Peripheral pressure stimulation quantitative evaluation method and system based on AI

Abstract

The invention provides an AI-based peripheral pressure stimulation quantitative evaluation method and system. Relates to the technical field of digital medical treatment and artificial intelligence, and the method comprises the following steps: collecting multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different body parts or different points of the same part of a user in real time, including pressure intensity, stimulation mode, period and frequency parameters; multi-modal physiological feedback data, including stress distribution change, local blood flow change and electromyographic signals, of skin and superficial tissue under pressure stimulation are obtained through the bionic skin sensing module; according to the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data, AI-driven physiological effect prediction is carried out, and a prediction result is obtained; dynamically adjusting pressure stimulation parameters through an adaptive optimization module according to a prediction result and individual characteristics; a personalized stimulation scheme is generated through an intelligent control interface, data are monitored in real time, and the curative effect is recorded. According to the invention, the accuracy and intelligent level of peripheral pressure stimulation intervention are improved.

Description

Technical Field

[0001] The present invention relates to the fields of digital medicine and artificial intelligence technology, and in particular to an AI-based peripheral pressure stimulation quantitative assessment method and system. Background Art

[0002] Peripheral pressure stimulation is widely used in fields such as chronic pain management, neurorehabilitation, sports injury repair, and biomechanics research. Traditional pressure stimulation assessment methods mainly rely on manual experience or simple physical parameter settings, and lack accurate response to individual physiological differences and real-time biofeedback. At the same time, existing physiological effect modeling mostly relies on static statistical methods, which are difficult to dynamically adapt to the complex and changing body response process, and have obvious deficiencies in data noise processing, multi-granularity information fusion, and modeling accuracy. In addition, the traditional Daubechies wavelet loses significant high-frequency details during noise reduction, which will cause the electromyographic signal characteristics to be over-smoothed.

[0003] With the development of artificial intelligence (AI) technology, flexible sensing technology and intelligent control theory, there is an urgent need for an intelligent peripheral pressure stimulation quantitative assessment system that can integrate multimodal physiological signals and has adaptive learning capabilities, so as to achieve high-precision prediction and intervention optimization of the body's physiological state and improve the personalization and scientificity of treatment. Summary of the Invention

[0004] The present invention aims to solve the above problems and proposes an AI-based quantitative assessment method and system for peripheral pressure stimulation, which systematically improves the accuracy and intelligence level of peripheral pressure stimulation intervention.

[0005] To achieve the above objectives, the following technical solutions are adopted: According to one aspect of the present invention, a method for quantitatively evaluating peripheral pressure stimulation based on AI is provided, comprising the following steps: S1: Real-time collection of multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different parts of the user's body or different points on the same part, including pressure intensity, stimulation mode, cycle and frequency parameters; S2: Obtain multimodal physiological feedback data of the skin and superficial tissue under pressure stimulation through the bionic skin sensing module, including stress distribution changes, local blood flow changes and electromyographic signals; S3: performing AI-driven physiological effect prediction based on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data to obtain a prediction result, including: performing data preprocessing on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data; inputting the preprocessed data into a CMBlock_Mamba model constructed based on the Mamba mechanism to perform deep feature extraction and physiological effect prediction, and outputting predicted values ​​of multiple physiological effect indicators, wherein the multiple physiological effect indicators include blood flow, muscle reaction, and pain perception; S4: Dynamically adjust the pressure stimulation parameters through the adaptive optimization module based on the prediction results and individual characteristics; S5: Generate personalized stimulation plans, monitor data in real time, and record therapeutic effects through the intelligent control interface.

[0006] According to another aspect of the present invention, there is also provided an AI-based peripheral pressure stimulation quantitative assessment method system, comprising: Stimulation data acquisition module: real-time acquisition of multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different parts of the user's body or different points on the same part, including pressure intensity, stimulation mode, cycle and frequency parameters; Bionic skin sensing module: used to obtain multimodal physiological feedback data of the skin and superficial tissue under pressure stimulation, including stress distribution changes, local blood flow changes and electromyographic signals; Physiological effect prediction module: performs AI-driven physiological effect prediction based on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data to obtain prediction results, including: performing data preprocessing on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data; inputting the preprocessed data into the CMBlock_Mamba model built based on the Mamba mechanism to perform deep feature extraction and physiological effect prediction, and outputting predicted values ​​of multiple physiological effect indicators, including blood flow, muscle response, and pain perception; Adaptive optimization module: dynamically adjusts pressure stimulation parameters based on prediction results and individual characteristics; Intelligent control interface: used to generate personalized stimulation plans, monitor data in real time, and record therapeutic effects.

[0007] Compared with the prior art, the present invention achieves the following beneficial effects: 1. This invention utilizes Coiflets wavelet decomposition combined with a minimax threshold adaptive algorithm and soft thresholding to achieve multi-scale feature extraction and efficient denoising. This ensures that the input signal retains its intrinsic physiological characteristics while effectively suppressing the effects of noise, providing high-fidelity input data for subsequent modeling. This overcomes the difficulty of balancing noise suppression with physiological signal feature preservation in traditional wavelet denoising (such as Daubechies wavelet), which often results in loss of high-frequency detail or low-frequency baseline drift.

[0008] 2. This paper addresses the redundant information present in physiological data of varying granularity by designing a residual redundancy removal module. Through three-dimensional tensorization, unified embedding space processing, and residual connections, this module systematically eliminates redundancy between multi-scale features, improving the independence and effectiveness of feature representation, and enhancing model convergence and generalization capabilities. This overcomes the problem that traditional methods (such as PCA dimensionality reduction) often fail to adequately address redundant multi-scale physiological data, leading to feature overlap or information loss.

[0009] 3. Based on the Mamba state modeling mechanism, this paper constructs the CMBlock_Mamba deep neural network architecture. This model integrates normalization, dynamic feature extraction (Mamba module), deep perception (Multi-layer Perceptron module), and feature fusion mechanisms. It can efficiently capture the temporal dynamics and cross-scale correlations of physiological signals, enabling accurate prediction of physiological effects such as local blood flow changes, muscle response intensity, and pain perception. The CMBlock architecture, based on the Mamba state space model, reduces computational complexity through lightweight convolution and state recursion, reduces errors in modeling long-term dependencies such as blood flow and electromyography, and improves real-time inference speed.

