response characteristics. In terms of data processing, traditional methods mostly rely on single modal data for analysis, which is difficult to effectively integrate brain electrical, physiological, molecular and behavioral data, resulting in incomplete disease characteristics extracted, inability to establish precise correlation between brain function and molecular mechanism, and thus affecting the accuracy of the diagnosis result. In the model training link, the traditional model is often trained with a fixed data set and directly applied, lacking individualized adjustment and dynamic incremental update capability for individual differences, and being difficult to adapt to the dynamic changes of the patient's condition. In the intervention execution, the traditional scheme mostly separately uses transcranial electrical stimulation or drug therapy, lacks the collaborative optimization design of the two, and lacks real-time effect feedback and parameter adjustment mechanism, resulting in insufficient pertinence of the intervention scheme, and difficulty in balancing efficacy and safety.

[0004] The existing adhesion technology for semi-dry brain electrodes presents significant challenges: balancing stability and comfort is difficult with mechanical fixation methods, while conventional adhesion materials lack electroresponsiveness, making it impossible to quickly regulate the adhesion state via external signals. Furthermore, the adhesion effect tends to decline after repeated use, affecting not only the continuity of data acquisition but also potentially causing signal distortion due to improper fixation. Therefore, developing semi-dry brain electrodes based on electroresponsiveness and employing a dual-network interpenetrating conductive hydrogel with a composite PVA and PAM network structure, achieving rapid and stable electroadhesion through the application of a continuous DC voltage, while simultaneously integrating multimodal data processing and closed-loop intervention technologies, is key to overcoming the current technological bottlenecks. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a medical data processing method and system for reversibly adhered hydrogel brain electrodes. This method collects multimodal data, including EEG and physiological parameters, using semi-dry brain electrodes. After preprocessing and cross-modal fusion, it constructs an EEG feature atlas using transfer learning and personalized incremental training, and outputs disease diagnosis and personalized neuromodulation prescriptions. The system includes modules for data acquisition, preprocessing and fusion, model training, collaborative intervention, and closed-loop optimization to achieve precision medicine and dynamic adjustment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, a method for medical data processing using reversible adhesion of hydrogel brain electrodes, the specific steps of which are as follows:

[0007] S100, Multimodal Dynamic Acquisition: A semi-dry brain electrode integrating a micro-programmable transcranial AC stimulation array is attached to the target functional area of ​​the patient's brain. The core conductive layer of the semi-dry brain electrode is a conductive hydrogel with a dual-network interpenetrating structure composed of PVA network and PAM network. By applying a continuous DC voltage to the conductive hydrogel, the electro-response characteristics of the material cause the conductive hydrogel to undergo physical changes and generate adhesion force, achieving rapid and stable electro-adhesion between the brain electrode and the scalp. Through the electrode and supporting acquisition equipment, EEG signals, physiological parameters, molecular expression imaging data and behavioral feature data are cyclically acquired to form a multi-dimensional dynamic dataset.

[0008] S200, Data Preprocessing and Fusion: Denoising, normalization, and spatial registration preprocessing are performed on the collected multimodal data respectively. A cross-modal spatiotemporal attention fusion algorithm is used to extract key features of each modality and weighted concatenation to generate cross-modal joint feature vectors and brain function-molecular mechanism association maps.

[0009] The specific steps of the brain function-molecular mechanism association map are as follows: Based on brain anatomy, define brain functional regions including the prefrontal, parietal, temporal, occipital, and hippocampal areas; determine the spatial coordinate range of each region using brain MRI images; extract the expression concentration, spatial distribution range, and expression intensity gradient data of specific molecular markers within each brain functional region from preprocessed molecular expression imaging data; calculate the correlation coefficient between the EEG core features of each brain functional region and the corresponding molecular marker expression data to quantify the correlation strength; construct a two-dimensional map framework with brain functional regions as the horizontal axis and molecular marker types as the vertical axis, filling the corresponding coordinate positions with numerical annotations or color gradients; incorporate spatial correlation matrix information from the cross-modal spatiotemporal attention fusion algorithm to correct the correlation weights between each region and molecular markers, forming a structurally complete and quantitatively accurate brain function-molecular mechanism association map.

[0010] S300, Model Training and Prescription Generation: Based on the transfer learning initialization model framework, a personalized adaptive incremental training algorithm is adopted, and the model parameters are iteratively updated using dynamic datasets to construct personalized EEG feature maps and output disease diagnosis results and personalized neuromodulation prescriptions.

[0011] The specific method for outputting disease diagnosis results and personalized neuromodulation prescriptions is as follows: Cross-modal joint feature vectors, personalized EEG feature maps, and brain function-molecular mechanism correlation maps are simultaneously input into a model optimized by a personalized adaptive incremental training algorithm. The model strengthens the correlation mapping between individual-specific EEG features and molecular expression patterns through a feature extraction network. After classification calculation by a fully connected layer, the disease diagnosis result is output. Based on the pathological mechanism correlation description in the diagnosis result, the expression patterns of molecular markers in the corresponding brain functional regions are extracted from abnormal EEG features. Combined with the brain functional region localization results, the core parameters of the personalized neuromodulation prescription are determined: the stimulation target is the brain functional region corresponding to the abnormal EEG features; the stimulation frequency matches the characteristics of the abnormal EEG bands; the stimulation intensity is adapted according to individual skin impedance and EEG signal amplitude; and the stimulation duration is correlated with the molecular marker expression concentration. The above parameters are integrated to form a complete personalized neuromodulation prescription, which is output to the system terminal simultaneously with the disease diagnosis result, providing instruction support for subsequent collaborative intervention.

[0012] S400, Collaborative Intervention Execution: The adhesion status is detected by sensors integrated with electrodes, and the optimal neuromodulation parameters and drug dosage are determined by a collaborative intervention parameter optimization algorithm. Transcranial electrical stimulation and drug delivery are executed sequentially, and monitoring data are continuously collected throughout the process.

[0013] S500, closed-loop optimization and update: Data after intervention is collected again at preset time nodes and added to the dynamic dataset. The closed-loop effect feedback optimization algorithm is used to calculate the parameter update amount, adjust the model parameters, intervention parameters and collection frequency, and form a cyclic optimization link.

