A method and system for regulating the current during nerve electrical stimulation.

Abstract

This invention relates to the field of neuroelectric stimulation technology, specifically a method and system for current regulation during neuroelectric stimulation. The method identifies the target nerve location and depth based on tissue impedance distribution data and patient characteristic parameters, using a deep neural network model to obtain a recommended initial stimulation current value and a personalized target amplitude. It acquires the deviation between the evoked potential amplitude and the personalized target amplitude, and analyzes this deviation with waveform characteristic parameters to obtain a first current regulation step value to adjust the coarse stimulation current until the evoked potential amplitude exceeds the personalized target amplitude. At this point, the current coarse stimulation current value is recorded as the optimized stimulation current value, and a comprehensive quality score is obtained. Based on the comprehensive quality score, a second current regulation step value is dynamically calculated to adjust the fine-tuned stimulation current value. When the comprehensive quality score exceeds a preset quality score threshold, the final stimulation current value is obtained. This effectively improves the accuracy of current regulation during neuroelectric stimulation for different patients.

Description

Technical Field

[0001] This invention relates to the field of neuroelectric stimulation technology, specifically a method and system for regulating the current during neuroelectric stimulation. Background Technology

[0002] Electrical nerve stimulation plays a crucial role in various fields, including neurological rehabilitation, pain management, and the diagnosis of neurological diseases. By applying precisely controlled electrical stimulation to specific nerve regions, it modulates the electrical activity of nerves, thereby achieving therapeutic or diagnostic purposes. In practical applications, the effectiveness of electrical nerve stimulation is highly dependent on the precise setting of the stimulation current parameters, which directly affects the depth, range, and activation effect on nerve tissue.

[0003] However, due to the complexity of human physiology and individual differences, patients exhibit significant variations in age, sex, body mass index (BMI), and subcutaneous fat thickness. These factors collectively influence the response characteristics of neural tissue to electrical stimulation. For example, younger and older patients may differ in nerve conduction velocity and excitability; men and women differ in fat distribution and nerve sensitivity; and patients with higher BMIs or thicker subcutaneous fat may have nerves buried deeper within the tissue, leading to alterations in the conduction pathways and effects of electrical stimulation. Currently, although some techniques attempt to optimize the neural electrical stimulation process through simple parameter presets or adjustments based on limited physiological indicators, these methods are often too crude and fail to fully consider the complex differences between individual patients, making precise control of the stimulation current difficult. This results in numerous cases where inadequate treatment outcomes or inaccurate diagnostic results are due to improper adjustment of stimulation current parameters.

[0004] Therefore, the present invention aims to propose a current regulation method for the nerve electrical stimulation process, which can effectively improve the accuracy of current regulation in nerve electrical stimulation. Summary of the Invention

[0005] (1) Technical problems to be solved The purpose of this invention is to provide a method and system for regulating the current during nerve electrical stimulation, in order to solve the problem that due to the large differences in the physical condition and physiological characteristics of different patients, it is difficult to achieve precise regulation of the stimulation current according to individual differences, resulting in poor stimulation effect.

[0006] (2) Technical solution To achieve the above objectives, in one aspect, the present invention provides a method for current regulation during nerve electrical stimulation, the method comprising: S1. Obtain patient characteristic parameters, including age, gender, body mass index, and subcutaneous fat thickness; apply microcurrent at the electrode placement location to perform tissue impedance scanning to obtain tissue impedance distribution data along the area surrounding the electrode; identify the target nerve location and target nerve depth based on the tissue impedance distribution data.

[0007] S2. The patient's characteristic parameters, target nerve location, and target nerve depth are used to obtain recommended initial stimulation current values ​​and personalized target amplitudes through a deep neural network model; the deep neural network model is trained using historical patient data.

[0008] S3. After applying electrical stimulation to the target nerve location according to the recommended initial stimulation current value, acquire the evoked potential waveform and extract the waveform feature parameters, including the evoked potential amplitude, latency, waveform morphology coefficient and signal-to-noise ratio.

[0009] S4. Obtain the deviation between the evoked potential amplitude and the personalized target amplitude, and analyze it with the waveform feature parameters to obtain the first current adjustment step value; adjust the recommended initial stimulation current value according to the first current adjustment step value to obtain the coarse-tuned stimulation current until the evoked potential amplitude exceeds the personalized target amplitude, then record the current coarse-tuned stimulation current value as the optimized stimulation current value, and extract the time domain features, frequency domain features and morphological features of the evoked potential waveform and obtain the comprehensive quality score through a convolutional neural network.

