Dynamic adjustable micro-current stimulation and training method and system for ophthalmology

Abstract

The application provides a dynamic adjustment micro-current stimulation and training method and system for ophthalmology. The application integrates various sensors through a wearable periocular device, collects physiological data such as corneal impedance, intraocular pressure signals, blinking behavior and visual evoked potential in real time, combines denoising, normalization, dimensionality reduction and multi-modal fusion processing, and constructs a standardized eye physiological state vector. The vector and incremental clustering are used to identify the micro-current stimulation tolerance inflection point, generate an individual safety upper limit initial value, combine a hierarchical fuzzy reasoning system and a dynamic comprehensive risk index, realize adaptive adjustment of the safety current limiting threshold, and finally smooth the micro-current output through a PID controller. The system can also update the model through cross-course memory data closed loop to improve the personalization and safety of the regulation. The application can dynamically adapt to individual physiological differences, effectively guarantee the biological safety of micro-current stimulation, and meet the needs of continuous and long-term adaptive fine intervention.

Description

Technical Field

[0001] The field of ophthalmic intelligent biofeedback and microcurrent stimulation modulation technology, especially a dynamic microcurrent stimulation and training method and system for ophthalmology. Background Technology

[0002] Currently, ophthalmic microcurrent stimulation technology has been applied in areas such as visual function recovery, improvement of visual fatigue, and myopia control. Most mainstream ophthalmic microcurrent stimulation systems adjust the stimulation current output based on fixed empirical thresholds or static safety limits set by experts. These systems generally use single or limited physiological parameters, such as a pre-set maximum allowable current intensity or an upper limit for a single treatment session, as safety boundaries, lacking consideration of the user's individual eye condition and dynamic changes. Some existing products integrate basic physiological monitoring functions, such as real-time measurement of intraocular pressure, surface temperature, or simple recording of the user's subjective feelings; however, this monitoring data is usually only used for abnormal alarms or post-treatment review, rather than for real-time dynamic adjustment of safety boundaries.

[0003] With the widespread adoption of wearable physiological sensors and biofeedback algorithms, some research has emerged that combines physiological monitoring such as ECG, intraocular pressure, and electroencephalography with neural stimulation, attempting to improve the intelligence of stimulation protocols. However, within the current technological framework, the safety control logic of most systems remains based on standard parameters plus empirical thresholds. This means that they rely on fixed physiological safety limits and a one-size-fits-all approach to power or current coverage, underestimating the differences in individual eye tolerance and dynamic state changes.

[0004] Existing related technologies mainly focus on the following application scenarios: One type of product targets visual training and myopia relief, promoting ocular blood flow and neural regulation through microcurrent stimulation. These typically employ fixed parameter schemes below generally accepted physiological thresholds, simultaneously monitoring basic intraocular pressure or tear secretion, and only shutting down when alarms are triggered by exceeding the limits. Another type of research-oriented equipment integrates simple physiological sensing, such as eye movement or blink detection, and initially achieves manual adjustment based on bandwidth-based individual differences; however, adaptive modeling for dynamic stimulation boundaries remains insufficient. Overall, these technologies are suitable for ordinary people with healthy eyes or patients without special physiological burdens. For individuals with extreme physiological sensitivity, underlying ocular diseases, or drastic short-term fluctuations in condition, existing static threshold strategies cannot accurately guarantee safety and are insufficient to fully leverage the individualized advantages of microcurrent stimulation.

[0005] Existing technologies suffer from the following prominent problems: First, the stimulation intensity boundary mainly relies on experience or standard physiological limits, making it difficult to automatically adapt to subtle changes in the user's real-time eye state. Differences in tolerance limits among different users are ignored, leading to insufficient timeliness for low stimulation intensity and easy induction of mild eye stress or even damage for high intensity. Second, multimodal physiological monitoring data is only used for reference or single-point alarms, failing to serve as input for deep fusion analysis in modeling, resulting in a severe lack of personalized real-time decision-making. Third, stimulation safety parameters are usually manually set or periodically adjusted, lacking deep learning, memory, and dynamic convergence capabilities based on historical data, making it impossible to track and optimize long-term control trajectories for individual users. Furthermore, most existing devices have limited functionality, relying primarily on simple negative feedback alarms, and cannot provide real-time, continuous, and uninterrupted intelligent dynamic control.

[0006] Therefore, there is an urgent need for a microcurrent stimulation modulation method that can integrate multi-source real-time physiological data, dynamically sense individual eye sensitivity, and adaptively determine safety boundaries. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for dynamically adjustable microcurrent stimulation and training in ophthalmology, thereby resolving the problems mentioned in the background section.

[0008] In a first aspect, the present invention provides a method for dynamically adjustable microcurrent stimulation and training in ophthalmology, comprising the following steps: S1: Acquire multimodal physiological state data of the user's eyes; S2: Preprocess the multimodal physiological state data to generate a standardized eye state vector; S3: Based on the standardized eye state vector, the incremental clustering algorithm is used to analyze the changing trends of the latency shift and amplitude variation coefficient of visual evoked potentials during the first low-intensity exploratory stimulation, identify the tolerance inflection point characteristics of the neurophysiological response from the stable zone to the stress transition zone, and form an individualized safety upper limit initial value. S4: Based on the individualized safety upper limit initial value, the multi-dimensional eye state vector collected in real time is weighted and fused using a sliding time window mechanism to generate a dynamic comprehensive risk index. S5: Input the dynamic comprehensive risk index into the hierarchical fuzzy inference system. The hierarchical fuzzy inference system maps the dynamic comprehensive risk index to the corresponding safe current limit threshold based on the preset physiological risk level mapping rules and multi-condition logic judgment mechanism, which serves as the target constraint condition for microcurrent stimulation intensity regulation. S6: Based on the aforementioned safe current limiting threshold, the microcurrent output signal is smoothly limited and adjusted by a PID controller to generate a personalized stimulation current intensity command that adapts to the current physiological state of the eye. S7: After each training cycle, monitor the regression curve changes of ocular state parameters during the recovery period, and store the regression curve changes as cross-treatment memory data in a local encrypted file for security boundary preloading before the next session. S8: When a subsequent session is started, the cross-treatment memory data in the local encrypted file is called to reconstruct the initial value of the initial individualized safety upper limit, and the judgment parameters of the hierarchical fuzzy inference system are dynamically optimized in combination with the real-time input multimodal physiological state data to achieve continuous adaptive updating of the safety boundary model.