[0010] 4. The present invention introduces Jensen-Shannon (JS) divergence and determination coefficient ( ) weighted loss function. The JS divergence is used to measure the similarity between the output distribution and the true distribution, combined with The metrics evaluate the quality of numerical fitting, forming a comprehensive optimization strategy that effectively improves the model's ability to learn and generalize complex physiological response patterns, and enhances the prediction model's ability to balance distribution consistency and numerical fitting accuracy. This avoids the shortcomings of single-task regression models (such as SVR) or mean squared error (MSE) loss functions, which struggle to balance distribution consistency and numerical accuracy.

[0011] In summary, the present invention has achieved a leap from "experience-driven" to "data intelligence-driven" in peripheral pressure stimulation assessment, and is significantly superior to existing technologies in terms of denoising accuracy, model efficiency, personalized adaptation and clinical safety, providing reliable intelligent solutions for chronic pain management, sports rehabilitation and other fields.

[0012] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which: Figure 1 1 is a schematic diagram of the specific steps of an AI-based quantitative assessment method for peripheral pressure stimulation according to an embodiment of the present invention; Figure 2 This is a flowchart of an AI-based peripheral pressure stimulation quantitative assessment method and system according to an embodiment of the present invention; Figure 3 It is the data processing and prediction model building process of the present invention; Figure 4 Schematic diagram of the CMBlock stacking structure of the present invention; Figure 5 3 is a module diagram of an AI-based peripheral pressure stimulation quantitative evaluation system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0015] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0016] In the present invention, peripheral pressure stimulation refers to the controllable mechanical pressure applied to the peripheral parts of the human body (such as skin, muscles, joints, etc.) through external equipment. The purpose is to induce specific physiological reactions through physical stimulation for treatment, rehabilitation or physiological status assessment.

[0017] Peripheral pressure refers to areas of the body outside the central nervous system, such as the skin, muscles, blood vessels, and peripheral nerves of the limbs, back, and soles. Unlike interventions targeting the central nervous system (such as the brain and spinal cord), peripheral pressure stimulation acts directly on the body's surface or superficial tissues, influencing physiological functions through local mechanical effects.

[0018] In the present invention, specific forms of pressure stimulation include: Pressure intensity: the amount of force applied (unit: kPa or mmHg), such as light touch (low intensity) or deep pressure (high intensity). Stimulation mode: continuous mode (continuous pressure), intermittent mode (periodic pressure application and release), pulse mode (high-frequency short-term impact). Frequency and period: the rhythm of pressure changes (such as once or five times per second), which affects tissue adaptability and physiological response speed. Spatial distribution: using multi-channel sensors to synchronously monitor the force applied to different parts of the body to optimize pressure distribution (such as to avoid damage caused by local high pressure).

[0019] To address the severe noise interference in collected physiological time series signals, this paper introduces the Coiflets wavelet decomposition and minimax threshold estimation method for the first time, combined with soft thresholding rules, to achieve multi-scale feature extraction and efficient denoising. This ensures that the input signal retains the intrinsic physiological characteristics while effectively suppressing the influence of noise, providing a clean and stable data foundation for subsequent modeling. To address the redundant information present in physiological data of different granularities, this paper designs a residual redundancy removal module. Through three-dimensional tensorization, unified embedding space processing, and residual connections, this module systematically eliminates redundancy between multi-scale features, improves the independence and effectiveness of feature representation, and enhances the convergence and generalization of modeling. Based on the Mamba state modeling mechanism, this paper innovatively constructs the CMBlock_Mamba deep neural network architecture. This model integrates normalization, dynamic feature extraction (Mamba module), deep perception (multi-layer perceptron module), and feature fusion mechanisms to efficiently capture the temporal dynamics and cross-scale correlations of physiological signals, enabling accurate prediction of physiological effects such as local blood flow changes, muscle response intensity, and pain perception. In order to improve the prediction model's ability to balance distribution consistency and numerical fitting accuracy, this paper introduces the Jensen-Shannon (JS) divergence and determination coefficient ( ) weighted loss function. The JS divergence is used to measure the similarity between the output distribution and the true distribution, combined with The indicators evaluate the quality of numerical fitting and form a comprehensive optimization strategy to effectively improve the model's ability to learn and generalize complex physiological response patterns.

[0020] Figure 1 A schematic diagram showing the specific steps of an AI-based quantitative assessment method for peripheral pressure stimulation is shown. Figure 2 The figure shows a flow chart of a method and system for quantitative evaluation of peripheral pressure stimulation based on AI. Figure 1 and Figure 2 As shown, an AI-based quantitative assessment method for peripheral pressure stimulation includes the following steps: S1: Real-time collection of multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different parts of the user's body or different points on the same part, including pressure intensity, stimulation mode, cycle and frequency parameters; This step S1 is mainly responsible for high-precision, comprehensive, real-time collection of various parameters during the application of peripheral pressure stimulation, ensuring the comprehensiveness and accuracy of the input data for the subsequent analysis model. It allows multi-point synchronous measurement, that is, collecting data simultaneously at different parts of the body or different points on the same part, to construct a complete pressure stimulation spatial distribution map, which is convenient for subsequent spatial feature analysis. The multi-dimensional pressure stimulation parameters are collected through the stimulation data acquisition module, which specifically includes the following units: (1) Pressure intensity acquisition unit: The pressure acting on the skin surface is measured in real time through a high-sensitivity pressure sensor (such as a capacitive or piezoelectric sensor), usually in kPa or mmHg. The acquisition range is wide, covering different treatment needs from gentle touch to deep compression. Its accuracy range is ±0.05 kPa and the range is 0-200 kPa. During acquisition, the sensors are evenly distributed on the body part to be measured to ensure that the pressure intensity at different points can be accurately measured. The system can alarm according to the set threshold to avoid excessive stimulation causing discomfort or tissue damage.

[0021] (2) Stimulation pattern recognition unit: By analyzing the timing characteristics of pressure stimulation, the applied stimulation mode is automatically identified and recorded, including continuous mode, intermittent mode, and pulse mode. At the same time, the stimulation rhythm (such as frequency, duty cycle) and stimulation duration are recorded to ensure that the subsequent model can fully understand the impact of different modes on physiological responses.

[0022] (3) Stimulation cycle and frequency acquisition unit: records the time points of stimulus application and removal, calculates the stimulation cycle (number of stimulations per unit time) and frequency (Hz), and is used to evaluate the body's response characteristics under different rhythms. The system supports millisecond-level time resolution, ensuring that dynamic stimulation patterns (such as rapid frequency changes) can also be accurately captured.