[0010] S5. Calculate the second current adjustment step value dynamically based on the comprehensive quality score, adjust the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value, until the comprehensive quality score of the fine-tuned stimulation current value is greater than the preset quality score threshold, then take the current fine-tuned stimulation current value as the final stimulation current value.

[0011] Furthermore, the method for identifying the location and depth of the target nerve based on the tissue impedance distribution data includes: Based on the tissue impedance distribution data, impedance abrupt change regions at the tissue interface are identified; connectivity analysis and morphological feature extraction are performed on the impedance abrupt change regions to obtain multiple candidate neural regions, the morphological features including the spatial extension direction of the impedance abrupt change region, the region aspect ratio, and impedance contrast.

[0012] The morphological features of the candidate neural regions are matched with the pre-stored standard impedance feature templates of neural tissue to obtain the matching confidence of each candidate neural region. Based on the matching confidence, the neural region with the highest matching confidence is selected as the target neural location. The local impedance distribution curve at the target neural location is obtained and the impedance attenuation characteristics are extracted. The impedance attenuation characteristics are mapped to the conduction path length from the electrode surface to the target neural location. The target neural depth is obtained by solving the conduction path length through inversion calculation.

[0013] Furthermore, the method for identifying impedance abrupt change regions at tissue interfaces based on the tissue impedance distribution data includes: The tissue impedance distribution data is filtered to eliminate measurement noise and artifact interference to obtain a smooth impedance distribution field; the impedance gradient field is constructed by calculating the impedance gradient values ​​in each spatial direction in the region around the electrode; and the impedance abrupt change region of the tissue interface is identified by the gradient magnitude and gradient direction distribution characteristics in the impedance gradient field.

[0014] Furthermore, the method for obtaining the deviation between the induced potential amplitude and the personalized target amplitude, and analyzing the waveform characteristic parameters to obtain the first current adjustment step value, includes: The difference between the current evoked potential amplitude and the personalized target amplitude is calculated and normalized to obtain the amplitude deviation coefficient; the latency, waveform morphology coefficient and signal-to-noise ratio in the waveform characteristic parameters are respectively calculated to obtain the latency deviation, morphology deviation and signal-to-noise ratio deviation from their corresponding normal reference ranges.

[0015] The latency deviation, morphological deviation, and signal-to-noise ratio deviation are fused by nonlinear weighting to obtain a comprehensive waveform anomaly index; the amplitude deviation coefficient is calculated by an adaptive step mapping function to obtain a first current adjustment step value, and the mapping slope of the adaptive step mapping function is adjusted according to the comprehensive waveform anomaly index.

[0016] Further, the method of calculating the first current adjustment step value by means of the amplitude deviation coefficient through an adaptive step mapping function, and adjusting the mapping slope of the adaptive step mapping function according to the comprehensive waveform anomaly index includes: The adaptive step mapping function adopts a piecewise linear mapping relationship, dividing the range of amplitude deviation coefficient into a fast adjustment segment and a fine adjustment segment. The mapping slope in each adjustment segment represents the increment of the current adjustment step value corresponding to a unit amplitude deviation coefficient.

[0017] Obtain the initial mapping slope baseline value for the current patient, and assign the initial mapping slope baseline value to the fast adjustment segment and the fine adjustment segment respectively as the base mapping slope; after normalizing the comprehensive waveform abnormality index, obtain the slope attenuation coefficient through inverse proportional function transformation.

[0018] The base mapping slope is adjusted according to the slope attenuation coefficient to obtain the adjusted mapping slope; the corresponding adjusted mapping slope is selected according to the adjustment segment to which the current amplitude deviation coefficient belongs, and the amplitude deviation coefficient and the adjusted mapping slope are linearly calculated to obtain the first current adjustment step value.