[0014] Furthermore, in the S100 multimodal dynamic acquisition, the conductive hydrogel of the semi-dry brain electrode is formed by combining a PVA network and a PAM network in a mass ratio of 2:1-3:1 to form a double-network interpenetrating structure. An electro-responsive property enhancer accounting for 3%-6% of the total mass of the hydrogel is added. After 3-5 freeze-thaw cycles, it is molded with an elongation ≥1200%, tensile strength ≥280kPa, and reversible electro-adhesion cycles ≥15 times. The adhesion force is released within 30 seconds after power failure. The target functional areas of the patient's brain include: the parietal lobe and hippocampus of Alzheimer's patients, and the temporal lobe and frontal lobe of epilepsy patients. The fitting position is determined according to the lesion-associated area located by brain MRI. The supporting acquisition equipment includes: an Ag / AgCl EEG sensor, a miniature piezoresistive pulse sensor, an NTC thermistor temperature sensor, a CdSe / ZnS core-shell quantum dot labeled probe, a near-infrared fluorescence imaging system, and a wearable accelerometer.

[0015] Furthermore, in S100, the multimodal dynamic acquisition includes brain electrical activity signals, physiological parameters including pulse rate and skin temperature, molecular expression imaging data including expression concentration data and spatial distribution data of specific molecular markers in the brain, and behavioral characteristic data including walking speed, limb tremor frequency, and limb movement amplitude data.

[0016] Furthermore, in S200, during data preprocessing and fusion, the mathematical expression for the cross-modal spatiotemporal attention fusion algorithm is: ,in, It is the final output cross-modal joint feature vector. Corresponding to four modalities: electroencephalography (EEG), molecular imaging, physiological, and behavioral. For modal baseline weights, For modal reliability factor, For each modal feature vector, It is a spatial incidence matrix. This is the time-series decay coefficient. This represents the data collection time interval.

[0017] Furthermore, in S300, the specific steps for constructing a personalized EEG feature atlas during model training and prescription generation are as follows: Core EEG-related features are selected from the cross-modal joint feature vector, including EEG time-domain features, frequency-domain features, and correlation features between EEG signals and molecular marker expressions; based on the model output initialized through transfer learning, and combined with brain anatomical structural partitions, a brain functional partition atlas framework including the prefrontal, parietal, temporal, occipital, and hippocampal regions is established; the feature values ​​corresponding to the EEG data acquired in each adhesion-acquisition cycle are filled into the atlas framework according to brain functional partitions to form an initial personalized EEG feature atlas; the model parameters updated using the personalized adaptive incremental training algorithm are used to correct the feature weights and correlation strengths of each brain functional partition, strengthening the representation of individual-specific EEG features; the atlas is updated synchronously according to the number of adhesion-acquisition cycles, recording the dynamic change trajectory of EEG features in each partition at different time points, forming a complete personalized EEG feature atlas.

[0018] Furthermore, in S300, during model training and prescription generation, the specific content of the transfer learning initialization of the model framework is as follows: The pre-training dataset is selected as the PhysioNet EEG public dataset; the basic model is the ResNet-18 network, which contains 16 convolutional layers, 4 pooling layers, and 3 fully connected layers, where the convolutional layers are used for feature extraction and the fully connected layers are used for classification output; the parameters of the 16 convolutional layers and 4 pooling layers trained on the public dataset in the pre-training model are retained as the initial weights for feature extraction; the original fully connected layers are replaced with a 3-layer adaptive structure, with the first layer containing 1024 neurons, the second layer containing 512 neurons, and the third layer containing 64 neurons. The output layer uses the Softmax activation function to adapt to the classification task of 3 disease types and 3 risk levels; the He initialization method is used to initialize the parameters of the replaced fully connected layers, setting the initial learning rate to 0.001, thus completing the transfer learning initialization of the model framework.

[0019] Furthermore, in S300, during model training and prescription generation, the model parameter update formula for the personalized adaptive incremental training algorithm is as follows: ,in, For the first Model parameters after the second data collection For the first Model parameters after the second data collection This is a personalized amnesia factor, dynamically adjusted based on the stability of the patient's historical data. The novelty coefficient for new data is calculated based on the difference between the new data and the old data. For adaptive learning rate, The gradient of the cross-entropy loss function. It is the final output cross-modal joint feature vector. A map of brain function-molecular mechanisms. For diagnostic labels, For the first The effective data volume of each collection This serves as the baseline value for the amount of data.

[0020] Furthermore, in S400, during the execution of the collaborative intervention, the mathematical expression of the collaborative intervention parameter optimization algorithm is: The constraints are: kPa ,in, The ultimate goal is to find the optimal combination of parameters. For mathematical operators, find the independent variable that maximizes the function. ) value, To achieve optimal neural regulation parameters, To achieve the optimal stimulation frequency, For optimal stimulus intensity, For the optimal stimulus duration, To achieve the optimal drug dosage, , These are the weighting coefficients. The amount of improvement in abnormal EEG characteristics, To intervene in the total duration, This refers to the fluctuation of physiological parameters. This represents the decrease in molecular marker concentration. This represents the actual adhesion pressure of the hydrogel electrode. The skin temperature at the electrode contact area.

[0021] Furthermore, in S500, during the closed-loop optimization update, the mathematical expression for the closed-loop effect feedback optimization algorithm is: ,in, For parameter update amount, To optimize the coefficients, The synergistic effect index is determined based on the degree of synergistic change in multimodal data before and after the intervention. To form a joint feature vector across modalities after intervention, To intervene in the joint feature vector of cross-modal approaches.

[0022] On the other hand, a power distribution system based on the aforementioned power big data automatic reasoning platform includes:

[0023] Data acquisition module: Composed of semi-dry EEG electrodes and matching acquisition equipment, it is used to fit the target functional area of ​​the patient's brain and cyclically collect EEG signals, physiological parameters, molecular expression imaging data and behavioral feature data to form a multi-dimensional dynamic dataset.

[0024] The data preprocessing and fusion module is used to perform denoising, normalization, and spatial registration preprocessing on the collected multimodal data, and to generate cross-modal joint feature vectors through a cross-modal spatiotemporal attention fusion algorithm.