[0009] Secondly, the present invention provides a dynamically adjustable microcurrent stimulation and training system for ophthalmology, comprising: The data acquisition module is used to acquire multimodal physiological state data of the user's eyes; The preprocessing module preprocesses the multimodal physiological state data to generate a standardized eye state vector. The individualized safety boundary modeling module, based on the standardized eye state vector, uses an incremental clustering algorithm to analyze the changing trends of the latency shift and amplitude variation coefficient of visual evoked potentials during the first low-intensity exploratory stimulation, identifies the tolerance inflection point characteristics of the neurophysiological response from the stable zone to the stress transition zone, and forms the initial value of the individualized safety upper limit. The dynamic risk assessment module, based on the individualized safety upper limit initial value, combines a sliding time window mechanism to perform weighted fusion processing on the real-time collected multi-dimensional eye state vector to generate a dynamic comprehensive risk index. The hierarchical fuzzy reasoning module maps the dynamic comprehensive risk index to the corresponding safe current limit threshold based on the preset physiological risk level mapping rules and multi-condition logic judgment mechanism, which serves as the target constraint condition for microcurrent stimulation intensity regulation. The PID control module performs smooth amplitude limiting adjustment on the microcurrent output signal based on the safe current limiting threshold, and generates a personalized stimulation current intensity command that adapts to the current physiological state of the eye. The cross-treatment memory storage module is used to encrypt and store the regression curves of eye state parameters and related training parameters during the recovery period after each training cycle, forming cross-treatment memory data. The adaptive update module is used to call cross-treatment memory data when subsequent sessions are started, reconstruct the initial value of the individualized safety upper limit, and dynamically optimize the judgment parameters of the hierarchical fuzzy inference module in combination with real-time input multimodal physiological state data, so as to realize the continuous update of the safety boundary model.

[0010] The beneficial effects of this invention are as follows: 1. This invention utilizes a multimodal sensor array integrated into a wearable device to collect multidimensional data such as corneal impedance, intraocular pressure, blinking behavior, and visual evoked potential (VEP) baseline at the initial stage of system startup. This establishes an initial profile of individual ocular physiological sensitivity. During the first low-intensity trial stimulation, it monitors the rate of change of neuroelectrophysiological feedback, such as VEP latency shift and amplitude variation coefficient, in real time. Combined with ocular surface environmental trends, it identifies the user-specific tolerance inflection point, i.e., the key critical value for the transition of physiological response from stability to stress, as the initial value of personalized safety upper limit. This invention eliminates the reliance on empirical fixed thresholds, realizing the modeling of safety boundaries from general group to individual specificity, effectively avoiding safety hazards caused by individual sensitivity differences, and significantly improving the clinical applicability and reliability of the system. 2. This invention introduces a sliding time window weighted fusion mechanism and a hierarchical fuzzy inference system, which realizes the continuous evolution and fine control of the safety boundary. The fused comprehensive risk index is mapped to the maximum allowable output current through hierarchical fuzzy inference, and the output is smoothed and limited by a PID controller to ensure that the stimulation intensity is close to the safety limit without interrupting the treatment process. The cross-treatment memory mechanism encrypts and stores the recovery curve of the eye state after each intervention in the local archive for the preloading of the safety boundary before the next session, which significantly shortens the adaptive convergence time and solves the problems of poor robustness and lag response of traditional systems in complex physiological environments. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the method of Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the initial value of the individualized safety upper limit in Embodiment 1 of the present invention; Figure 3 This is a flowchart illustrating the generation process of the dynamic comprehensive risk index in Embodiment 1 of the present invention. Detailed Implementation

[0012] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings: Example 1 like Figure 1 As shown, this embodiment provides a method for dynamically adjustable microcurrent stimulation and training in ophthalmology, including the following steps: S1: Acquire multimodal physiological state data of the user's eyes; In this embodiment, the multimodal physiological state data includes corneal impedance, non-contact intraocular pressure monitoring signals, blink frequency and amplitude, and baseline response characteristics of visual evoked potentials.

[0013] In this embodiment, the acquisition of multimodal physiological data is achieved through multiple sensor arrays integrated into a wearable periocular device, including an impedance corneal wetness sensor, a non-contact intraocular pressure sensor, an ocular surface behavior imaging acquisition unit, and a surface electrode array. A synchronous sampling control algorithm is adopted, with a synchronization period of 5ms and a clock tolerance of ±0.1ms, to achieve time-series synchronous acquisition of corneal impedance, intraocular pressure signals, blink behavior video streams, and visual evoked potential electrical signals. Multiple raw signals are transmitted to the edge computing unit in real time through a low-latency data bus to avoid phase mismatch in cross-modal data acquisition.

[0014] In this embodiment, the acquired corneal impedance signal is processed by environmental temperature and humidity compensation. The data collected synchronously by the temperature and humidity sensor is used to correct the measurement deviation caused by environmental fluctuations based on the preset compensation function model, and an environmentally corrected corneal impedance value is generated to improve the accuracy of humidity assessment. Motion artifact filtering was performed on the non-contact intraocular pressure monitoring signal. An adaptive notch filter algorithm was used to eliminate interference components introduced by head micro-movements or blinking. A stable intraocular pressure fluctuation trend sequence was extracted as an effective input parameter to reflect the dynamic changes of intraocular pressure. Based on time-domain analysis, features of blinking behavior signals are extracted, the blinking frequency per unit time and the envelope curve of the closing duration and amplitude of each blink are calculated, and a blinking dynamic feature vector that quantitatively represents the fatigue state of the ocular surface is generated. Baseline drift correction and signal-to-noise ratio enhancement were performed on the visual evoked potential signal. Wavelet denoising combined with average superposition technology was used to extract stable VEP latency and N75 and P100 wave amplitude features, generating a baseline response feature parameter set that can be used for neural response sensitivity assessment.

[0015] S2: Preprocess the multimodal physiological state data to generate a standardized eye state vector; specifically including the following steps: S2.1: Based on the corneal impedance, non-contact intraocular pressure monitoring signal, blink frequency and amplitude, and visual evoked potential baseline response characteristics obtained in S1, the original multimodal physiological state data is decomposed in the time-frequency domain using a wavelet threshold denoising algorithm. The detail coefficients of each modality of physiological state data are extracted and adaptive soft thresholding is applied to suppress high-frequency noise and motion artifact interference, thereby obtaining a denoised multimodal physiological signal sequence. In this embodiment, a hierarchical decomposition is performed on each modal signal using discrete wavelet transform, decomposing the signal into approximate coefficients and detail coefficients of different frequency bands. An adaptive soft threshold function is applied in the detail coefficient domain to suppress high-frequency noise components while preserving the effective feature structure of the physiological response, resulting in a multi-band coefficient group with minimized denoising residual.