[0019] Furthermore, the method for extracting the temporal, frequency, and morphological features of the evoked potential waveform and obtaining a comprehensive quality score through a convolutional neural network includes: Temporal features are extracted from the evoked potential waveform through temporal domain analysis, including the peak occurrence time, time interval between adjacent peaks, rising slope, and falling slope. A fast Fourier transform is performed on the evoked potential waveform to obtain frequency domain spectra, from which frequency domain features are extracted, including the dominant frequency component, the proportion of spectral energy within a preset physiological frequency band, and the concentration of spectral energy distribution. Morphological features are extracted from the evoked potential waveform through morphological analysis, including waveform left-right symmetry indices, waveform envelope smoothness indices, and baseline drift amplitude.

[0020] After normalizing the time-domain features, frequency-domain features, and morphological features, they are organized into three feature vectors according to feature type. The three feature vectors are then mapped to three independent feature channels to construct a multi-channel feature matrix. The fusion feature pattern across feature channels is extracted through the convolutional and pooling layers of a convolutional neural network. Finally, the fusion feature pattern is mapped to the scoring space through a fully connected layer to obtain a comprehensive quality score.

[0021] Furthermore, the method of dynamically calculating the second current adjustment step value based on the comprehensive quality score, and adjusting the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value includes: The difference between the overall quality score and the preset quality score threshold is calculated to obtain the quality improvement space. The quality improvement space is then transformed by a logarithmic function to obtain the nonlinear quality gain coefficient.

[0022] Based on the optimized stimulation current value, positive micro-perturbation current and negative micro-perturbation current are applied respectively. The positive micro-perturbation current is the optimized stimulation current value plus a preset micro-perturbation amplitude, and the negative micro-perturbation current is the optimized stimulation current value minus the preset micro-perturbation amplitude.

[0023] The evoked potential waveforms of the positive and negative micro-perturbation currents are obtained, and the quality improvement direction and effective quality gradient are determined. The second current adjustment step value is calculated based on the effective quality gradient and the nonlinear quality gain coefficient. The second current adjustment step value is applied to the optimized stimulation current value according to the determined quality improvement direction to obtain the fine-tuned stimulation current value.

[0024] Furthermore, the method for obtaining the induced potential waveforms of the positive and negative micro-perturbation currents and determining the quality improvement direction and effective quality gradient includes: The evoked potential waveforms corresponding to positive and negative micro-perturbation currents were collected separately, and their time-domain, frequency-domain, and morphological features were extracted. A convolutional neural network was used to obtain the quality scores for the positive and negative micro-perturbation currents, respectively. The difference between the positive micro-perturbation quality score and the overall quality score was calculated as the positive quality gradient, and the difference between the negative micro-perturbation quality score and the overall quality score was calculated as the negative quality gradient. When the positive quality gradient was positive, the direction of current increase was determined as the quality improvement direction; when the negative quality gradient was positive, the direction of current decrease was determined as the quality improvement direction. The quality gradient corresponding to the quality improvement direction was selected as the effective quality gradient.

[0025] On the other hand, based on the same inventive concept, the present invention also provides a current regulation system for a nerve electrical stimulation process, the system comprising: a tissue impedance distribution data acquisition and analysis module, an initial stimulation parameter generation module, a waveform feature parameter extraction module, a coarse-tuning stimulation current analysis module, and a fine-tuning stimulation current analysis module, wherein each module is sequentially connected in communication. The tissue impedance distribution data acquisition and analysis module is used to acquire patient characteristic parameters, including age, gender, body mass index, and subcutaneous fat thickness; apply microcurrent at the electrode placement location to perform tissue impedance scanning to obtain tissue impedance distribution data along the area surrounding the electrode; and identify the location and depth of the target nerve based on the tissue impedance distribution data.

[0026] The initial stimulation parameter generation module is used to obtain recommended initial stimulation current values ​​and personalized target amplitudes from the patient characteristic parameters, target nerve location, and target nerve depth through a deep neural network model; the deep neural network model is trained using historical patient data.

[0027] The waveform feature parameter extraction module is used to acquire evoked potential waveforms and extract waveform feature parameters after applying electrical stimulation to the target nerve location according to the recommended initial stimulation current value. The waveform feature parameters include evoked potential amplitude, latency, waveform morphology coefficient, and signal-to-noise ratio.