[0023] S2: Obtain multimodal physiological feedback data of the skin and superficial tissue under pressure stimulation through the bionic skin sensing module, including stress distribution changes, local blood flow changes and electromyographic signals; In step S2, the bionic skin sensing module simulates the multi-layered structure and sensory properties of human skin, combining various advanced sensing technologies to collect real-time biophysical and biochemical feedback signals under peripheral pressure stimulation. The entire bionic skin sensing module is made of highly flexible, low-modulus materials, with a thickness that can be controlled to the millimeter or even submillimeter level to ensure fit and wear comfort. It adapts to the pressure sensing needs of complex curved surfaces in different parts of the body (such as the palm, sole, and back), preventing the sensor itself from affecting physiological status. The bionic skin sensing module includes the following units: (1) Tissue stress distribution sensing unit: A large-area stress sensing array is made using flexible sensing materials (such as graphene, nanosilver wire, and flexible capacitor grids) to capture the stress distribution changes of the skin and superficial tissues under pressure stimulation in real time. Through two-dimensional or three-dimensional pressure imaging technology, the system can accurately identify local high-pressure points and areas of concentrated shear force, providing a basis for avoiding pressure injuries and optimizing the stimulation location. Alternatively, the Myoton Pro muscle tester can be used to measure muscle mechanical indicators.

[0024] (2) Local blood flow change monitoring unit: integrated optical sensors (such as near-infrared spectroscopy (NIRS) and laser speckle imaging (LSI)) or micro-ultrasound sensors to monitor changes in blood flow and blood flow velocity in the pressure area in real time. NIRS detects changes in blood oxygen saturation, reflecting tissue ischemia or congestion, and LSI can dynamically observe changes in microcirculatory blood flow and capture blood flow recovery or blockage in a short period of time. For example, when using the fNIRS near-infrared spectroscopy imaging system to collect blood flow change data, the probe is fixed at the frontal lobe position, and the signal is collected in a quiet state. The collection time is 5 minutes and the sampling frequency is 10 Hz. During the collection process, the subject is kept in a quiet state to avoid interference with the data caused by head movement, and the subject's blood flow change data is recorded.

[0025] (3) Electromyographic signal acquisition unit: Using a surface electromyography (sEMG) electrode array, the electrophysiological responses of local muscle groups are synchronously recorded during pressure stimulation. The unit monitors muscle activation, relaxation, and reflex contraction responses; analyzes the effects of different stimulation modes (e.g., continuous pressure versus intermittent pressure) on muscle neural control; and provides important indicators for neurorehabilitation and muscle function assessment. For example, using the American DELSYS Trigno surface electromyography testing system, electrodes are attached to appropriate locations on the subject's right triceps, long head of the biceps, and short head of the biceps.

[0026] S3: performing AI-driven physiological effect prediction based on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data to obtain a prediction result, including: performing data preprocessing on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data; inputting the preprocessed data into a CMBlock_Mamba model constructed based on the Mamba mechanism to perform deep feature extraction and physiological effect prediction, and outputting predicted values ​​of multiple physiological effect indicators, wherein the multiple physiological effect indicators include blood flow, muscle reaction, and pain perception; This step S3 establishes a mapping relationship between pressure stimulation and physiological response based on a deep learning algorithm, and can predict physiological indicators such as local blood flow changes, muscle response, and pain perception under different pressure stimulation conditions.

[0027] The input data is collected by step S1 and step S2 to obtain two parts of data: one is the collected multi-dimensional pressure stimulation parameters, such as the magnitude of the applied force, frequency, action time, etc.; the other is the multi-modal physiological feedback data collected by the bionic skin sensing module, such as stress changes, blood flow, electromyographic signals (EMG) and other time series data. First, the multi-dimensional pressure stimulation data and multi-modal physiological feedback data of the above process are pre-processed to ensure that the various signals are strictly aligned on the time axis, which is convenient for subsequent feature extraction and model training. Figure 3 This is the data processing and prediction model building process of the present invention, and step S3 specifically includes the following sub-steps: S31: Data preprocessing of multi-dimensional pressure stimulation data and multi-modal physiological feedback data, including noise reduction and residual redundancy removal; S311: The collected time series data is subjected to Coiflets wavelet denoising. The processing flow includes using Coiflets wavelet decomposition, minimax threshold estimation and soft threshold rule for denoising and Coiflets wavelet reconstruction. Specifically, it includes: S3111: Use Coiflets wavelet basis function to perform multi-scale decomposition of the original signal; Coiflets wavelet has compact support, orthogonality and approximate symmetry, and is suitable for physiological signal processing. The Coiflets wavelet basis function is defined as: ; in, Indicates the Layer, Wavelet basis functions of the translation positions; is the number of wavelet decomposition layers; is the translation factor is the time variable; S3112: Decomposing the original signal into multi-scale approximation coefficients by determining the optimal decomposition layer number and wavelet order and detail coefficient ; The determination of the optimal number of decomposition layers includes: (1) calculating the maximum feasible number of layers: according to the signal length n (i.e., the number of signal sample points), according to the formula: -YL, where YL is a safety margin, and the optional YL=3 can be used to avoid excessive decomposition leading to too short low-frequency approximation coefficients. (2) Energy ratio analysis: Decompose each layer , calculate the approximation coefficient energy ratio layer by layer: ,in They are the approximation coefficient and detail coefficient of the sth layer respectively; select The minimum s ≥ 0.95 is used as the optimal decomposition level CS to ensure that more than 95% of the signal energy is retained.

[0028] (3) Termination condition: If the detail coefficient If the variance of changes less than 10% between two consecutive layers, the noise is considered to be fully separated and the decomposition is terminated.

[0029] The wavelet order is determined, choosing an order that strikes a balance between frequency domain resolution and time domain compactness. The feasible range of the Coiflet wavelet order JS is generally empirically limited to 1 ≤ JS ≤ 5. Higher-order wavelets (such as Coif5) offer better frequency domain resolution but at the expense of increased computational cost. Therefore, frequency band matching can be used to further analyze the dominant frequency range of physiological signals (e.g., the high-frequency components of EMG signals range from 20-500 Hz). The order JS is selected to maximize the overlap between the wavelet passband and the dominant frequency of the signal. Specifically, Coif1 is suitable for low-frequency signals (e.g., blood flow changes, 0.1-2 Hz); Coif3 is suitable for mid-frequency signals (e.g., electromyography, 20-150 Hz); and Coif5 is suitable for suppressing high-frequency noise (e.g., electrode motion artifacts).