[0025] Model training and prescription generation module: used to initialize the model framework based on transfer learning, iteratively update the model parameters through a personalized adaptive incremental training algorithm, construct a personalized EEG feature map, and output disease diagnosis results and personalized neuromodulation prescriptions;

[0026] Collaborative intervention execution module: Used to detect electrode adhesion status through sensors, determine optimal neuromodulation parameters and drug dosage based on collaborative intervention parameter optimization algorithm, execute transcranial electrical stimulation and drug delivery, and continuously collect monitoring data;

[0027] Closed-loop optimization and update module: It is used to collect post-intervention data at preset time nodes, supplement the dynamic dataset, and adjust the model parameters, intervention parameters and collection frequency through closed-loop effect feedback optimization algorithm to form a cyclic optimization link.

[0028] Compared with existing technologies, this method and system for reversibly adhering hydrogel brain electrodes to medical data processing has the following advantages:

[0029] I. This invention integrates reversible adhesive electrodes with multimodal dynamic acquisition technology to comprehensively acquire EEG signals, physiological parameters, molecular expression imaging data, and behavioral feature data. After denoising, normalization, and spatial registration preprocessing, a cross-modal spatiotemporal attention fusion algorithm is used to extract key features from each modality and weighted splicing them to construct a brain function-molecular mechanism correlation map. This multi-dimensional data integration mode breaks through the information limitations of single-modal data, deeply explores the intrinsic correlation between EEG features and molecular expression patterns, physiological states, and behavioral performance, and allows disease diagnosis to break free from reliance on isolated indicators. By accurately mapping the correlation strength between brain functional areas and molecular mechanisms, it clearly presents the individual specificity of disease pathological mechanisms, providing a comprehensive and accurate basis for diagnosis, significantly improving the specificity and accuracy of disease identification, laying a solid foundation for subsequent personalized intervention, and solving the problems of fragmented information and insufficient targeting in traditional diagnosis.

[0030] II. This invention initializes the model framework through transfer learning, iteratively updates model parameters using a personalized adaptive incremental training algorithm, constructs a dynamically evolving personalized EEG feature atlas, accurately outputs disease diagnosis results and customized neuromodulation prescriptions, and achieves synergistic coordination between transcranial electrical stimulation and drug delivery through a collaborative intervention parameter optimization algorithm. Simultaneously, it monitors electrode adhesion status and intervention process data in real time using sensors, and continuously integrates post-intervention data to adjust model parameters, intervention plans, and acquisition frequency using a closed-loop effect feedback optimization algorithm, forming a full-process cyclical optimization link. This combination of personalization and dynamic optimization ensures that the intervention plan aligns with individual pathological characteristics and physical condition, while also responding in real time to changes in intervention effects, improving the targeting and safety of the intervention. It effectively solves the problems of traditional intervention plans lacking personalization and lagging adjustments, promoting the development of neuromodulation therapy towards precision and dynamism, and significantly improving treatment outcomes and patient experience.

[0031] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0033] Figure 1 A framework diagram of a reversible adhesion medical data processing method for hydrogel brain electrodes;

[0034] Figure 2 A flowchart for a reversible adhesion medical data processing system for hydrogel brain electrodes. Detailed Implementation

[0035] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0036] Example 1:

[0037] Example of medical data processing for Alzheimer's disease patients.

[0038] Multimodal dynamic acquisition: A semi-dry brain electrode made of PVA and PAM networks in a 3:1 mass ratio was used, with 5% of the total mass of the hydrogel added as an electroresponsiveness enhancer. After four freeze-thaw cycles, the electrode was molded. The electrode has a tensile strength of 1580%, a tensile strength of 320 kPa, and reversible electroadhesion for ≥15 cycles, with complete release of adhesion within 25 seconds after power deactivation. The electrode integrates a micro-programmable transcranial AC stimulation array. Its high tensile strength and elasticity allow it to adapt to dynamic changes in brain contours, and its reversible adhesion characteristics support long-term cyclic acquisition, avoiding frequent electrode changes that could irritate or cause discomfort to the patient's scalp. Based on the lesion-related regions located by the patient's brain MRI, the electrode was precisely attached to the parietal lobe and hippocampus, two target functional areas closely related to the pathological changes of Alzheimer's disease. Data is collected cyclically using a combination of an Ag / AgCl EEG sensor, a miniature piezoresistive pulse sensor, an NTC thermistor temperature sensor, a CdSe / ZnS core-shell quantum dot labeled probe, a near-infrared fluorescence imaging system, and a wearable accelerometer. The Ag / AgCl EEG sensor captures brain electrical activity signals, reflecting the activity state of neurons. The miniature piezoresistive pulse sensor and the NTC thermistor temperature sensor monitor physiological parameters such as pulse rate and skin temperature to understand the patient's basic physical condition. The CdSe / ZnS core-shell quantum dot labeled probe, combined with the near-infrared fluorescence imaging system, obtains data on the expression concentration and spatial distribution of specific molecular markers in the brain, revealing pathological changes at the molecular level. The wearable accelerometer records behavioral characteristics such as walking speed, limb tremor frequency, and limb movement amplitude, reflecting the patient's actual functional performance. Multi-dimensional data collectively form a dynamic dataset, providing comprehensive and three-dimensional information support for subsequent diagnosis and intervention. Electrode placement strictly follows the lesion-associated regions located on brain MRI, ensuring precise coverage of pathologically relevant areas in the parietal lobe and hippocampus. An integrated micro-programmable transcranial AC stimulation array is synchronously ready, providing hardware support for the transcranial electrical stimulation component in subsequent collaborative interventions. During data acquisition, the acquisition time intervals of each modality are recorded simultaneously, providing temporal parameters for cross-modal spatiotemporal attention fusion algorithms, such as... Figure 1 The diagram shown illustrates the framework of the reversible adhesion hydrogel brain electrode medical data processing method, which intuitively presents the entire process of multimodal dynamic acquisition, data preprocessing and fusion, model training and prescription generation, collaborative intervention execution, and closed-loop optimization and updating: First, multi-dimensional data is collected using reversible adhesion electrodes. After processing and fusion, the model is trained to output diagnosis and prescription. Then, collaborative intervention is executed. Finally, the parameters are adjusted based on the data after intervention, forming a complete loop of cyclic optimization.