[0028] The coarse-adjustment stimulation current analysis module is used to obtain the deviation between the evoked potential amplitude and the personalized target amplitude, and to obtain a first current adjustment step value by analyzing the waveform feature parameters; the recommended initial stimulation current value is adjusted according to the first current adjustment step value to obtain the coarse-adjustment stimulation current until the evoked potential amplitude exceeds the personalized target amplitude, then the current coarse-adjustment stimulation current value is recorded as the optimized stimulation current value, and the time domain features, frequency domain features and morphological features of the evoked potential waveform are extracted and a comprehensive quality score is obtained through a convolutional neural network.

[0029] The fine-tuning stimulation current analysis module is used to dynamically calculate the second current adjustment step value based on the comprehensive quality score, and adjust the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value until the comprehensive quality score of the fine-tuned stimulation current value is greater than the preset quality score threshold. Then, the current fine-tuned stimulation current value is taken as the final stimulation current value.

[0030] (3) Beneficial effects Compared with the prior art, the beneficial effects of the present invention are: 1. By comprehensively considering the patient's multi-dimensional characteristic parameters and combining tissue impedance distribution data, the target nerve location and depth are accurately identified. A deep neural network model then provides recommended initial stimulation current values ​​and personalized target amplitudes. This adjustment method, based on comprehensive individual characteristic information, allows for the customization of the most suitable initial stimulation parameters for each patient. This effectively avoids the inaccuracy caused by traditional methods neglecting individual differences, significantly improving the accuracy and effectiveness of initial stimulation current adjustment.

[0031] 2. Based on the initial stimulation, characteristic parameters of the evoked potential waveform are collected. Combined with deviation analysis from the personalized target amplitude, a first current adjustment step value is dynamically calculated for coarse adjustment, gradually bringing the stimulation current closer to the ideal range. Further, multi-dimensional features of the evoked potential waveform are extracted and a comprehensive quality score is obtained through a convolutional neural network. A second current adjustment step value is then dynamically calculated based on the comprehensive quality score for fine adjustment. This hierarchical dynamic adjustment mechanism can respond in real time to changes in the patient's neural tissue response to stimulation, adjusting the stimulation current promptly. This effectively avoids situations where excessive or insufficient fluctuations in the stimulation current affect the treatment or diagnostic effect, enhancing the stability and reliability of the electrical stimulation process. Attached Figure Description

[0032] Figure 1 This is a flowchart of a current regulation method for a nerve electrical stimulation process according to Embodiment 1 of the present invention.

[0033] Figure 2 This is a schematic diagram of the module composition of a current regulation system for a nerve electrical stimulation process according to Embodiment 2 of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Example 1: As Figure 1 As shown, this embodiment provides a method for current regulation during nerve electrical stimulation, the method comprising: S1. Obtain patient characteristic parameters, including age, gender, body mass index (BMI), and subcutaneous fat thickness. Apply a microcurrent to the electrode placement location to perform tissue impedance scanning and obtain tissue impedance distribution data along the area surrounding the electrode. Identify the target nerve location and depth based on the tissue impedance distribution data. Obtaining patient characteristic parameters is routine and well-established in clinical practice. Age and gender are obtained directly from the patient's medical record. BMI is calculated using a standard formula after measuring the patient's height and weight. Subcutaneous fat thickness is measured using an ultrasound measuring instrument in the predetermined electrode placement area. These four parameters directly affect the effect of nerve electrical stimulation. Age affects nerve conduction velocity and excitability; elderly patients typically have slower nerve conduction velocity, requiring different stimulation current parameters compared to younger patients. Gender differences are reflected in fat distribution patterns and nerve sensitivity; women usually have more even and relatively thicker subcutaneous fat distribution. BMI reflects the patient's overall body shape; patients with higher BMIs have different electrical impedance characteristics in the tissues surrounding the nerves compared to patients with normal body shape. The thickness of subcutaneous fat directly determines the degree of attenuation of current as it travels from the skin surface to deep nerves; the thicker the fat layer, the more significant the current attenuation. After acquiring the patient's characteristic parameters, the operator places electrodes on the corresponding parts of the patient's body according to standard nerve electrical stimulation procedures. After electrode placement, a microcurrent with an extremely small amplitude is applied to the electrodes to perform tissue impedance scanning. The amplitude of this microcurrent is far below the threshold current that can cause nerve excitation, usually at the microampere level, and will not produce any stimulation or physiological response in the patient; its sole purpose is to measure the tissue impedance at different locations. During the scanning process, the microcurrent is controlled to sequentially sweep across different areas around the electrodes along a preset spatial path. At each location, the voltage response is measured, and the tissue impedance value at that location is calculated according to Ohm's law. The complete scan covers a certain range of three-dimensional space around the electrodes, ultimately obtaining a set of spatially continuously distributed impedance values, forming tissue impedance distribution data. This tissue impedance distribution data actually reflects the distribution of different tissue types around the electrodes, because the conductivity of different tissues varies significantly; the impedance values ​​of nerve tissue, muscle tissue, adipose tissue, and bone are all different.