[0030] S3113: Adaptively calculate the optimal threshold of each layer detail coefficient based on the minimax criterion; Threshold processing is performed on the detail coefficients (high-frequency components) of each layer after decomposition to remove noise interference. The optimal threshold λ is determined adaptively using the minimax criterion. The calculation of the threshold is based on the statistical characteristics of the noise. The optimal threshold λ is: ; Where n is the number of signal sample points; λ is the optimal threshold; Is the noise standard deviation estimate, noise estimation usually uses the median absolute deviation method, by the first layer detail coefficient Estimated Implementation: ; in, is the median absolute deviation operation; is the first layer detail coefficient.

[0031] The minimax criterion can minimize the signal error under unknown noise distribution conditions, ensuring the robustness and universality of denoising.

[0032] S3114: performing soft threshold processing on the detail coefficients of each layer according to the optimal threshold of the detail coefficients of each layer to obtain the denoised detail coefficients; Threshold processing uses a soft threshold strategy, specifically for each layer detail coefficient Perform soft threshold processing, the formula is as follows: ; in, is the original detail coefficient; is the detail coefficient after threshold processing, i.e., denoising; λ is the optimal threshold; sign is the sign function; max is the maximum value operation; To take the absolute value operation.

[0033] Compared with the hard threshold method, the soft threshold method can effectively avoid the oscillation phenomenon in the denoising process, smoothly suppress small-amplitude noise, and retain important features, ensuring the continuity and smoothness of the signal after denoising.

[0034] S3115: The detail coefficient after denoising and approximation coefficient Combination, reconstruction of denoised signal : ; in, is the denoised signal; is the coefficient after threshold processing; Indicates the Layer, Wavelet basis functions of the translation positions; is the approximation coefficient of the lowest decomposition level, is the corresponding scaling function.

[0035] After data denoising, residual de-redundancy is further used to eliminate redundant information between multi-granularity data to ensure the purity of data input.

[0036] S312: Perform residual redundancy removal on the denoised signal data, specifically eliminating multi-granularity redundant information through tensor conversion, unified embedding space mapping, and residual connection, including: S3121: Tensor dimension conversion: The multi-scale features extracted from the same segment of physiological signal data at different time / frequency resolution levels are divided into coarsest-grained data and fine-grained data. Among them, the coarsest-grained data refers to the feature data extracted from the time series with a longer time window or lower resolution, which provides the global change characteristics of the physiological response and is suitable as the "backbone feature" or "reference signal" of the model. Fine-grained data refers to the feature data extracted from the same physiological signal with a shorter time window or higher resolution. This type of data has high dimensionality and changes rapidly, but is more susceptible to noise interference or redundant repetition. Its role is to supplement detailed information and reflect short-term fluctuations and local reactions. Specifically: Divide the multi-scale features of the same physiological signal after denoising into the coarsest granularity data and fine-grained data ; For the coarsest-grained data and fine-grained data The dimension conversion is performed as follows: For the coarsest-grained data Perform dimension conversion: ; Num is the number of samples; is the characteristic dimension; is the number of channels ( ); For fine-grained data Perform dimension conversion: ; It is the characteristic dimension of fine-grained data; is the number of channels, which is consistent with the coarsest granularity data.

[0037] Assuming the original fine-grained data (100 samples, 256-dimensional features), converted to a three-dimensional tensor (Number of channels 64, feature length 4).

[0038] Through the above process, the coarsest granularity data and fine-grained data Perform dimensionality conversion, converting two-dimensional data into three-dimensional tensors. The three-dimensional tensor structure (number of samples × number of channels × feature length) is more suitable for the input requirements of convolutional networks (CNNs) or state-space models (such as Mamba), improving computational efficiency. Each granularity in the dimensionally converted fine-grained data is then segmented and sampled into subsequences. After conversion, each granularity data is sampled in segments. subsequences The expression is: ; is the length of a single subsequence.

[0039] Segmented sampling divides fine-grained data into subsequences of length c , capturing short-term fluctuations or local spatial patterns. For example, in a time series, if c = 10, each subsequence contains data from 10 time points, and the model can learn local temporal correlations.

[0040] S3122: Unified Embedding Space Mapping: Use a multi-layer perceptron to adjust the size of each fine-grained tensor to a unified embedding space to obtain the mapped feature tensor , the mapping formula is: ; are the fine-grained tensors after conversion in step S3121; is the feature tensor after mapping; represents a fully connected layer; is the activation function; Represents a flattening operation. Since the fully connected layer (FC) usually accepts two-dimensional input, the three-dimensional tensor needs to be flattened before entering the fully connected layer. Flattened to a 2D matrix, for example: , .

[0041] By jointly modeling the channel and feature dimensions through the fully connected layer, cross-channel information interaction can be enhanced. For example, each fine-grained tensor converted in step S3121 The dimension is 100×64×4 (Num=100, Channel=64, L=4); after flattening, it becomes 100×256, and is mapped to a unified embedding space (such as 256 dimensions) through a fully connected layer, and the output is .

[0042] Through the above process, the model can effectively extract the local features of fine-grained signals and fuse them with coarse-grained features, ultimately improving the accuracy of physiological effect prediction.

[0043] S3123: Residual connection eliminates redundancy: The residual connection is constructed according to the following rules to eliminate multi-granularity redundant information: ; in, Indicates the The feature tensor after layer processing is passed to the CMBlock_Mamba model in step S32 after information embedding and processing; represents the redundant features used by the next residual connection, , which means adjusting the number of channels by 1x1 convolution. The feature tensor after mapping After eliminating redundancy through residual connection, it is finally input into the CMBlock_Mamba model.

[0044] Through dimensionality conversion and segmented sampling, high-dimensional fine-grained data is regularized into a parseable tensor form, improving the model's ability to perceive local features. The channel number consistency design ensures that multi-granularity data can be seamlessly integrated, enhancing the model's ability to jointly model global and local physiological responses. The three-dimensional tensor structure is compatible with mainstream deep learning frameworks (such as PyTorch and TensorFlow), reducing implementation complexity. Through residual connections, redundant features between adjacent layers are dynamically subtracted ( ), preserving differential information across data granularities and avoiding feature duplication or noise interference. By suppressing redundant feature transfer, the complexity of network parameter training is reduced, model convergence is accelerated, and generalization ability is improved, ultimately enhancing the accuracy and robustness of physiological effect predictions.