[0039] Data Preprocessing and Fusion: For the four modalities of collected EEG, physiological, molecular expression imaging, and behavioral feature data, denoising, normalization, and spatial registration preprocessing were performed respectively. Denoising removed irrelevant signals such as environmental electromagnetic interference and electromyographic interference, ensuring the purity of EEG and other data. Normalization unified the dimensions of data from different modalities, avoiding the impact of data range differences on subsequent feature fusion results. Spatial registration accurately mapped each modal data to brain anatomical structures, ensuring consistency in spatial location and laying an accurate foundation for cross-modal data correlation analysis. Based on brain anatomical structures, functional brain regions—prefrontal, parietal, temporal, occipital, and hippocampal—were defined, and the spatial coordinate range of each region was determined using brain MRI images. From the preprocessed molecular expression imaging data, the expression concentration, spatial distribution range, and expression intensity gradient data of specific molecular markers within each functional brain region were extracted. By calculating the correlation coefficient between the core EEG features of each functional brain region and the corresponding molecular marker expression data, the correlation strength between the two was quantified, clearly presenting the correspondence between brain function changes and molecular mechanisms, providing a reference for understanding the molecular roots of brain dysfunction. A two-dimensional atlas framework was constructed with brain functional regions as the horizontal axis and molecular marker types as the vertical axis. Association strengths were filled using numerical annotations or color gradients, and spatial association matrix information was incorporated to adjust association weights, resulting in a structurally complete and quantitatively accurate brain function-molecular mechanism association atlas. Subsequently, a cross-modal spatiotemporal attention fusion algorithm was employed. This algorithm automatically identifies and highlights key features relevant to disease diagnosis in each modality, while weakening irrelevant information. By weighted concatenation of key features from each modality, a cross-modal joint feature vector is generated. The mathematical expression of the cross-modal spatiotemporal attention fusion algorithm is as follows: ,in, It is the final output cross-modal joint feature vector. Corresponding to four modalities: electroencephalography (EEG), molecular imaging, physiological, and behavioral. For modal baseline weights, For modal reliability factor, For each modal feature vector, It is a spatial incidence matrix. This is the time-series decay coefficient. To collect data at different time intervals, integrate multi-dimensional information, avoid the limitations of single-modal data, and provide comprehensive feature support for subsequent model training.

[0040] Model Training and Prescription Generation: The PhysioNet EEG public dataset was selected as the pre-training dataset. The basic model used was a ResNet-18 network containing 16 convolutional layers, 4 pooling layers, and 3 fully connected layers. Transfer learning was performed to initialize the model framework: The parameters of the 16 convolutional layers and 4 pooling layers trained on the public dataset in the pre-training model were retained. These parameters contain general features of brain electrical signal processing, which can reduce the computational cost of training the model from scratch, speed up the initialization, and improve the model's basic ability to extract EEG features. The original fully connected layers were replaced with a 3-layer adaptive structure. The first layer contains 1024 neurons, the second layer contains 512 neurons, and the third layer contains 64 neurons. The output layer uses the Softmax activation function to adapt to the classification task of 3 disease types and 3 risk levels, meeting the specific disease diagnosis needs. The He initialization method was used to initialize the parameters of the replaced fully connected layers to ensure that the parameters of the new layers are reasonably distributed and to avoid gradient vanishing or exploding in the early stage of training. The initial learning rate was set to 0.001 to balance the model convergence speed and training stability, thus completing the model framework initialization. EEG time-domain features, frequency-domain features, and correlation features between EEG signals and molecular marker expressions are selected from cross-modal joint feature vectors. Based on the model output initialized by transfer learning, a brain functional zoning atlas framework including the prefrontal, parietal, temporal, occipital, and hippocampal regions is established by combining brain anatomical structural zoning. The feature values ​​corresponding to the EEG data acquired in each adhesion-acquisition cycle are filled into the atlas framework according to brain functional zoning to form an initial personalized EEG feature atlas. The model parameters are updated using a personalized adaptive incremental training algorithm. The model parameter update formula for the personalized adaptive incremental training algorithm is as follows: ,in, For the first Model parameters after the second data collection For the first Model parameters after the second data collection For personalized forgetting factors, The novelty coefficient of the new data. For adaptive learning rate, The gradient of the cross-entropy loss function. It is the final output cross-modal joint feature vector. A map of brain function-molecular mechanisms. For diagnostic labels, For the first The effective data volume of each collection Using the data volume as a baseline, the feature weights and correlation strengths of each brain functional region are adjusted to enhance the representation of patient-specific EEG features and avoid the general model's neglect of individual differences. The atlas is updated synchronously according to the number of adhesion-acquisition cycles, recording the dynamic changes of EEG features in each region at different time points, forming a complete personalized EEG feature atlas, providing accurate basis for personalized diagnosis and prescription. The cross-modal joint feature vector, personalized EEG feature atlas, and brain function-molecular mechanism correlation atlas are synchronously input into the model optimized by the personalized adaptive incremental training algorithm: the model strengthens the correlation mapping between patient-specific EEG features and molecular expression rules through the feature extraction network, improving the understanding of individual patient conditions. After classification calculation by fully connected layers, the model outputs disease diagnosis results that are more consistent with the patient's actual situation. Based on the pathological mechanism correlation in the diagnostic results, the expression patterns of molecular markers in the abnormal EEG features and corresponding brain functional areas are extracted. Combined with the brain functional area localization results, the core parameters of the personalized neuromodulation prescription are determined: the stimulation target is the brain functional area corresponding to the abnormal EEG features to ensure that the stimulation directly acts on the lesion-related area; the stimulation frequency matches the characteristics of the abnormal EEG band to enhance the regulatory effect of stimulation on abnormal EEG; the stimulation intensity is adapted according to the individual's skin impedance and EEG signal amplitude to avoid discomfort caused by excessive intensity or the effect being affected by insufficient intensity; the stimulation duration is correlated with the expression concentration of molecular markers to match the intervention with the changes at the molecular level; the above parameters are integrated to form a complete personalized neuromodulation prescription, which is output to the system terminal simultaneously with the disease diagnosis results to provide precise instruction support for subsequent collaborative intervention.