[0036] S2. The patient's characteristic parameters, target nerve location, and target nerve depth are used to obtain a recommended initial stimulation current value and a personalized target amplitude through a deep neural network model. This deep neural network model is trained using historical patient data. It is a multi-layer feedforward neural network, and its training data comes from a large number of real treatment records of historical patients. Each training sample contains the patient's characteristic parameters, nerve location and depth information, as well as a clinically validated initial stimulation current value and the evoked potential amplitude at which the ideal treatment effect is achieved. The deep neural network model learns the patterns implicit in this historical data to establish a mapping relationship from individual patient characteristics to optimal current stimulation parameters. The deep neural network model outputs two key values: the recommended initial stimulation current value and the personalized target amplitude. The recommended initial stimulation current value is a suitable initial stimulation current intensity predicted based on the current patient's specific characteristics, which is more in line with the patient's actual situation than traditional fixed initial values ​​or values ​​set solely based on experience. The personalized target amplitude refers to the amplitude of the evoked potential waveform that should appear when the stimulation effect reaches an ideal state for that patient. This target amplitude varies from person to person, depending on the patient's nerve sensitivity and tissue conduction characteristics.

[0037] S3. After applying electrical stimulation to the target nerve location according to the recommended initial stimulation current value, evoked potential waveforms are acquired and waveform feature parameters are extracted. These waveform feature parameters include evoked potential amplitude, latency, waveform morphology coefficient, and signal-to-noise ratio. Electrical stimulation is typically performed in pulse form, with pulse width and frequency set according to specific treatment or diagnostic needs. After electrical stimulation is applied, nerve tissue is activated and generates nerve impulses. These impulses propagate along nerve fibers, causing corresponding physiological responses at the distal end of the conduction path and simultaneously producing detectable potential changes locally—this is the evoked potential. The system acquires evoked potential signals using recording electrodes placed on the nerve conduction path or the surface of the nerve-innervated muscle to obtain the evoked potential waveform. The evoked potential waveform is a voltage curve that changes over time, containing rich information about the nerve response. Four key waveform feature parameters are extracted from the acquired evoked potential waveform. The evoked potential amplitude refers to the voltage amplitude of the main peak in the waveform, directly reflecting the degree of nerve activation; a larger amplitude indicates a greater number of activated nerve fibers or a deeper degree of activation. The latency period refers to the time delay between the application of a stimulus and the appearance of the main wave of the evoked potential waveform. This delay is mainly determined by the conduction time of the nerve impulse and reflects the state of nerve conduction function. The waveform morphology coefficient is a numerical value describing the overall shape characteristics of the waveform. It is quantified by calculating parameters such as kurtosis, skewness, or waveform width. Normal evoked potential waveforms have specific morphological characteristics; abnormal morphology often indicates inappropriate stimulus parameters or abnormal nerve function. The signal-to-noise ratio (SNR) is the ratio of the evoked potential signal intensity to the background noise intensity. A high SNR indicates a clear and reliable acquired signal, while a low SNR may indicate insufficient stimulus current resulting in a weak evoked potential signal or significant electromagnetic interference.