[0045] S32: Inputting the pre-processed data into the CMBlock_Mamba model built based on the Mamba mechanism to perform deep feature extraction and physiological effect prediction, and outputting predicted values ​​of multiple physiological effect indicators, wherein the multiple physiological effect indicators include blood flow, muscle reaction, and pain perception; After the noise reduction and redundancy removal preprocessing in step S31, the data is fed into the CMBlock_Mamba model built based on the Mamba mechanism for deep feature extraction and physiological effect prediction. The CMBlock_Mamba model consists of multiple CMBlocks.

[0046] S321: Constructing CMBlock Each CMBlock consists of a Mamba module, a residual connection module, and a two-layer multilayer perceptron (MLP) module. The Mamba module is responsible for modeling temporal dynamic features. The residual connection module preserves the original input features to prevent gradient vanishing. The multilayer perceptron (MLP) module is a two-layer fully connected network used to enhance the feature expression capabilities of the current CMBlock. Specifically: (1) Dynamic feature modeling based on Mamba module The Mamba module is a state-space modeling structure that can effectively capture long-term dependencies and dynamic changes in time series. The state update formula (expressed in discretized form) is: ; :time The hidden state vector of :Input feature vector, the initial value is the output of step S3123 ; : The state transfer matrix and the observation matrix are both learnable parameters. Initialization method: A and B are initialized with normal distribution, C and D are set as learnable diagonal matrices; training process: Adam optimizer is used, learning rate 1e-4, and batch size 32.

[0047] The Mamba module combines convolution with state recursion, and is lightweight, has good parallelism, and strong dynamic perception capabilities.

[0048] (2) Residual Connection Module After the dynamic feature modeling output based on the Mamba module, residual connections are added to alleviate the gradient disappearance problem and maintain the original information flow: ; Represents the features extracted by the Mamba module; is the output feature after the residual connection module.

[0049] (3) Constructing a multi-layer perceptron module After each CMBlock, a two-layer multilayer perceptron (MLP) module is set to further improve the feature expression capability: (3.1) First layer fully connected (with activation function ReLU): ; is the output feature after the residual connection module; is the activation function; is the weight matrix; is the bias term; is the output after the first layer of full connection.

[0050] (3.2) Layer 2 fully connected (no activation function): ; in, is the weight matrix; is the bias term; the final output As the output result of the current CMBlock.

[0051] S322: Layer stacking and feature fusion By stacking multiple CMBlocks, a deep feature extraction structure is formed, and the following is obtained: Figure 4The CMBlock stacking structure shown in the figure has a number of CMBlocks NUM. The implementation of the hierarchical stacking and feature fusion of the CMBlock_Mamba model is as follows: generating multi-level features by stacking multiple CMBlocks; performing weighted fusion on the multi-level features to obtain the fused features , thus achieving the integration of features output at different levels through feature fusion operation. The fusion formula is as follows: ; in, Indicates the CMBlock output; is the total number of CMBlocks; is the learnable fusion weight; the final fusion feature As input for physiological response prediction.

[0052] The model was trained using a historical dataset that included pressure stimulation parameters and corresponding physiological feedback data for 500 patients. The training parameters of the model were: Adam optimizer, learning rate 10 −4 , batch size 32, training epochs 200.

[0053] S34: Output prediction Assume that the final prediction target has M subtasks, corresponding to different physiological effect indicators (such as local blood flow changes, muscle reaction intensity, pain perception, etc.), then the output is: ; in, is the fusion feature obtained in step S33; is the output layer weight matrix; is the output layer bias; The predicted value of the physiological effect indicator output by the model; is the Tanh activation function.

[0054] This step builds a separate regressor for each physiological indicator, sharing a set of forward networks (sharing underlying representations), allowing the model to simultaneously output multiple continuous physiological responses. It is a typical multi-task regression architecture with shared features and independent outputs, which is conducive to improving modeling efficiency and generalization capabilities.

[0055] S35: Loss function optimization In order to simultaneously measure the probability distribution deviation and numerical fitting effect of the model output, the CMBlock_Mamba model training adopts the weighted loss function of Jensen-Shannon divergence and determination coefficient. Expressed as: ; in, , They are Divergence and Weight coefficient of loss term; empirical value range: α∈[0.3,0.7], β=1-α, used to balance distribution and numerical loss, preferably, α=0.5, β=0.5.

[0056] Represents the predicted distribution The Jensen-Shannon divergence between the probability distribution Y and the true distribution Y, where Y is the true value of a physiological effect indicator, such as blood flow changes, muscle reaction intensity, pain perception, etc. Based on the definition of Kullback-Leibler divergence, it is used to measure the similarity between two probability distributions. Its calculation formula is as follows: ; in, The true distribution Normalized vectors (such as blood flow changes, muscle reaction intensity, etc.), is the predicted distribution Normalized vector; is the middle distribution, which is the mean of P and Q, used to balance and differences; It represents the Kullback-Leibler divergence, which reflects the degree of deviation between the true distribution P and the intermediate distribution M. The smaller the value, the closer P is to M, that is, the closer the model prediction distribution is to the true distribution. It is defined as: ; in, represents the probability value of the i-th element (i.e., sample) in vector P. For example, to predict blood flow changes, represents the probability of the true blood flow change value of the i-th element; Represents the probability value of the i-th element (i.e., sample) in vector Q, for example, it represents the probability of the blood flow change value of the i-th element predicted by the model; , represents the i-th element of the intermediate distribution M; the multi-task prediction of the present invention simultaneously predicts blood flow, muscle reaction and pain perception, etc., and i corresponds to the joint distribution dimension of different physiological indicators.

[0057] JS divergence is symmetric (KL divergence is asymmetric) and its value range is [0,1], which is more suitable for measuring the similarity of two distributions. When predicting multiple physiological indicators (such as blood flow, muscle response, pain perception), JS divergence can uniformly measure the global differences of multi-task distributions, while Providing an intermediate reference benchmark for each task. By balancing the differences between PP and Q, the model's modeling accuracy and generalization ability for complex physiological response patterns are improved.

[0058] It represents the coefficient of determination, which is used to measure the consistency between the model prediction value and the true value and is defined as: ; in, is the sample size; The true distribution The mean of .

[0059] This step S35 minimizes the total loss , the model simultaneously optimizes: distribution consistency (JS divergence): ensuring that the probability distribution of the predicted results matches the real physiological response pattern; numerical fitting accuracy (R 2 Coefficient of determination): Ensures that the absolute error of the predicted value is minimized.