[0041] Collaborative intervention execution: The adhesion status of the hydrogel electrodes is monitored in real time using sensors integrated into the electrodes, simultaneously monitoring the skin temperature at the electrode contact area. This ensures that the actual adhesion pressure of the hydrogel electrodes is ≥5 kPa, guaranteeing stable adhesion between the electrodes and the brain skin and preventing electrode detachment during stimulation, which could affect the intervention effect. It also ensures that the skin temperature is ≤38℃ to prevent overheating due to prolonged stimulation, ensuring patient safety during the intervention. A collaborative intervention parameter optimization algorithm is employed, comprehensively considering the ratio of improvement in EEG abnormalities to total intervention duration, and the ratio of physiological parameter fluctuations to decreases in molecular marker concentrations. This, combined with weighted coefficients, balances the intervention effect with patient physiological stability to determine the optimal neuromodulation parameters and drug dosage. The mathematical expression for the collaborative intervention parameter optimization algorithm is: The constraints are: kPa ,in, To achieve optimal neural regulation parameters, To achieve the optimal stimulation frequency, For optimal stimulus intensity, For the optimal stimulus duration, To achieve the optimal drug dosage, , These are the weighting coefficients. The amount of improvement in abnormal EEG characteristics, To intervene in the total duration, This refers to the fluctuation of physiological parameters. This represents the decrease in molecular marker concentration. This represents the actual adhesion pressure of the hydrogel electrode. The optimal neuromodulation parameters, including optimal stimulation frequency, intensity, and duration, are used to measure the skin temperature at the electrode contact area. These parameters ensure synergistic effects between electrical stimulation and drug delivery, enhancing intervention efficacy while minimizing adverse impacts on the patient's physiological state. Transcranial electrical stimulation and drug delivery are performed sequentially, with continuous data collection and monitoring throughout the process. This allows for real-time monitoring of changes in the patient's EEG, physiological, molecular, and behavioral characteristics, providing a basis for subsequent efficacy evaluation and parameter adjustments.

[0042] Closed-loop optimization and update: Multimodal data on patients after intervention are collected again at preset time points and added to the dynamic dataset. This enriches the dataset's coverage of the post-intervention state, enabling the model to adjust its understanding of the patient's condition based on the latest data. A closed-loop effect feedback optimization algorithm is employed. By comparing the differences between the cross-modal joint feature vectors after intervention and those before intervention, and combining the optimization coefficient and synergistic effect index, the parameter update amount is calculated to adjust model parameters, intervention parameters, and data collection frequency. The mathematical expression of the closed-loop effect feedback optimization algorithm is: ,in, For parameter update amount, To optimize the coefficients, The synergistic effect index, To form a joint feature vector across modalities after intervention, The system incorporates cross-modal joint feature vectors before intervention; optimized model parameters enhance adaptability to post-intervention data; adjusted intervention parameters ensure subsequent interventions better align with changes in the patient's condition after intervention; and updated collection frequency optimizes data collection efficiency based on the rhythm of disease progression, avoiding unnecessary frequent collections or data gaps, thus forming a cyclical optimization loop to continuously improve diagnostic accuracy and intervention effectiveness.

[0043] In summary, this embodiment focuses on the medical data processing of Alzheimer's disease patients, forming a complete workflow through five steps: First, semi-dry EEG electrodes with specific parameters are attached to the parietal lobe and hippocampus, and multi-dimensional data is collected using various devices to provide comprehensive information support for diagnosis; then, through preprocessing and a cross-modal spatiotemporal attention fusion algorithm, joint feature vectors and brain function-molecular mechanism correlation maps are generated to improve data effectiveness; next, a model is built based on transfer learning and personalized adaptive incremental training algorithms to output accurate diagnosis and personalized neuromodulation prescriptions; subsequently, safe intervention is performed through sensor monitoring and collaborative intervention parameter optimization algorithms; finally, a closed-loop effect feedback optimization algorithm is used to cyclically adjust parameters to continuously improve diagnostic accuracy and intervention effects, adapting to the dynamic changes in the patient's condition.

[0044] Example 2:

[0045] An example of a medical data processing system for epilepsy patients.

[0046] Data Acquisition Module: This module consists of semi-dry EEG electrodes and supporting acquisition equipment. Its core function is to acquire multi-dimensional dynamic data of epilepsy patients. The core conductive layer of the semi-dry EEG electrodes is a double-network interpenetrating structure formed by combining PVA and PAM networks at a mass ratio of 2.5:1. An electroresponsive property enhancer, accounting for 4% of the total hydrogel mass, is added, and the electrode is formed after three freeze-thaw cycles. The electrode has a tensile strength of 1350%, a tensile strength of 290 kPa, and reversible electroadhesion cycles of ≥16 times. Adhesion is released within 28 seconds after power is cut off. It also integrates a micro-programmable transcranial AC stimulation array. The high tensile strength and tensile strength allow it to adapt to the subtle brain movements during epileptic seizures, closely adhering to the temporal and frontal lobes. The reversible adhesion characteristics support long-term cyclic acquisition, reducing scalp stimulation from frequent electrode changes and minimizing patient resistance. The accompanying data acquisition equipment includes an Ag / AgCl EEG sensor, a miniature piezoresistive pulse sensor, an NTC thermistor temperature sensor, a CdSe / ZnS core-shell quantum dot labeled probe, a near-infrared fluorescence imaging system, and a wearable accelerometer. The Ag / AgCl EEG sensor accurately captures abnormal brain electrical discharge signals related to epileptic seizures, providing core electrophysiological evidence for diagnosis. The miniature piezoresistive pulse sensor and the NTC thermistor temperature sensor monitor the patient's pulse rate and skin temperature in real time, allowing for the assessment of the patient's basic physiological state during intervention. The CdSe / ZnS core-shell quantum dot labeled probe, combined with the near-infrared fluorescence imaging system, clearly acquires data on the expression concentration and spatial distribution of specific molecular markers related to epilepsy in the brain, revealing pathological changes at the molecular level. The wearable accelerometer conveniently records behavioral characteristics such as the patient's walking speed, limb tremor frequency, and limb movement amplitude, reflecting the patient's daily functional status. The module collects the aforementioned multi-dimensional data in a loop to form a dynamic dataset, providing comprehensive and continuous raw data input for subsequent modules. This avoids diagnostic biases caused by single data types. During the collection process, the time interval of each data collection is precisely recorded, and the collection duration and timestamp of each modality are stored synchronously. This provides the basic data required for calculating the temporal decay coefficient of the cross-modal spatiotemporal attention fusion algorithm in the subsequent data preprocessing and fusion modules, ensuring the temporal accuracy of cross-modal feature fusion. Figure 2 The flowchart of the medical data processing system shown clearly illustrates the collaborative relationship between the five modules: the data acquisition module collects multimodal data and transmits it to the preprocessing and fusion module; the fused data is input into the model training and prescription generation module to output diagnoses and prescriptions; the collaborative intervention execution module implements interventions according to the prescriptions; and the closed-loop optimization and update module uses the intervention data to adjust the parameters of each module to ensure that the system dynamically adapts to the needs of diagnosis and treatment.