[0038] S4. Obtain the deviation between the evoked potential amplitude and the personalized target amplitude, and analyze it with the waveform feature parameters to obtain a first current adjustment step value. Adjust the recommended initial stimulation current value according to the first current adjustment step value to obtain a coarse-tuned stimulation current until the evoked potential amplitude exceeds the personalized target amplitude. Then, record the current coarse-tuned stimulation current value as the optimized stimulation current value, and extract the time-domain features, frequency-domain features, and morphological features of the evoked potential waveform, and obtain a comprehensive quality score through a convolutional neural network. Adjust the recommended initial stimulation current value according to the first current adjustment step value to obtain a new stimulation current value, called the coarse-tuned stimulation current value. Apply stimulation again using the coarse-tuned stimulation current value, collect a new evoked potential waveform, extract the evoked potential amplitude, and determine whether it exceeds the personalized target amplitude. If it has not yet exceeded it, continue to calculate a new first current adjustment step value and perform the next round of adjustment. This iterative process continues, and each iteration makes the stimulation current closer to the ideal value. When the amplitude of the evoked potential measured after a certain adjustment first exceeds the personalized target amplitude, it indicates that the electrical stimulation intensity has entered the target range, and the coarse-tuned stimulation current value at this time is recorded as the optimized stimulation current value. The coarse-tuning phase ends here, and the system has adjusted the stimulation current from its initial value to a level close to the optimal value. Based on the completion of coarse-tuning, the system enters a more refined fine-tuning phase. At this point, it is no longer sufficient to simply focus on whether the evoked potential amplitude meets the target; it is also necessary to evaluate the overall quality of the evoked potential waveform. The system performs more comprehensive feature extraction on the evoked potential waveform acquired under the current optimized stimulation current, including time-domain features, frequency-domain features, and morphological features. These three types of features reflect the quality of the evoked potential from different perspectives. These multi-dimensional features are input into a convolutional neural network, which outputs a comprehensive quality score between 0 and 100. The higher the score, the better the waveform quality, indicating that the neural response under the current stimulation parameters is closer to the ideal state.

[0039] S5. Calculate the second current adjustment step value dynamically based on the comprehensive quality score. Adjust the optimized stimulation current value according to the second current adjustment step value to obtain a fine-tuned stimulation current value. Continue until the comprehensive quality score of the fine-tuned stimulation current value exceeds a preset quality score threshold. Then, use the current fine-tuned stimulation current value as the final stimulation current value. The calculation method of the second current adjustment step value differs from that of the first current adjustment step value; it is more refined and based on the changing trend of the comprehensive quality score. Apply stimulation to the fine-tuned stimulation current value obtained by fine-tuning the optimized stimulation current value according to the calculated second current adjustment step value, collect the evoked potential waveform, extract features again, and calculate the comprehensive quality score. If the new comprehensive quality score exceeds the preset quality score threshold, it indicates that the evoked potential waveform has reached a high quality standard. The fine-tuned stimulation current value at this time is the final stimulation current value, and the entire adjustment process ends. If the comprehensive quality score has not yet met the standard, extract the multi-dimensional features of the evoked potential waveform from the current fine-tuned stimulation current value and calculate the comprehensive quality score. Then, continue to calculate a new second current adjustment step value and perform the next round of fine-tuning until the comprehensive quality score meets the requirements.

[0040] The method for identifying the location and depth of a target nerve based on the tissue impedance distribution data includes: Based on the tissue impedance distribution data, impedance abrupt change regions at the tissue interface are identified. Connectivity analysis and morphological feature extraction are performed on these impedance abrupt change regions to obtain multiple candidate neural regions. The morphological features include the spatial extension direction, aspect ratio, and impedance contrast of the impedance abrupt change regions. The purpose of connectivity analysis is to determine which impedance abrupt change points are spatially connected and belong to the same continuous tissue structure, thereby aggregating scattered impedance abrupt change points into complete candidate neural regions. Morphological feature extraction quantifies the geometric shape and spatial distribution characteristics of each candidate neural region. The extracted morphological features include three key parameters: Spatial extension direction describes the main orientation of the candidate neural region in three-dimensional space. Nerves typically present as elongated structures with a clear extension direction, while other non-neural tissues may exhibit irregular shapes or clustered distributions. The aspect ratio is the ratio of the length of the candidate neural region in its main extension direction to its width perpendicular to that direction. The aspect ratio of neural tissue is usually large, often exceeding 5:1, while the aspect ratio of fat or muscle tissue is smaller. Impedance contrast reflects the degree of difference in impedance values ​​between candidate neural regions and surrounding tissues. Due to their unique electrophysiological characteristics, neural tissues typically exhibit significant impedance differences compared to surrounding tissues. By extracting these morphological features, the system can quantify the characteristics of multiple candidate neural regions.