[0060] S4: Dynamically adjust the pressure stimulation parameters through the adaptive optimization module based on the prediction results and individual characteristics; In step S4, the adaptive optimization module is the core of the system's dynamic adjustment and personalization. It aims to continuously optimize peripheral pressure stimulation parameters based on the subject's individual characteristics and real-time physiological feedback to improve the intervention effect. Its main functions and designs include: dynamically adjusting pressure stimulation parameters based on prediction results and individual characteristics through the adaptive optimization module, including: (1) Individual characteristic modeling: Before use, the system inputs or measures the subject's basic physiological characteristics, including but not limited to: Age, sex, weight, height; Skin thickness, muscle condition (muscle mass, tone); Nerve reflex sensitivity indicators (such as nerve conduction velocity, reflex threshold); Past medical history (e.g., diabetic neuropathy, history of chronic pain, records of previous treatment responses, etc.).

[0061] These features are used to establish a personalized basic model, providing a preliminary basis for adjusting pressure stimulation parameters. Specifically: Encode the above features into feature vectors , which means that the feature vector after encoding the basic physiological characteristics of the subject is a d-dimensional real number space, which is input into the preset rule engine or lightweight neural network to generate the initial stimulation parameter set: ,in, Including initial pressure intensity, stimulation mode (continuous / intermittent / pulse), frequency range, action area, etc. Optionally, the lightweight neural network used is a three-layer fully connected neural network, and the number of neurons in the input layer is the individual feature dimension d=10, = [age, weight, height, skin thickness, muscle mass, nerve conduction velocity, blood pressure, blood glucose level, pain threshold, motor ability score], with 32 hidden layer neurons and the number of output layer neurons equal to the dimension of the initial stimulus parameters. Training was performed using the ReLU activation function and the Adam optimizer, with a learning rate of 0.001 and a batch size of 16.

[0062] Safety constraints: Set an upper limit on pressure intensity based on skin thickness and medical history to avoid tissue damage; set a minimum stimulation interval based on nerve reflex sensitivity to prevent overstimulation.

[0063] (2) Real-time feedback analysis: The physiological feedback such as tissue stress distribution, local blood flow, and electromyographic signals collected by the bionic skin sensing module are compared with the prediction results of the AI-driven physiological effects in step S4; a dynamic difference analysis algorithm is used to evaluate the stimulation effect; and short-term adaptability (such as immediate blood flow improvement) and long-term trend changes (such as muscle tension relief) are monitored.

[0064] A dynamic difference analysis algorithm is used to evaluate the stimulation effect and provide real-time feedback on the analysis results, including calculating the mean square sum of the instantaneous deviations between the actual physiological data and the predicted results to obtain the difference measurement index D: ; Where T is the total number of time steps; The real data collected at each time step t, namely, tissue stress, blood flow, electromyographic signal (sEMG), etc. collected by the bionic skin sensor; is the predicted value at each time step t.

[0065] Difference type judgment: Short-term adaptability (smaller T): evaluate the immediate intervention effect (such as the speed of blood flow improvement); long-term trend analysis (larger T): monitor chronic responses such as muscle tension relief and changes in pain perception.

[0066] The smaller the difference D, the more accurate the model prediction is; otherwise, the stimulation parameters need to be adjusted or the model needs to be optimized.

[0067] Multi-task prediction output: The model of the present invention predicts multiple physiological indicators (such as blood flow, electromyography, pain, etc.) at the same time. is a vector, for example: , each subtask calculates the loss independently and is finally optimized through the weighted loss function.

[0068] (3) Dynamic adjustment of stimulation parameters: According to the difference metric D and individual characteristics, the adaptive optimization module adjusts the pressure stimulation parameters through a policy gradient algorithm, including at least one of: fine-tuning the pressure intensity, switching the stimulation mode, optimizing the stimulation frequency / period, and local offsetting the stimulation area. Specifically: Fine adjustment of pressure intensity (80-110 mmHg); Stimulation mode switching (e.g., switching from continuous pressure to pulsed pressure); Optimization of stimulation frequency and period (extending or shortening the stimulation interval, 20-70s); Local deflection of the stimulation area (to avoid prolonged pressure on local tissue).

[0069] If the difference metric D exceeds a safety threshold or an abnormal signal (such as local tissue ischemia) is detected, an emergency stop or parameter reset is triggered.

[0070] according to and The adaptive optimization module dynamically adjusts the pressure stimulation parameters through the policy gradient algorithm, with the goal of minimizing the difference D and improving the intervention effect.

[0071] Using a policy gradient-based approach, the stimulation scheme is iteratively optimized during the continuous stimulation process to ensure that each application is in the optimal state. The goal is to maximize the cumulative reward or maximize the expected return. (Treatment effect score): ; Among them, R(s,a) is the action taken Immediate rewards obtained (e.g., improvement in physiological state); is an action (such as increasing pressure , switch mode); is the current status (e.g., degree of blood flow improvement, level of myoelectric recovery); are policy network parameters.

[0072] The reward function R(s,a) is defined as follows: ; in, is the weight, is the indicator improvement amount, and D is the instantaneous deviation.

[0073] Use the policy gradient update formula: ; Express Parameter derivatives; Strategies for adjusting stimulation parameters.

[0074] This formula represents the optimization strategy of adjusting the stimulation parameters according to the actual reward feedback at each step.

[0075] Through dynamic feedback and comparison between prediction and actual measurement, treatment is optimized to achieve precise and personalized control of peripheral pressure stimulation.

[0076] (4) Adaptive threshold update: During the long-term intervention process, the system automatically adjusts the response threshold according to the subject's physiological changes (such as increased tolerance and decreased sensitivity), avoiding early setting failures and improving the personalized adaptability of long-term intervention. , defining the tolerance score :Calculate pressure intensity based on historical data combined with subjective feedback (such as pain score), etc. ; Among them, the current maximum tolerable pressure: the pressure value (unit: kPa) is determined by gradually increasing the pressure intensity until physiological abnormalities (such as blood flow obstruction, abnormal electromyographic signals) are detected or the patient actively reports discomfort. Initial tolerable pressure: the maximum safe pressure value that the patient can withstand during the first treatment cycle (initially set by the doctor or the system). Pain score: uses the visual analog scale (VAS, 0-10 points), where 0 is painless and 10 is severe pain. Decline rate: the percentage of decrease in the pain score of the current cycle relative to the initial score, that is: : Dividing by 10 is a normalization process, scaling the descent rate to a range of 0-10, consistent with the pressure intensity ratio dimension.