[0047] Data Preprocessing and Fusion Module: As the core component of the system's data processing, this module receives multimodal data transmitted from the data acquisition module and first performs denoising, normalization, and spatial registration preprocessing. Denoising removes irrelevant signals such as environmental electromagnetic interference and patient electromyography (EMG) interference, ensuring the accuracy of core data such as EEG. Normalization unifies the dimensions of different modalities, preventing data range differences from affecting the fairness of feature fusion. Spatial registration precisely aligns each modality of data with the anatomical structures determined by brain MRI, ensuring the consistency of molecular expression imaging data and EEG data in spatial location, laying the foundation for subsequent correlation analysis. Subsequently, the module defines functional brain regions—prefrontal, parietal, temporal, occipital, and hippocampal—based on brain anatomy, determines the spatial coordinate range of each region using MRI images, extracts expression information of molecular markers from the preprocessed molecular expression imaging data, and constructs and corrects a brain function-molecular mechanism correlation map by quantifying the correlation strength between core EEG features and molecular markers. This clearly presents the correspondence between abnormal brain function and molecular changes, providing an intuitive reference for the model to understand the molecular mechanisms of epilepsy pathogenesis. The final module employs a cross-modal spatiotemporal attention fusion algorithm. This algorithm automatically identifies key features related to epilepsy diagnosis and intervention in each modality, weakens irrelevant information, and generates a cross-modal joint feature vector through weighted concatenation. The mathematical expression of the cross-modal spatiotemporal attention fusion algorithm is as follows: ,in, It is the final output cross-modal joint feature vector. Corresponding to four modalities: electroencephalography (EEG), molecular imaging, physiological, and behavioral. For modal baseline weights, For modal reliability factor, For each modal feature vector, It is a spatial incidence matrix. This is the time-series decay coefficient. To establish a data collection time interval, multi-dimensional information from electrophysiology, physiology, molecular dynamics, and behavior is integrated to avoid the limitations of single-modality data and provide high-quality, high-dimensional feature data support for model training and prescription generation modules.

[0048] Model Training and Prescription Generation Module: This module incorporates a model initialization framework based on transfer learning. The PhysioNet EEG dataset is selected as the pre-training dataset. The base model is a ResNet-18 network containing 16 convolutional layers, 4 pooling layers, and 3 fully connected layers. The parameters of the 16 convolutional layers and 4 pooling layers in the pre-trained model are retained; these parameters contain general EEG feature extraction capabilities, reducing the computational cost of training the model from scratch and accelerating the initialization speed. Simultaneously, the broad representativeness of the public dataset improves the model's basic performance. The original fully connected layers are replaced with a 3-layer adaptive structure, and the output layer uses the Softmax activation function to adapt to classification tasks involving 3 disease types and 3 risk levels, meeting the specific needs of epilepsy diagnosis. The He initialization method is used to initialize the parameters of the new fully connected layers, ensuring a reasonable parameter distribution and avoiding gradient problems in the early stages of training. The initial learning rate is set to 0.001 to balance model convergence speed and training stability, completing the model framework initialization. After receiving the cross-modal joint feature vector and the brain function-molecular mechanism association map, the module selects EEG time-domain features, frequency-domain features, and EEG-molecular association features from the feature vector. Combined with the output of the transfer learning model, it establishes a brain function zoning map framework, filling in the EEG features collected each time according to the zoning to form an initial personalized EEG feature map. The model parameters are iteratively updated through a personalized adaptive incremental training algorithm. The model parameter update formula for the personalized adaptive incremental training algorithm is: ,in, For the first Model parameters after the second data collection For the first Model parameters after the second data collection For personalized forgetting factors, The novelty coefficient of the new data. For adaptive learning rate, The gradient of the cross-entropy loss function. It is the final output cross-modal joint feature vector. A map of brain function-molecular mechanisms. For diagnostic labels, For the first The effective data volume of each collection Using the data volume as a baseline, the feature weights and correlation strengths of each partition are adjusted to enhance the representation of individual patient-specific epilepsy EEG characteristics. The atlas is updated synchronously according to the acquisition cycle, recording feature trajectories at different time points to form a complete personalized EEG feature atlas, accurately capturing individual patient differences. The module inputs the above data into the optimized model, which strengthens individual-specific correlation mapping through a feature extraction network, improving the accuracy of epilepsy diagnosis results. Based on the pathological mechanism description in the diagnosis results, combined with brain functional partition localization, the core parameters of neuromodulation are determined, integrated to form a personalized neuromodulation prescription, and output synchronously with the diagnosis results to the collaborative intervention execution module, providing precise instructions for intervention.

[0049] Collaborative Intervention Execution Module: This module is equipped with an electrode adhesion status sensor, a transcranial electrical stimulation execution component, and a drug delivery component. Its core function is to safely and accurately execute intervention procedures. After receiving the diagnostic results and neuromodulation prescription, the module uses integrated sensors to monitor the actual adhesion pressure of the hydrogel electrodes and the skin temperature of the electrode contact area in real time. Adhesion pressure monitoring ensures stable electrode contact, preventing detachment during stimulation that could interrupt the intervention or weaken its effect. Skin temperature monitoring prevents skin burns caused by prolonged electrical stimulation, ensuring patient safety. Subsequently, the module employs a collaborative intervention parameter optimization algorithm, comprehensively considering the improvement in abnormal EEG features, total intervention duration, physiological parameter fluctuations, and decrease in molecular marker concentrations. This is combined with weighted coefficients to balance the intervention effect and the patient's physiological stability, determining the optimal neuromodulation parameters and drug dosage. The mathematical expression for the collaborative intervention parameter optimization algorithm is: The constraints are: kPa ,in, The ultimate goal is to find the optimal combination of parameters. For mathematical operators, find the independent variable that maximizes the function. ) value, To achieve optimal neural regulation parameters, To achieve the optimal stimulation frequency, For optimal stimulus intensity, For the optimal stimulus duration, To achieve the optimal drug dosage, , These are the weighting coefficients. The amount of improvement in abnormal EEG characteristics, To intervene in the total duration, This refers to the fluctuation of physiological parameters. This represents the decrease in molecular marker concentration. This represents the actual adhesion pressure of the hydrogel electrode. The skin temperature at the electrode contact area is monitored to ensure synergistic effects between electrical stimulation and medication, enhancing the suppression of abnormal epileptic discharges while minimizing drug side effects and interference with physiological states. The module control execution component sequentially performs transcranial electrical stimulation and drug delivery, continuously collecting EEG, physiological, molecular, and behavioral monitoring data throughout the process. Real-time feedback on changes in the patient's condition during intervention provides data support for closed-loop optimization and updating of the module.