[0016] S2.2: For the denoised corneal impedance signal, a temperature and humidity compensation model is introduced to correct environmental parameters. The temperature and humidity compensation model constructs a three-dimensional mapping lookup table of temperature-humidity-impedance based on historical calibration data. The compensation factor is obtained by interpolation based on the real-time collected environmental temperature and humidity values. The corneal impedance signal is corrected point by point to eliminate measurement drift caused by non-physiological factors and output an environmentally compensated corneal impedance signal. In this embodiment, a bilinear interpolation algorithm (parameters: ΔT≤0.5°C for nearby temperature sampling points, ΔH≤1% for nearby humidity sampling points) is used to locate the compensation factor corresponding to the real-time collected ambient temperature and humidity values ​​in the lookup table, thereby estimating the environmental offset of the current impedance sampling point and obtaining a point-by-point compensation coefficient matrix.

[0017] The point-by-point multiplication correction method is adopted to multiply the compensation coefficient matrix with the corresponding elements of the noise-reduced corneal impedance signal to generate an impedance signal sequence corrected for temperature and humidity, thereby offsetting the measurement drift effect caused by non-physiological environmental factors.

[0018] The corrected impedance signal sequence is smoothed by a sliding window smoothing algorithm to suppress high-frequency numerical fluctuations introduced by the compensation process, thus forming a smooth environmental compensation impedance dataset.

[0019] S2.3: Perform piecewise moving mean filtering on the environmentally compensated corneal impedance signal, non-contact intraocular pressure monitoring signal, blink behavior parameters and visual evoked potential baseline response characteristics to further smooth low-frequency fluctuation components, improve the signal-to-noise ratio, and resample each modal signal to a unified time reference to generate a time-aligned multidimensional physiological signal matrix. S2.4: Based on the time-aligned multidimensional physiological signal matrix, the maximum-minimum normalization method is used to map the signals of each dimension to the [0,1] interval. The normalization parameter is dynamically updated according to the historical maximum and minimum values ​​of the ontology to avoid cross-session scale bias and generate a normalized multimodal signal vector with a unified numerical range. S2.5: Perform principal component analysis to reduce the dimensionality of the normalized multimodal signal vector, retain the principal components with a cumulative contribution rate greater than 95%, remove redundant information and collinearity interference, and output a low-dimensional, compact, standardized eye state vector.

[0020] S3: Based on the standardized eye state vector, an incremental clustering algorithm is used to analyze the changing trends of the latency shift and amplitude variation coefficient of visual evoked potentials during the first low-intensity exploratory stimulus, identifying the tolerance inflection point characteristics of the neurophysiological response transitioning from the stable zone to the stress transition zone, and forming an individualized initial safety upper limit value; such as Figure 2 As shown, the specific steps include the following: S3.1: Acquire the visual evoked potential signals continuously collected during the first low-intensity exploratory stimulation, and perform bandpass filtering and artifact removal processing on the visual evoked potential signals to eliminate electromyographic interference and baseline drift effects, thereby obtaining a clean visual evoked potential time series data sequence. In this embodiment, during the initial low-intensity exploratory stimulation phase, the continuous visual evoked potential signal collected by the surface electrode is used as the input object, and the signal sampling frequency is set to 1024Hz to ensure the integrity of waveform details.

[0021] A bandpass filtering algorithm (parameters: passband range 0.5Hz-30Hz, filter type linear phase FIR, order 64) is used to achieve frequency domain limitation of the original visual evoked potential signal, retain the low-frequency and mid-frequency components related to the neural response, and suppress high-frequency noise and DC drift.

[0022] In some embodiments, the Independent Component Analysis (ICA) algorithm (parameters: decomposition dimension set to the number of sampling channels, non-Gaussian metric function selected as Kurtosis) is used to perform source separation processing on the bandpass filtered signal, separating out and removing electromyographic interference sources unrelated to visual stimuli, and obtaining clean neural response signal components.

[0023] In some embodiments, a baseline correction processing method is used (parameter: baseline interval selected 200ms before stimulation), the average signal value within the interval is calculated and the data in the entire time domain is subtracted point by point to achieve the function of eliminating low-frequency baseline drift and obtain a signal sequence with a stable baseline.

[0024] In some embodiments, an artifact detection and removal algorithm (parameter: threshold selected as average amplitude ± 3 times standard deviation) is used to identify transient abnormal waveforms in the baseline-corrected signal, and to perform linear interpolation replacement on bands exceeding the threshold, thus preserving the continuity and authenticity of the neural response.

[0025] S3.2: Based on the pure visual evoked potential time series data sequence, calculate the latency shift and amplitude variation coefficients of N1 and P1 bands within each stimulation cycle, and generate a time series feature vector containing the dynamic response characteristics of neurophysiology, as a quantitative input index for evaluating stimulation tolerance. Based on the pure visual evoked potential time series data sequence output by step S3.1, a window segmentation extraction algorithm (parameters: window length is consistent with the trial stimulus period, step size is equal to the sampling period) is used to realize the time series data segmentation of each independent stimulus period, ensuring the period boundary alignment of subsequent feature calculations.

[0026] Using peak detection and waveform calibration methods (parameters: target band N1 / P1, threshold set by the root mean square value of baseline noise), the characteristic peak and trough times within each stimulation cycle are identified, generating an initial latency sequence as a basis for subsequent offset calculations.

[0027] The change in latency relative to the baseline is calculated using the latency offset formula. Quantification, that is: ; in, For the first Latency measurement value of each stimulation cycle This is the baseline latency period.

[0028] In some embodiments, the variance normalization method is used to calculate the coefficient of variation of the periodic variation of the amplitudes in the N1 and P1 bands. Quantification, that is: ; in, The standard deviation of the amplitude of the current stimulation cycle group. This represents the average amplitude.

[0029] By performing periodic traversal operations, the latency offset and amplitude variation coefficient of each period are combined in chronological order to form two-dimensional feature points, thus creating a time-series feature vector containing the dynamic response characteristics of neurophysiology.

[0030] The missing value imputation algorithm (parameters: a combination of forward filling and spline interpolation) is used to correct feature gaps caused by abnormal periods, ensuring the integrity of time series feature vectors and data continuity.

[0031] S3.3: Input the time series feature vector into the incremental clustering algorithm model. The incremental clustering algorithm model dynamically divides the clusters based on the change in data point density within the sliding window, identifies the key time nodes when the response features transition from a stable clustered state to an outlier dispersed state, and outputs the initially determined tolerance state transition interval. S3.4: Combine corneal impedance trend and blink frequency feedback to perform multimodal verification on the tolerance state transition interval, eliminate false stress response caused by ocular surface dryness or blinking, optimize the tolerance inflection point positioning accuracy, and generate a tolerance inflection point candidate set corrected by physiological context. This embodiment uses a multimodal consistency verification algorithm (parameters: corneal impedance trend sequence, blink frequency time vector, tolerance state transition interval) to perform physiological context association analysis on the transition interval based on the tolerance state transition interval and the standardized ocular state vector.