[0077] According to the changes in patient tolerance (tolerance score ) Dynamically update threshold: ; in, Is the current cycle the stimulation threshold (e.g., maximum permissible pressure); is the learning rate, used to adjust the amplitude; Is the current cycle The tolerance score, is the target tolerance score (set according to individual rehabilitation goals), which is the preset ideal tolerance score (e.g. =1.2), indicating the expected improvement in tolerance and pain relief.

[0078] like , indicating that tolerance increases, raising the threshold ρ to enhance the stimulus intensity; like , indicating a decrease in tolerance, the threshold ρ is lowered to ensure the safety of personalized stimulation.

[0079] Updated threshold As the constraint condition for parameter adjustment in the next cycle, it is written into the individual model.

[0080] S5: Generate personalized stimulation plans, monitor data in real time, and record therapeutic effects through the intelligent control interface.

[0081] The intelligent control interface is an important window for interaction between the system and the user (operator / doctor / patient). Its design focuses on intuitiveness, personalization, and feedback to ensure user experience and data visualization. Its main functions include: (1) Personalized stimulation plan generation interface: Automatically generates personalized pressure stimulation plans based on the individual information of the subjects and system recommendations; provides three options: default plan, recommended plan, and manual editing mode, supporting doctors or professionals to adjust according to specific needs; displays expected physiological effects (such as the percentage of blood flow improvement and the expected value of muscle tension reduction).

[0082] (2) Real-time monitoring and feedback display interface: Visually present pressure distribution diagrams, local blood flow dynamic diagrams, and electromyographic signal change curves; provide real-time stimulation status monitoring (such as current stimulation intensity, mode, and remaining time); and have abnormal alarms (such as detection of abnormal reflexes or high-voltage risks) to promptly remind users.

[0083] (3) Interactive parameter adjustment function: Users can instantly adjust key parameters such as pressure intensity, stimulation rhythm, and stimulation mode through the interface; it supports multiple interaction methods such as touch and voice commands to improve convenience, and is particularly suitable for use by patients with limited mobility.

[0084] (4) Recording and tracking of therapeutic effects: Automatically record stimulation parameters, subject feedback (such as pain scores), and physiological effect change data for each treatment process; provide timeline efficacy tracking charts to facilitate doctors or users to evaluate long-term intervention effects and adjust treatment strategies; support one-click export of treatment records to facilitate medical file management or scientific research data accumulation.

[0085] (5) Intelligent recommendation and self-learning functions: Based on the treatment data and efficacy feedback accumulated by the system, federated learning or local training mechanisms are used to continuously optimize the recommendation algorithm; better stimulation plans are automatically recommended for patients with similar characteristics, improving the overall intelligence level of the system and the success rate of individual interventions.

[0086] Suppose an athlete receives rehabilitation treatment for a muscle strain. The present invention can monitor muscle tension and blood flow velocity in real time by attaching a flexible sensor to the affected area. After the AI ​​model analyzes the data, it automatically controls the pressure equipment to apply intermittent pulse pressure to stimulate local microcirculation. If insufficient blood flow recovery is detected, the system will increase the pulse frequency while avoiding secondary damage caused by excessive pressure. The traditional method only relies on the therapist to manually press the affected area and adjust the strength based on experience. The effect cannot be quantified and may lead to slow recovery due to uneven strength. Therefore, compared with traditional methods, the present invention has obvious technical advantages. The present invention uses AI technology to transform traditional empirical operations into precise "data-driven" treatments, which can improve the safety, personalization and scientificity of the treatment effect.

[0087] Figure 5 This is a module diagram of a peripheral pressure stimulation quantitative evaluation method system based on AI in an embodiment of the present invention. Figure 5 As shown, an AI-based peripheral pressure stimulation quantitative assessment method system 200 includes: Stimulation data acquisition module 210: collects in real time multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different parts of the user's body or different points on the same part, including pressure intensity, stimulation mode, cycle and frequency parameters; Bionic skin sensing module 220: used to obtain multimodal physiological feedback data of the skin and superficial tissue under pressure stimulation, including stress distribution changes, local blood flow changes and electromyographic signals; Physiological effect prediction module 230: performs AI-driven physiological effect prediction based on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data to obtain prediction results, including: performing data preprocessing on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data; inputting the preprocessed data into a CMBlock_Mamba model constructed based on the Mamba mechanism to perform deep feature extraction and physiological effect prediction, and outputting predicted values ​​of multiple physiological effect indicators, wherein the multiple physiological effect indicators include blood flow, muscle response, and pain perception; Adaptive optimization module 240: dynamically adjusts pressure stimulation parameters based on prediction results and individual characteristics; Intelligent control interface 250: generates personalized stimulation plans, monitors data in real time, and records therapeutic effects.

[0088] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0089] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the methods.

[0090] It should also be noted that, in the embodiments of the present application, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device comprising the elements.

[0091] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the embodiments of the present application may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown in the embodiments of the present application, but rather will conform to the widest scope consistent with the principles and novel features disclosed in the embodiments of the present application.

Claims

1. An AI-based quantitative assessment method for peripheral pressure stimulation, characterized in that: The following steps are involved: S1: Real-time collection of multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different parts of the user's body or different points on the same part, including pressure intensity, stimulation mode, cycle and frequency parameters; S2: Obtain multimodal physiological feedback data of the skin and superficial tissue under pressure stimulation through the bionic skin sensing module, including stress distribution changes, local blood flow changes and electromyographic signals; S3: performing AI-driven physiological effect prediction based on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data to obtain a prediction result, including: performing data preprocessing on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data; The pre-processed data is input into the CMBlock_Mamba model built based on the Mamba mechanism for deep feature extraction and physiological effect prediction, and the predicted values ​​of multiple physiological effect indicators are output, including blood flow, muscle reaction and pain perception; S4: Dynamically adjust the pressure stimulation parameters through the adaptive optimization module based on the prediction results and individual characteristics; S5: Generate personalized stimulation plans, monitor data in real time, and record therapeutic effects through the intelligent control interface.

2. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 1, characterized in that: in, The data preprocessing includes noise reduction processing and residual redundancy removal processing; wherein, the noise reduction processing: using Coiflets wavelet decomposition, minimax threshold estimation and soft threshold rule to perform noise reduction and Coiflets wavelet reconstruction, specifically including: Coiflets wavelet basis function is used to perform multi-scale decomposition of the original signal; Determining the optimal decomposition layer number and wavelet order, and decomposing the original signal into multi-scale approximation coefficients and detail coefficients; Adaptively calculate the optimal threshold of each layer's detail coefficient based on the minimax criterion; According to the optimal threshold of the detail coefficient of each layer, soft threshold processing is performed on the detail coefficient of each layer to obtain the denoised detail coefficient; The denoised detail coefficients are combined with the approximation coefficients to reconstruct the denoised signal.

3. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 2, characterized in that: in, The residual redundancy removal process is to eliminate multi-granularity redundant information of the denoised signal data through tensor conversion, unified embedding space mapping and residual connection, including: Divide the multi-scale features of the same physiological signal after denoising into coarse-grained data and fine-grained data; Performing dimension conversion on the coarsest-grained data and the fine-grained data respectively; Each granularity data in the dimensionally converted fine-grained data is segmented and sampled into subsequences; Use a multi-layer perceptron to adjust the size of each fine-grained tensor to a unified embedding space to obtain the mapped feature tensor; The residual connection is constructed according to the following rules to eliminate multi-granularity redundant information: ； in, Indicates the Feature tensor after layer processing; is the feature tensor after mapping; Represents redundant features for use by the next residual connection.

4. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 3, characterized in that: in, The CMBlock_Mamba model consists of multiple CMBlocks, each of which contains a Mamba module, a residual connection module and a two-layer multilayer perceptron module. The multi-layer perceptron module includes: setting two fully connected layers after each CMBlock, and the outputs of the two fully connected layers are respectively: ； in, , is the weight matrix, , is the bias term, is the activation function; is the output feature after residual connection.

5. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 4, characterized in that: in, The CMBlock_Mamba model further includes: forming a deep feature extraction structure by stacking the multiple CMBlocks; The implementation of the layer stacking and feature fusion of the CMBlock_Mamba model is as follows: Generate multi-level features by stacking multiple CMBlocks; Perform weighted fusion on the multi-level features to obtain fusion features .

6. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 5, characterized in that: in, The data after data preprocessing is input into the CMBlock_Mamba model built based on the Mamba mechanism for deep feature extraction and physiological effect prediction, and the predicted values ​​of multiple physiological effect indicators are output. The multiple physiological effect indicators include blood flow, muscle reaction and pain perception, including: Fusion features For multi-task regression, the output layer formula is: ； in, It is a fusion feature; is the output layer weight matrix; is the output layer bias; The predicted value of the physiological effect indicator output by the model; is the Tanh activation function; Each physiological effect indicator corresponds to an independent regressor, sharing the underlying network features, and the model simultaneously outputs multiple continuous physiological effect indicators.

7. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 6, characterized in that: in, The CMBlock_Mamba model training adopts the weighted loss function of Jensen-Shannon divergence and determination coefficient, and the total loss Expressed as: ； in, , They are Divergence and The weight coefficient of the loss term; Represents the predicted distribution and the true distribution The Jensen-Shannon divergence between The calculation formula is as follows: ； in, The true distribution The normalized vector, is the predicted distribution Normalized vector; is the middle distribution, which is the mean of P and Q; represents the Kullback-Leibler divergence and is defined as: ； in, Represents the probability value of the i-th element in vector P; Represents the probability value of the i-th element in vector Q; , represents the i-th element of the intermediate distribution M; It represents the coefficient of determination, which is used to measure the consistency between the model prediction value and the true value and is defined as: ； in, is the sample size; The true distribution The mean of .

8. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 7, characterized in that: S4: Dynamically adjust the pressure stimulation parameters through the adaptive optimization module according to the prediction results and individual characteristics, include: Individual characteristic modeling: Establish a personalized basic model by inputting or measuring the subject's age, gender, weight, height, skin thickness, muscle condition, nerve reflex sensitivity index and medical history; A dynamic difference analysis algorithm is used to evaluate the stimulation effect and provide real-time feedback on the analysis results, including calculating the mean square sum of the instantaneous deviations between the actual physiological data and the predicted results to obtain the difference measurement index D: ； Where T is the total number of time steps; is the real data collected at time step t; is the prediction result of the CMBlock_Mamba model at time step t; According to the difference metric D and individual characteristics, the adaptive optimization module adjusts the pressure stimulation parameters through a policy gradient algorithm, including at least one of: fine-tuning the pressure intensity, switching the stimulation mode, optimizing the stimulation frequency / period, and local offsetting the stimulation area; The objective function is to maximize the expected cumulative reward and dynamically adjust the stimulation threshold according to the tolerance score; If the difference metric D exceeds a safety threshold, an emergency stop or parameter reset is triggered.

9. The AI-based quantitative assessment method for peripheral pressure stimulation according to claim 8, characterized in that: in, The dynamically adjusting the stimulation threshold according to the tolerance score includes: ； in, Is the current cycle The stimulation threshold; is the learning rate, used to adjust the amplitude; Is the current cycle The tolerance score, is the target tolerance score set; like , indicating that tolerance increases, raising the threshold ρ to enhance the stimulus intensity; like , indicating a decrease in tolerance, the threshold ρ is lowered to ensure the safety of personalized stimulation.

10. An AI-based peripheral pressure stimulation quantitative assessment method system, characterized in that: include: Stimulation data acquisition module: real-time acquisition of multi-dimensional pressure stimulation data of peripheral pressure stimulation applied to different parts of the user's body or different points on the same part, including pressure intensity, stimulation mode, cycle and frequency parameters; Bionic skin sensing module: used to obtain multimodal physiological feedback data of the skin and superficial tissue under pressure stimulation, including stress distribution changes, local blood flow changes and electromyographic signals; Physiological effect prediction module: performs AI-driven physiological effect prediction based on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data to obtain prediction results, including: performing data preprocessing on the multi-dimensional pressure stimulation data and the multi-modal physiological feedback data; inputting the preprocessed data into the CMBlock_Mamba model built based on the Mamba mechanism to perform deep feature extraction and physiological effect prediction, and outputting predicted values ​​of multiple physiological effect indicators, including blood flow, muscle response, and pain perception; Adaptive optimization module: dynamically adjusts pressure stimulation parameters based on prediction results and individual characteristics; Intelligent control interface: used to generate personalized stimulation plans, monitor data in real time, and record therapeutic effects.

Images