[0050] Closed-loop optimization and update module: As the core of continuous system optimization, this module triggers the data acquisition module at preset time nodes to collect multimodal data of patients after intervention, supplementing the dynamic dataset with new data, enriching the dataset's coverage of post-intervention states, and enabling the system to adjust strategies based on the latest changes in the patient's condition. The module employs a closed-loop effect feedback optimization algorithm, comparing the cross-modal joint feature vectors before and after intervention, and calculating the parameter update amount by combining the optimization coefficient and the synergistic effect index. The mathematical expression of the closed-loop effect feedback optimization algorithm is: ,in, For parameter update amount, To optimize the coefficients, The synergistic effect index, To form a joint feature vector across modalities after intervention, To obtain the cross-modal joint feature vector before intervention, adjustment instructions are sent to the data acquisition module, data preprocessing and fusion module, model training and prescription generation module, and collaborative intervention execution module, respectively: adjusting the data acquisition frequency to adapt the acquisition rhythm to changes in the patient's condition; optimizing model parameters to improve the model's adaptability to post-intervention data and diagnostic accuracy; revising intervention parameters to make subsequent stimulation and drug dosages more consistent with the patient's post-intervention pathological state; and updating the preprocessing and fusion strategy to ensure that data processing always matches the latest data features. Through cyclical adjustment of parameters across multiple modules, a closed-loop optimization chain is formed, continuously improving the system's diagnostic accuracy and intervention effect, and adapting to the dynamic changes in the patient's condition.

[0051] In summary, this embodiment of the medical data processing system for epilepsy patients relies on the coordinated operation of five modules: the data acquisition module uses specialized hydrogel electrodes and multiple devices to accurately collect multi-dimensional data related to the temporal and frontal lobes, reducing patient discomfort; the data preprocessing and fusion module generates high-quality features and correlation maps through preprocessing operations and cross-modal spatiotemporal attention fusion algorithms; the model training and prescription generation module outputs personalized diagnoses and prescriptions based on transfer learning models and incremental algorithms; the collaborative intervention execution module monitors adhesion and temperature, and achieves safe collaborative intervention through optimization algorithms; and the closed-loop optimization and update module supplements data at each node, adjusts the parameters of multiple modules through algorithms, forms a cyclical optimization link, ensures that the system continuously adapts to the condition of epilepsy patients, and improves the diagnosis and treatment effect.

[0052] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A hydrogel electrode reversible adhesion medical data processing method, characterized in that, The specific steps of the method are: S100, multi-modal dynamic acquisition: a semi-dry EEG electrode integrated with a micro-programmable transcranial alternating current stimulation array is attached to the target functional area of the patient's brain, the core conductive layer of the semi-dry EEG electrode is a double-network interpenetrating structure conductive hydrogel composed of PVA network and PAM network, and a continuous direct current voltage is applied to the conductive hydrogel; through the electrode and the matching acquisition equipment, the electroencephalogram signal, the physiological parameter, the molecular expression imaging data and the behavior characteristic data are cyclically acquired to form a multi-dimensional dynamic data set; S200, data preprocessing and fusion: the collected multi-modal data are respectively preprocessed by denoising, normalization and spatial registration, the cross-modal spatio-temporal attention fusion algorithm is used to extract key features of each modality and to weight and splice them to generate a cross-modal joint feature vector and a brain function-molecular mechanism correlation graph; S300, model training and prescription generation: based on the transfer learning to initialize the model framework, the individualized adaptive incremental training algorithm is used to iteratively update the model parameters using the dynamic data set, to construct an individualized electroencephalogram feature map, and to output a disease diagnosis result and an individualized neuromodulation prescription; S400, collaborative intervention execution: the sensor integrated with the electrode detects the adhesion state, the collaborative intervention parameter optimization algorithm is used to determine the optimal neuromodulation parameters and drug dosage, and the transcranial electrical stimulation and drug delivery are executed in turn, and the monitoring data are continuously acquired throughout the process; S500, closed-loop optimization and update: the post-intervention data are collected again at the preset time node, supplemented to the dynamic data set, the parameter update amount is calculated by the closed-loop effect feedback optimization algorithm, the model parameters, the intervention parameters and the acquisition frequency are adjusted, and a cyclic optimization link is formed.

2. The medical data processing method of hydrogel electrode reversible adhesion according to claim 1, wherein, In the S100, multi-modal dynamic acquisition, the conductive hydrogel of the semi-dry EEG electrode is composed of PVA network and PAM network in a mass ratio of 2:1-3:1 to form a double-network interpenetrating structure, and an electric response characteristic enhancer accounts for 3%-6% of the total mass of the hydrogel. After 3-5 freeze-thaw cycles, the tensile rate is ≥1200%, the tensile strength is ≥280kPa, the electric adhesion reversible times are ≥15, and the adhesion force is released within 30s after power-off; the target functional area of the patient's brain includes the parietal lobe and the hippocampal region of Alzheimer's disease patients, the temporal lobe and the frontal lobe of epilepsy patients, and the lesion associated area determined by brain MRI positioning; the matching acquisition equipment includes Ag / AgCl EEG sensor, micro piezoresistive pulse sensor, NTC thermistor temperature sensor, CdSe / ZnS core-shell structure quantum dot marker probe, near-infrared fluorescence imaging system and wearable acceleration sensor.

3. The method of claim 1, wherein the hydrogel electrode reversible adhesion is achieved by, In the S100, multi-modal dynamic acquisition, the acquired electroencephalogram signal includes brain electrical activity signal, the physiological parameter includes pulse frequency and skin temperature, and the molecular expression imaging data includes expression concentration data and spatial distribution data of specific molecular markers in the brain, and the behavior characteristic data includes walking speed, limb tremor frequency and limb activity amplitude data.