[0032] In some embodiments, a time trend matching method for environmentally compensated corneal impedance signals (parameters: sliding window length 10 seconds, matching threshold 0.8) is used to achieve synchronous comparison of corneal wetness fluctuations and transition interval time series, and obtain the time offset data of impedance change rate peak and tolerance interval.

[0033] In some embodiments, a blink dynamics anomaly detection algorithm (parameters: frequency deviation threshold 3 times standard deviation, closing duration deviation threshold 1.5 times standard deviation) is used to identify transient anomalies in blinking behavior and generate a blink anomaly probability sequence as a basis for judging pseudo-stress response.

[0034] In some embodiments, a joint verification decision function is constructed to logically cross-match the peak value of corneal impedance change rate with the blink abnormal probability sequence. If the impedance drops sharply and is accompanied by a high probability blinking event within the transition interval, the interval is marked as pseudo-stress and the effective tolerance transition candidate set is removed.

[0035] S3.5: Determine the initial value of the individualized safety upper limit based on the minimum effective stimulus intensity value in the candidate set of tolerance inflection points.

[0036] S4: Based on the aforementioned individualized safety upper limit initial value, and combined with a sliding time window mechanism, the real-time collected multi-dimensional eye state vector is weighted and fused to generate a dynamic comprehensive risk index; such as... Figure 3 As shown, the specific steps include the following: S4.1: Based on the initial value of the individualized safety upper limit and the standardized eye state vector, an initial multimodal decision weight matrix is ​​constructed, in which the decision weights of each modality are initialized to be equally distributed to ensure unbiased fusion in the early stage where historical feedback data is lacking, and the initial weight configuration parameters are generated. In this embodiment, based on the initial value of the individualized security upper limit and the standardized eye state vector collected for the first time in the current session, a matrix initialization method (parameters: total number of modalities, initial weight value) is used to construct the multimodal decision weight matrix.

[0037] In some embodiments, a weight balancing allocation algorithm is used to initialize the weights of each modality to an equal distribution, thereby obtaining initial weight matrix data that satisfies the normalization constraint.

[0038] In some embodiments, a modality index mapping table generation program is used to realize the corresponding binding relationship between each modality physiological signal and the elements of the decision weight matrix, and to generate a mapping matrix structure with address identifiers.

[0039] In some embodiments, a matrix consistency check algorithm is used to check the numerical stability of the initial weight matrix and output a copy of the weight configuration matrix that has passed the check.

[0040] S4.2: The real-time acquired multidimensional eye state vector is segmented by a sliding time window mechanism. The time window length is set to 30 seconds and slides in 5-second increments to generate a time-series sequence of eye state segments in order to capture the dynamic evolution trend of physiological parameters and obtain a continuous state observation dataset. S4.3: Using the eye state segment sequence output by the sliding time window, combined with the cross-treatment memory data recorded in the previous training cycle, calculate the correlation coefficient and error contribution rate of each modality signal in historical risk prediction, execute the weight optimization algorithm to dynamically update the multimodal decision weight matrix, and generate the adaptive weight vector under the current session. The temporal sequence of eye state fragments obtained by sliding time window segmentation is associated and bound with cross-treatment memory data input as the original dataset for weight optimization; The Pearson correlation coefficient calculation method (parameters: the current value sequence of each modal signal and the historical risk prediction result sequence within the sliding time window) is used to measure the linear correlation of each modal signal in risk prediction and output the correlation coefficient matrix.

[0041] In this embodiment, the mean square error calculation method is used. (Parameter: the difference between the predicted risk index and the actual stimulus tolerance response), to evaluate the error contribution rate of each modality signal and obtain the error contribution rate vector; that is: In the formula, For the first The actual stimulus tolerance response value at time t, For the corresponding risk prediction output value, This represents the number of samples within the time window.

[0042] In this embodiment, the new weights are calculated using a weighted update formula (parameters: current modality weights, correlation coefficients, and error contribution rates), and an adaptive weight vector is generated. ,Right now: in, For the first The current weights of the modality, For the first Modal correlation coefficient, For the first Modal error contribution rate.

[0043] In this embodiment, the weight vector is mapped to a standardized space with a sum of 1 by a normalization process (parameter: all updated modal weights) to maintain the relative stability of the influence of multimodal decisions and achieve a biased weight distribution.

[0044] In this embodiment, the correlation and error indicators from the previous step are transformed into an adaptive weight vector through a weight optimization algorithm and normalization processing, thereby achieving the expected technical effect of dynamically adjusting the modal contribution ratio for risk index calculation in the current session.

[0045] S4.4: Perform a weighted inner product operation on the current eye state segment and the adaptive weight vector, and fuse the corneal impedance trend, intraocular pressure fluctuation, blink abnormality index and visual evoked potential variability to highlight the decision influence of high-risk modalities and output the weighted fusion multidimensional state comprehensive evaluation value. S4.5: Based on the initial value of the individualized safety upper limit, the multidimensional state comprehensive evaluation value is subjected to nonlinear normalization mapping, which is compressed to the [0,1] interval and transformed into an interpretable risk index output to generate a dynamic comprehensive risk index; In this embodiment, the comprehensive evaluation value is monotonically and nonlinearly compressed using an exponential normalization function to ensure that the risk growth rate significantly increases as the mapping curve approaches the initial value of the safety upper limit, i.e.: in, This is the normalized risk index; This is a weighted and integrated comprehensive evaluation value; A scaling factor to control the steepness of the curve; This is a displacement factor used to align the initial values ​​of the individualized safety upper limit.

[0046] In this embodiment, the mapping slopes of the low-risk area and the high-risk area are set differently by using a piecewise nonlinear correction function, so as to achieve smooth output in the low-risk area and rapid warning output in the high-risk area.

[0047] In this embodiment, the peak fluctuations of the instantaneous risk index are suppressed by smoothing the normalization result, thereby improving the stability of the risk index as an input to the fuzzy inference engine.

[0048] S5: Input the dynamic comprehensive risk index into the hierarchical fuzzy inference system. The hierarchical fuzzy inference system maps the dynamic comprehensive risk index to the corresponding safe current limiting threshold based on the preset physiological risk level mapping rules and multi-condition logic judgment mechanism, which serves as the target constraint condition for microcurrent stimulation intensity regulation. Specifically, it includes the following steps: S5.1: Based on the dynamic comprehensive risk index, construct the input variable set of the hierarchical fuzzy inference system; the input variable set includes corneal impedance change rate, intraocular pressure fluctuation amplitude, abnormal blink frequency and visual evoked potential variation coefficient. The Z-score standardization method is used to normalize the variables of each dimension to eliminate the difference in dimensions and generate a fuzzy input vector under a unified scale. In this embodiment, based on a dynamic comprehensive risk index, corneal impedance change rate, intraocular pressure fluctuation amplitude, abnormal blink frequency, and visual evoked potential variation coefficient are selected as the input data sources for the hierarchical fuzzy inference system. A feature selection algorithm is used to filter out input dimensions that are not significantly correlated with the risk index, effectively simplifying the input set. A feature statistics method (parameters: window length = 60 seconds, step size = 10 seconds) is used to calculate the mean and standard deviation of each input dimension within a set time window to capture short-term fluctuation patterns and enhance risk sensitivity.

[0049] In this embodiment, the Z-score normalization method is used to perform a normalization operation on each input dimension signal x, that is: In the formula, This is the current input value; This is the historical average for this dimension; This represents the historical standard deviation of this dimension; This is the data after standardization.

[0050] For the standardized data, an outlier suppression algorithm is used to replace Z-score data with absolute values ​​greater than 3 with the median value of the nearest time point in order to reduce the interference of abnormal fluctuations on the blurring.

[0051] The processed standardized data from each dimension are combined into a fuzzy input vector at a unified scale and passed to the data entry point of the fuzzy inference engine through the dimension mapping interface, achieving format compatibility with subsequent multi-level fuzzy logic judgments.

[0052] S5.2: Perform two-level fuzzification mapping processing on the fuzzified input vector. The first level is based on the preset eye physiological state membership function library, which maps the parameters of each dimension to low, medium and high risk fuzzy sets to generate a preliminary risk attribution matrix. The second stage introduces a weight adjustment factor to dynamically adjust the contribution ratio of each modal parameter in the fusion decision based on the historical prediction accuracy, and calculates a weighted fuzzy comprehensive evaluation vector to improve the response sensitivity to key risk factors. S5.3: Based on the weighted fuzzy comprehensive evaluation vector, a hierarchical multi-condition logic judgment mechanism is executed. First, in the primary judgment layer, parallel rule matching is performed according to the preset physiological risk level mapping rule set to identify whether there is a high-risk combination mode that triggers emergency current limiting. If it is satisfied, the minimum safe current limiting threshold is immediately output as an emergency protection response. Based on the weighted fuzzy comprehensive evaluation vector, a rule matching algorithm (parameter: preset physiological risk level mapping rule set) is used to achieve parallel retrieval and trigger condition determination of high-risk combination patterns; The pattern combination generator (parameter: risk factor lookup table) combines corneal impedance, intraocular pressure fluctuation, blink abnormality and VEP variation coefficient into a multidimensional risk pattern vector, and matches it one by one with the high-risk pattern library defined in the rule set to obtain the pattern matching result matrix.

[0053] The Boolean logic operation unit (parameters: AND / OR logic and weight threshold) is used to perform logical operations to identify high-risk patterns that meet the combined triggering conditions in the pattern matching result matrix and generate triggering flag signals.

[0054] The safety current limiting module (parameter: minimum safe current threshold) is controlled by a trigger flag signal to force the output of a preset minimum safe current limit as an emergency protection response when a high-risk mode trigger condition is met.

[0055] S5.4: If the emergency current limiting mode is not triggered, the secondary judgment layer is entered, and the nonlinear mapping rule table in the individualized risk response knowledge base is called to map the weighted fuzzy comprehensive evaluation vector into a continuous safety current limiting threshold. In this embodiment, based on the weighted fuzzy comprehensive evaluation vector, the nonlinear mapping rule table in the individualized risk response knowledge base (parameters: historical physiological response curve set, modal weight coefficient vector, cross-treatment memory data index) is called to realize the function of converting multidimensional risk fuzzy quantitative data into continuous safe current limiting threshold.

[0056] In this embodiment, a multidimensional spline interpolation algorithm (parameters: the node set is derived from the tolerance characteristic curves of different sensitive groups, the order is set to third order, and the boundary condition adopts natural boundary) is used to achieve a smooth mapping of the weighted fuzzy evaluation vector to the safe current limiting space, and obtain a continuous correspondence surface between cross-modal risk and stimulation tolerance.

[0057] In this embodiment, parameter adaptation processing is performed using the recovery period regression features in the cross-treatment memory data, and the local gradient in the mapped surface is calibrated using the least squares fitting method, so that the generated amplitude limit output is more in line with the current user's recent physiological recovery pattern, thereby reducing state drift between sessions.

[0058] In this embodiment, based on the combination of the weighted fuzzy comprehensive evaluation vector R and the nonlinear mapping rule table F, the continuous safety current limiting threshold is calculated using the following mathematical relationship: in, For safe current limiting threshold, For weighted fuzzy comprehensive evaluation vector, A table of nonlinear mapping functions adapted to the parameters.

[0059] In this embodiment, a constrained optimization algorithm is used to generate a personalized output target flow limit threshold for the current user state. The objective function is to minimize the squared difference between the limit value and the physiological tolerance threshold, and the constraint condition is that it does not exceed the initial value of the individualized safety upper limit.

[0060] S5.5: Perform a smooth transition check on the safe current limiting threshold, compare the rate of change between the current output value and the limiting value of the previous period, and if it exceeds the preset allowable gradient range of PID regulation, then use the exponential decay interpolation algorithm to generate an intermediate transition threshold.

[0061] In this embodiment, the rate of change of the stimulus intensity is quantitatively evaluated by using the differential rate comparison method (parameters: current threshold, current output value, previous time period limit value) on the safe current limit threshold calculated by the hierarchical fuzzy inference system.

[0062] In this embodiment, the consistency between the current threshold change magnitude and the preset allowable gradient range of PID control is determined by calculating the differential rate formula, i.e.: in This is the current limiting threshold. The threshold value for the previous time period; In this embodiment, an exponential decay interpolation algorithm is used to smooth out threshold changes that exceed the gradient range, i.e.: in, The decay rate constant is This is the interpolation time step; This is the amplitude limiting threshold after smoothing.

[0063] By using a dynamic time step adjustment strategy (parameters: physiological response delay time, shortest recovery period, stimulation cycle length), the time of the decay interpolation process is controlled to ensure that the interpolation curve matches the user's neurophysiological response buffer.