4. The method of claim 1, wherein the hydrogel-based electrode reversible adhesion is used for medical data processing. The mathematical expression of the cross-modal spatio-temporal attention fusion algorithm in the data preprocessing and fusion S200 is: Wherein, is the final output cross-modal joint feature vector, Corresponding to four modes of electroencephalogram, molecular imaging, physiology and behavior, is the mode reference weight, is the mode reliability factor, is the feature vector of each mode, is the spatial correlation matrix, is the time attenuation coefficient, is the collection time interval.

5. The medical data processing method of hydrogel electrode reversible adhesion according to claim 1, characterized in that, The specific steps of constructing the individualized electroencephalogram feature map in the S300, model training and prescription generation, are as follows: screening electroencephalogram-related core features from the cross-modal joint feature vector, including electroencephalogram time domain features, frequency domain features, and the correlation features between electroencephalogram signals and molecular marker expression; Based on the model output initialized by transfer learning, combined with brain anatomical structure partitioning, a brain function partitioning map framework is established, including the prefrontal lobe, parietal lobe, temporal lobe, occipital lobe, and hippocampal region; the feature values corresponding to the electroencephalogram data obtained in each adhesion-capture cycle are filled into the map framework according to the brain function partitioning, forming an initial individualized electroencephalogram feature map; Using the model parameters updated by the individualized adaptive incremental training algorithm, the feature weights and correlation strengths of each brain function partition are corrected, and the representation of individual-specific electroencephalogram features is strengthened; the map is updated synchronously according to the number of adhesion-capture cycles, and the dynamic change trajectories of electroencephalogram features in each partition at different time nodes are recorded, forming a complete individualized electroencephalogram feature map.

6. The hydrogel electrode reversible adhesion medical data processing method according to claim 1, characterized in that, The specific content of initializing the model framework by transfer learning in the S300, model training and prescription generation, is as follows: selecting the pre-training data set as the PhysioNet electroencephalogram public data set; the base model uses the ResNet-18 network, which includes 16 convolutional layers, 4 pooling layers, and 3 fully connected layers, where the convolutional layers are used for feature extraction, and the fully connected layers are used for classification output; the parameters of the 16 convolutional layers and the 4 pooling layers in the pre-trained model trained with the public data set are retained as the initial weights for feature extraction; Replace the original fully connected layer with a 3-layer adaptive structure, the first layer contains 1024 neurons, the second layer contains 512 neurons, and the third layer contains 64 neurons, and the output layer uses the Softmax activation function, which adapts to the classification task of 3 disease types and 3 risk levels; the He initialization method is used to initialize the parameters of the replaced fully connected layer, and the initial learning rate is set to 0.001, completing the transfer learning initialization of the model framework.

7. The hydrogel electrode reversible adhesion medical data processing method according to claim 1, wherein, The model parameter update formula of the personalized adaptive incremental training algorithm in the model training and prescription generation is: wherein, is the model parameter after the first collection, is the model parameter after the first collection, is the model parameter after the first collection, is the model parameter after the first collection, is the personalized forgetting factor, is the new data novelty coefficient, is the adaptive learning rate, is the cross-entropy loss function gradient, is the final output cross-modal joint feature vector, is the brain function-molecular mechanism correlation graph, is the diagnostic label, is the effective data amount of the first collection, is the effective data amount of the first collection, is the data amount reference value.

8. The hydrogel electrode reversible adhesion medical data processing method according to claim 1, wherein, The S400, in cooperation with the intervention execution, the mathematical expression of the cooperative intervention parameter optimization algorithm is: ; the constraint condition is kPa, , wherein is the optimal parameter combination to be finally found, is a mathematical operator, and the value of the independent variable that makes the function reach the maximum value is found , wherein is the optimal neuromodulation parameter, is the optimal stimulation frequency, is the optimal stimulation intensity, is the optimal stimulation duration, is the optimal drug dose, , is a weight coefficient, is the improvement amount of the abnormal characteristics of the brain electrical activity, is the total duration of the intervention, is the physiological parameter fluctuation amount, is the molecular marker concentration reduction amount, is the actual adhesion pressure of the hydrogel electrode, is the skin temperature of the electrode attachment area.

9. The hydrogel-based electrode-reversible-adhesion medical data processing method according to claim 1, wherein, The S500, in the closed loop optimization update, the mathematical expression of the closed loop effect feedback optimization algorithm is: wherein, is a parameter update amount, is an optimization coefficient, is a synergistic effect index, is a cross-modal joint feature vector after intervention, is a cross-modal joint feature vector before intervention.

10. A hydrogel brain electrode reversible adhesion medical data processing system, which is suitable for a hydrogel brain electrode reversible adhesion medical data processing method according to any one of claims 1-9, characterized in that, The system comprises: A data acquisition module composed of a semi-dry electroencephalogram electrode and a matching acquisition device, the conductive layer of the semi-dry electroencephalogram electrode is a double-network interpenetrating structure conductive hydrogel composed of PVA network and PAM network, which has electrical response characteristics and realizes electrical adhesion with the scalp by applying a continuous direct current voltage; it is used to adhere to the target functional area of the patient's brain, and cyclically acquires electroencephalogram signals, physiological parameters, molecular expression imaging data, and behavior characteristic data, forming a multi-dimensional dynamic data set; A data preprocessing and fusion module for performing denoising, normalization, and spatial registration preprocessing on the collected multi-modal data, and generating a cross-modal joint feature vector through a cross-modal spatio-temporal attention fusion algorithm; A model training and prescription generation module for initializing the model framework based on transfer learning, iteratively updating the model parameters through an individualized adaptive incremental training algorithm, constructing an individualized electroencephalogram feature map, and outputting disease diagnosis results and individualized neuroregulation prescriptions; Synergistic intervention execution module: for detecting the electrode adhesion state through the sensor, determining the optimal neuromodulation parameters and drug dose based on the synergistic intervention parameter optimization algorithm, executing transcranial electrical stimulation and drug delivery, and continuously collecting monitoring data; Closed-loop optimization update module: for collecting post-intervention data at preset time nodes, supplementing to the dynamic data set, adjusting the model parameters, intervention parameters and collection frequency through the closed-loop effect feedback optimization algorithm, and forming a cyclic optimization link.

Images