[0064] S6: Based on the aforementioned safe current limiting threshold, a PID controller is used to smoothly limit the microcurrent output signal, generating a personalized stimulation current intensity command adapted to the current physiological state of the eye; specifically including: S6.1: Based on the safe current limiting threshold output from the previous steps, obtain the set target current intensity command of the current microcurrent stimulation system and use it as the initial reference input for the proportional-integral-derivative adjustment of the PID controller to establish a stimulation intensity benchmark consistent with the individualized safety boundary. The safe current limiting threshold output by the hierarchical fuzzy inference system, together with the target setting parameters of the current session of the microcurrent stimulation system, constitutes the input conditions for this sub-step.

[0065] The parameter parsing module is used to perform structured parsing of the target current intensity command to achieve a precise quantitative expression of the original set value.

[0066] In this embodiment, a safety boundary comparison algorithm (parameters: safety current limit threshold, target current intensity command) is used to quantitatively evaluate the difference between the set value and the safety upper limit and obtain the allowable initial adjustment range data.

[0067] In this embodiment, the preprocessing module performs preliminary amplitude limiting calculations on the target current intensity to generate a pre-adjusted target value that meets the safety threshold constraints, thereby avoiding out-of-bounds stimulus settings at the input of the PID controller.

[0068] In this embodiment, the time sequence of the pre-adjusted target value is aligned with the main clock of the control system based on the signal synchronization algorithm, forming a reference input vector that can be directly used for proportional-integral-differential operations.

[0069] S6.2: The target current intensity command and the actual output micro-current feedback signal are sampled and compared. The real current value in the output circuit is collected in real time using a high-precision current sensor. The deviation between the two is calculated, and an error time series signal is generated as the dynamic adjustment input basis for the PID control algorithm.

[0070] S6.3: Based on the error time series signal, execute the incremental PID control algorithm to calculate the contribution of the proportional term (P), integral term (I), and derivative term (D) to the current control cycle, respectively. The proportional term is used to quickly respond to instantaneous deviations, the integral term is used to eliminate steady-state error accumulation, and the derivative term is used to predict trend changes and suppress overshoot, thereby generating a comprehensive control incremental signal.

[0071] Based on the error time series signal acquired and generated by a high-precision current sensor, an incremental PID control algorithm (parameters: proportional coefficient Kp, integral coefficient Ki, derivative coefficient Kd) is adopted to realize the independent calculation of the three components of the current control cycle, so as to ensure the sensitivity and stability of the adjustment process.

[0072] The proportional term calculation module performs a linear amplification operation on the instantaneous deviation at the end of the current cycle to obtain the proportional component used to directly correct the output trend. ,Right now: in, This represents the instantaneous error value for the current sampling period.

[0073] In this embodiment, the integral term calculation module performs a weighted summation operation on the cumulative error of all historical sampling periods to obtain the integral component used to eliminate steady-state error. ,Right now: ; in, This is the time variable for integration operations.

[0074] In this embodiment, the differential term calculation module performs a weighted operation on the current period error change rate to obtain the differential component used to suppress overshoot and improve dynamic response. ,Right now: In this embodiment, the proportional component, integral component, and derivative component are numerically superimposed using a multi-component weighted summation module to obtain the comprehensive control increment signal. As the output current command adjustment amount under the current control cycle: that is: 。

[0075] S6.4: The integrated control incremental signal is superimposed on the output current value of the previous control cycle to generate an updated real-time micro-current output command. The command is then subjected to amplitude clamping to ensure that it does not exceed the safe current limit threshold output by the hierarchical fuzzy inference system, thus forming a smooth and restricted personalized stimulation current intensity output.

[0076] The integrated control increment signal output by the incremental PID control algorithm and the micro-current output value of the previous control cycle are used as input conditions to initialize the operation parameters of the amplitude limiting processing module, including the safe current limiting threshold and the allowable gradient change range output by the hierarchical fuzzy inference system.

[0077] A cycle-by-cycle superposition method is used to predict and update the real-time microcurrent output value for the new cycle, and an unlimited initial output command matrix is ​​generated in the digital control domain through a numerical mapping strategy.

[0078] The amplitude clamping algorithm constrains the command value within the safety boundary; the amplitude-limited smooth transition algorithm ensures the continuity of the output current across cycles; and the combination of clamping and smoothing control transforms the updated microcurrent output command into a smooth and safe personalized stimulation intensity signal, achieving a technical effect that conforms to the physiological adaptation of the eye.

[0079] S6.5: The microcurrent output command after amplitude limiting is transmitted to the digital-to-analog converter module and the constant current source drive circuit to complete the conversion of digital control signal to analog current signal, and output a low-fluidity, high-stability microcurrent stimulation signal that is adapted to the current physiological state of the eye, so as to realize the closed-loop precise mapping from safety decision to physical execution.

[0080] S7: After each training cycle, monitor the regression curve changes of ocular state parameters during the recovery period, and store these regression curve changes as cross-treatment memory data in a local encrypted archive for security boundary preloading before the next session; specifically including: S7.1: Based on the real-time multimodal ocular status data stream after the training cycle ends, start the recovery period monitoring time sequence window, set the sampling frequency to once per minute, and continuously collect corneal impedance, intraocular pressure fluctuation value, blink frequency and visual evoked potential baseline drift for at least 30 minutes as the original input sequence for the recovery process.

[0081] S7.2: The original input sequence of the recovery process is detrended and outlier is removed. The moving median filtering algorithm is used to eliminate instantaneous interference signals. Time alignment and interpolation compensation are performed on the parameters of each dimension to generate a structured eye state matrix during the recovery period.

[0082] S7.3: Based on the structured recovery period ocular state matrix, calculate the normalized decay exponent and half-recovery time (T1 / 2) for each parameter dimension, fit a double exponential decay model to extract key morphological features of the regression curve, including initial offset, decay rate constant and steady-state residual term, to form a quantitative characterization vector comparable across treatment courses.

[0083] S7.4: Associate and bind the quantized representation vector with the safe current limiting threshold, stimulation intensity instruction sequence and individualized safety upper limit initial value used in this training, encapsulate it into a structured cross-treatment memory data package, and attach a timestamp and user identity encryption identifier.

[0084] In this embodiment, based on the quantized representation vector, the safe current limit threshold in the current training cycle, the complete stimulus intensity instruction sequence, and the initial value of the individualized safety upper limit, a data association mapping method is used to realize the logical binding of multidimensional recovery period features and execution parameters.

[0085] In this embodiment, a fingerprint is generated for the safe current limiting threshold and the stimulus intensity instruction sequence using a hash verification algorithm, and the fingerprint is embedded into the metadata area of ​​the quantization representation vector to achieve parameter consistency verification and anti-tampering protection.

[0086] In this embodiment, a serialization encoding method is used to achieve structured encapsulation of various types of data, and a timestamp field and a user identity encryption identifier field are inserted during the encapsulation process. The timestamp field is generated based on a high-precision real-time clock module, and the user identity encryption identifier field is generated with a unique code based on a public key encryption algorithm.

[0087] In this embodiment, the field mapping verification strategy is used to verify whether the field positions and type definitions of the encapsulated data packet are consistent with the preset data packet structure template, and the data packet is marked as cross-treatment memory data after the verification is passed.

[0088] S7.5: The structured cross-treatment memory data packet is transmitted to the local encrypted storage module, and the data encryption writing operation is performed using the AES-256 algorithm. The data is stored in a dedicated memory archive area in the non-volatile memory for subsequent session startup to achieve preloading and fast convergence of the security boundary model.

[0089] S8: Upon initiation of a subsequent session, the cross-treatment memory data from the locally encrypted archive is invoked to reconstruct the initial value of the individualized safety upper limit. This is combined with real-time input multimodal physiological state data to dynamically optimize the decision parameters of the hierarchical fuzzy inference system, achieving continuous adaptive updates to the safety boundary model. Specifically, this includes: S8.1: Based on the cross-treatment memory data stored in the local encrypted archive, read the regression curve of ocular state parameters during the recovery period after each intervention in the historical training cycle. The parameters include corneal impedance recovery slope, intraocular pressure stabilization time, duration of blink frequency returning to normal, and baseline drift of visual evoked potentials. Use curve fitting algorithm to generate individualized physiological recovery pattern feature vectors as prior knowledge input for the initialization of the safety boundary model.

[0090] S8.2: Based on the individualized physiological recovery pattern feature vector and the standardized eye state vector collected for the first time in the current session, perform similarity matching calculation, use the dynamic time warping (DTW) algorithm to evaluate the temporal proximity between the current initial state and the historical recovery pattern, output the pattern matching index, and reconstruct the initial value of the initial individualized safety upper limit accordingly to reflect the current potential tolerance level of the eye to microcurrent stimulation.

[0091] S8.3: The initial value of the individualized safety upper limit is used as the boundary benchmark and input into the high-level rule base of the hierarchical fuzzy inference system to trigger the initial weight configuration process based on the physiological risk level mapping rule, and generate the initial comprehensive risk-current limiting mapping surface as the starting decision surface for safety boundary modeling in this round of the session.

[0092] The initial value of the individualized safety upper limit output via S8.2 is used as a benchmark parameter input to the high-level rule base of the hierarchical fuzzy inference system, and the pre-set physiological risk level mapping rule set is called to perform boundary condition loading.

[0093] The parameter normalization method is used to unify the scale of the vector composed of the initial safety upper limit value and the corresponding corneal impedance change rate, intraocular pressure fluctuation amplitude, abnormal blink frequency, and visual evoked potential variation coefficient, so as to achieve dimensional consistency of rule matching.

[0094] In this embodiment, the initial membership matrix of each input parameter under low, medium and high risk fuzzy sets is generated by the risk level membership function in the high-level rule base, and the matrix is ​​weighted and corrected by combining high-risk physiological state combination triggering factors to improve the sensitivity of the initial mapping.

[0095] In this embodiment, a weight allocation algorithm (parameter: weight allocation coefficient β based on pattern matching degree index) is used to initially set the contribution ratio of each modality parameter, wherein the value of β is jointly determined by historical pattern similarity and current input stability to achieve initial equilibrium of risk assessment.

[0096] In this embodiment, based on the membership matrix and weight allocation coefficients mentioned above, the three-dimensional surface fitting module is called to generate a comprehensive risk-current limiting mapping surface. This surface forms a continuous mapping space that can cover different physiological states, with risk level as the horizontal axis, current output limiting as the vertical axis, and weight correction factor as the depth axis.

[0097] In this embodiment, the initial membership matrix after weight correction is transformed into a current limiting decision surface with continuous risk response capability through the mapping surface construction process, thereby achieving the initialization effect of the security boundary model in the current session startup phase.

[0098] S8.4: During formal stimulation, based on the real-time acquired multidimensional eye state vector and its corresponding dynamic comprehensive risk index, the least mean square error (LMS) adaptive algorithm is used to compare the deviation between the actual neurophysiological response and the expected stable zone response, calculate the correction increment of each risk level threshold in the hierarchical fuzzy inference system, and perform online gradient adjustment on the initial mapping surface to form an optimized safe current limiting threshold output.

[0099] S8.5: Store the fuzzy inference parameter update trajectory corresponding to the optimized safe current limiting threshold into a local encrypted file, mark it as the learning result of this session, and use it as historical learning experience to participate in a new round of model reconstruction in the next session, thereby realizing the closed-loop iteration and continuous adaptive update of the decision parameters of the hierarchical fuzzy inference system.

[0100] Example 2 This embodiment provides a dynamically adjustable microcurrent stimulation and training system for ophthalmology, including: The data acquisition module is used to acquire multimodal physiological state data of the user's eyes; The preprocessing module preprocesses the multimodal physiological state data to generate a standardized eye state vector. The individualized safety boundary modeling module, based on the standardized eye state vector, uses an incremental clustering algorithm to analyze the changing trends of the latency shift and amplitude variation coefficient of visual evoked potentials during the first low-intensity exploratory stimulation, identifies the tolerance inflection point characteristics of the neurophysiological response from the stable zone to the stress transition zone, and forms the initial value of the individualized safety upper limit. The dynamic risk assessment module, based on the individualized safety upper limit initial value, combines a sliding time window mechanism to perform weighted fusion processing on the real-time collected multi-dimensional eye state vector to generate a dynamic comprehensive risk index. The hierarchical fuzzy reasoning module maps the dynamic comprehensive risk index to the corresponding safe current limit threshold based on the preset physiological risk level mapping rules and multi-condition logic judgment mechanism, which serves as the target constraint condition for microcurrent stimulation intensity regulation. The PID control module performs smooth amplitude limiting adjustment on the microcurrent output signal based on the safe current limiting threshold, and generates a personalized stimulation current intensity command that adapts to the current physiological state of the eye. The cross-treatment memory storage module is used to encrypt and store the regression curves of eye state parameters and related training parameters during the recovery period after each training cycle, forming cross-treatment memory data. The adaptive update module is used to call cross-treatment memory data when subsequent sessions are started, reconstruct the initial value of the individualized safety upper limit, and dynamically optimize the judgment parameters of the hierarchical fuzzy inference module in combination with real-time input multimodal physiological state data, so as to realize the continuous update of the safety boundary model.

[0101] The embodiments and descriptions above are merely illustrative of the principles and preferred embodiments of the present invention. Various changes and modifications may be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A method for dynamically adjustable microcurrent stimulation and training in ophthalmology, characterized in that, Includes the following steps: S1: Acquire multimodal physiological state data of the user's eyes; S2: Preprocess the multimodal physiological state data to generate a standardized eye state vector; S3: Based on the standardized eye state vector, the incremental clustering algorithm is used to analyze the changing trends of the latency shift and amplitude variation coefficient of visual evoked potentials during the first low-intensity exploratory stimulation, identify the tolerance inflection point characteristics of the neurophysiological response from the stable zone to the stress transition zone, and form an individualized safety upper limit initial value. S4: Based on the initial value of the individualized safety upper limit, the multi-dimensional eye state vector collected in real time is weighted and fused using a sliding time window mechanism to generate a dynamic comprehensive risk index. S5: Input the dynamic comprehensive risk index into the hierarchical fuzzy inference system. The hierarchical fuzzy inference system outputs the corresponding safe current limit threshold based on the preset physiological risk level mapping rules and multi-condition logic judgment mechanism, which serves as the target constraint condition for microcurrent stimulation intensity regulation. S6: Based on the safe current limiting threshold, the microcurrent output signal is smoothly limited and adjusted by the PID controller to generate a personalized stimulation current intensity command that adapts to the current physiological state of the eye. S7: After each training cycle, monitor the regression curve changes of ocular state parameters during the recovery period.

2. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 1, characterized in that: In step S1, the multimodal physiological state data includes corneal impedance, non-contact intraocular pressure monitoring signals, blink frequency and amplitude, and baseline response characteristics of visual evoked potentials.

3. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 2, characterized in that: The acquisition of the multimodal physiological data is achieved through multiple sensor arrays integrated into the wearable periocular device, including an impedance corneal wetness sensor, a non-contact intraocular pressure sensor, an ocular surface behavior imaging acquisition unit, and a surface electrode array.

4. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 1, characterized in that: Step S3 specifically includes: Bandpass filtering and artifact removal were performed on the visual evoked potential signals to obtain a clean visual evoked potential time series data sequence. Based on the pure visual evoked potential time series data sequence, the latency shift and the amplitude variation coefficients of the N1 and P1 bands within each stimulation cycle are calculated to generate a time series feature vector containing the dynamic response characteristics of neurophysiology. Input the time series feature vector into the incremental clustering algorithm model to identify the key time nodes when the response features transition from a stable clustered state to an outlier dispersed state, and output the preliminary determined tolerance state transition interval. By combining corneal impedance trends and blink frequency feedback, multimodal verification of tolerance state transition intervals is performed to generate a candidate set of tolerance inflection points corrected by physiological context. Based on the minimum effective stimulus intensity value in the candidate set of tolerance inflection points, an initial value for the individualized safety upper limit is determined.

5. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 4, characterized in that: Step S4 specifically includes: Based on the initial value of the individualized safety upper limit and the standardized eye state vector, an initial multimodal decision weight matrix is ​​constructed, and initial weight configuration parameters are generated; The multidimensional eye state vector acquired in real time is segmented and truncated to generate a time-series sequence of eye state segments; By combining the sequence of eye state fragments with cross-treatment memory data recorded in the previous training cycle, the correlation coefficient and error contribution rate of each modality signal in historical risk prediction are calculated, the multimodal decision weight matrix is ​​dynamically updated, and an adaptive weight vector for the current session is generated. The current eye state segment is weighted and inner productd with the adaptive weight vector. The corneal impedance trend, intraocular pressure fluctuation, blink abnormality index and visual evoked potential variability are fused and processed to output a weighted fused multidimensional state comprehensive evaluation value. Based on the initial value of the individualized safety upper limit, a nonlinear normalization mapping is applied to the multidimensional state comprehensive evaluation value, compressing it to the [0,1] interval to generate a dynamic comprehensive risk index.

6. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 5, characterized in that: Step S5 specifically includes: The input variable set for constructing a hierarchical fuzzy inference system based on a dynamic comprehensive risk index; A two-level fuzzification mapping process is performed on the fuzzified input vector. The first level generates a preliminary risk attribution matrix based on a preset membership function library of ocular physiological states. The second stage introduces a weighted adjustment factor to calculate the weighted fuzzy comprehensive evaluation vector; Based on the weighted fuzzy comprehensive evaluation vector, a hierarchical multi-condition logic judgment mechanism is used to identify whether there is a high-risk combination mode that triggers emergency current limiting. If the condition is met, the minimum safe current limiting threshold is immediately output as an emergency protection response. If the emergency current limiting mode is not triggered, the system enters the secondary judgment layer, calls the nonlinear mapping rule table in the individualized risk response knowledge base, and maps the weighted fuzzy comprehensive evaluation vector to a continuous safe current limiting threshold. A smooth transition check is performed on the safe current limiting threshold. The rate of change between the current output value and the limiting value of the previous period is compared. If it exceeds the preset allowable gradient range of PID regulation, an exponential decay interpolation algorithm is used to generate an intermediate transition threshold.

7. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 1, characterized in that: In step S7, after each training cycle, the monitoring of eye parameters during the recovery period is automatically started. The corneal impedance, intraocular pressure, blink frequency, and visual evoked potential baseline drift results are collected for at least 30 minutes. The normalized decay index and half recovery time are calculated and fitted with a double exponential decay model. Finally, the data is stored in a local encrypted file as cross-treatment memory data for subsequent session security boundary preloading.

8. The method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 7, characterized in that: After step S7, the method further includes: S8: When a subsequent session is started, the cross-treatment memory data in the local encrypted file is called to reconstruct the initial value of the initial individualized safety upper limit, and the judgment parameters of the hierarchical fuzzy inference system are dynamically optimized in combination with the real-time input multimodal physiological state data to achieve continuous adaptive updating of the safety boundary model.

9. A method for dynamically adjustable microcurrent stimulation and training in ophthalmology according to claim 8, characterized in that: In step S7, the cross-treatment memory data is encrypted and stored in the local archive area of ​​the non-volatile device using the AES-256 algorithm. When the session starts, the cross-treatment memory data is automatically read to reconstruct the initial value of the security upper limit. The parameters of the hierarchical fuzzy inference system are iteratively optimized using dynamic time warping and LMS adaptive algorithms to achieve continuous dynamic updates of the individualized security boundary.

10. A dynamically adjustable microcurrent stimulation and training system for ophthalmology, characterized in that, The system described herein utilizes the method described in any one of claims 1-9 to achieve dynamic adjustment of microcurrent stimulation and training. The system includes a data acquisition module, a preprocessing module, an individualized safety boundary modeling module, a dynamic risk assessment module, a hierarchical fuzzy inference module, a PID control module, a cross-treatment memory storage module, and an adaptive update module.

Images