[0041] The morphological features of the candidate nerve regions are matched with pre-stored standard impedance feature templates for nerve tissue to calculate the matching confidence of each candidate nerve region. Based on the matching confidence, the nerve region with the highest matching confidence is selected as the target nerve location. The local impedance distribution curve at the target nerve location is obtained, and the impedance attenuation characteristics are extracted. A mapping relationship is established between the impedance attenuation characteristics and the conduction path length from the electrode surface to the target nerve location. The target nerve depth is obtained by inversion calculation to solve for the conduction path length. The system pre-stores standard impedance feature templates for nerve tissue. These templates are established based on a large amount of anatomical and clinical measurement data, describing the typical characteristics that nerves in different locations should present under impedance scanning. The morphological features of each candidate nerve region are matched with these standard impedance feature templates. The matching process uses a multi-feature weighted comparison method, calculating the angular deviation in the spatial extension direction, the numerical difference in aspect ratio, and the relative error of impedance contrast. Then, different weights are assigned according to the importance of each feature for comprehensive analysis, resulting in the matching confidence of each candidate nerve region with the standard impedance feature template for nerve tissue. A higher match confidence score indicates that the candidate nerve region is more likely to be actual nerve tissue. The nerve region with the highest match confidence score is selected from all candidate nerve regions and identified as the target nerve location. For example, when searching for the ulnar nerve, the system may identify three candidate nerve regions. One is located on the medial side of the elbow, extending longitudinally with an aspect ratio of 8:1 and an impedance contrast of 0.42, achieving a match confidence score of 0.91 with the ulnar nerve template. The other two candidate regions have irregular shapes with aspect ratios of 2.3:1 and 1.8:1, respectively, and match confidence scores of only 0.35 and 0.28. Based on this, the system determines the first candidate region as the target ulnar nerve location. The local impedance distribution curve at the target nerve location is obtained; this is an impedance change curve extending from the electrode surface along the depth direction to the nerve location. The impedance attenuation characteristics, i.e., the law of impedance value change with increasing depth, are extracted from the local impedance distribution curve. Human tissues typically exhibit a layered structure along the vertical depth direction, consisting of layers from superficial to deep, such as the skin, subcutaneous fat, fascia, and muscle. Each layer possesses a specific electrical conductivity. A pre-established spatial layering model of tissue conductivity describes the spatial distribution of conductivity under typical tissue layering conditions. According to this model, a definite mathematical relationship exists between the conduction path length of current from the electrode surface to deep nerves and the impedance characteristics of each tissue layer along the way. The system establishes a mapping relationship between the actually measured impedance attenuation characteristics and the spatial layering model of tissue conductivity, and then calculates the conduction path length through inversion.Inversion calculation is a numerical solution process that infers model parameters from measurement results. That is, a series of possible nerve depth values ​​are set, and for each nerve depth value, the theoretical impedance attenuation curve that should be generated is calculated according to the tissue conductivity spatial stratification model. These theoretical curves are compared with the measured curves to find the nerve depth value that minimizes the error between the two. This nerve depth value is the target nerve depth.

[0042] The method for identifying impedance abrupt change regions at tissue interfaces based on the tissue impedance distribution data includes: The tissue impedance distribution data is filtered to eliminate measurement noise and artifact interference, resulting in a smooth impedance distribution field. An impedance gradient field is constructed by calculating the impedance gradient values ​​in each spatial direction within the electrode region. The impedance abrupt change regions at the tissue interface are identified by the gradient amplitude and gradient direction distribution characteristics in the impedance gradient field. Tissue impedance distribution data inevitably contains various interference factors during measurement, including electromagnetic environmental noise, measurement fluctuations caused by unstable electrode contact, and artifact interference from patient breathing or minor movements. These interferences superimpose high-frequency noise components and irregular outliers into the tissue impedance distribution data. Direct analysis without processing would lead to the misidentification of numerous spurious abrupt changes, affecting the accuracy of nerve localization. The system filters the raw tissue impedance distribution data using digital signal processing techniques. Appropriate filters are designed to separate noise and artifact components from the useful signal. Since the impedance distribution of tissue structures is spatially continuous and relatively smooth, while noise and artifacts typically manifest as high-frequency random fluctuations or isolated spikes, low-pass filters can effectively suppress these interference components. The cutoff frequency of the filter is set according to the spatial scale of the actual tissue structure, aiming to preserve the true tissue boundary information while removing noise. After filtering, a smooth impedance distribution field is obtained, in which data points exhibit a continuous and smooth transition. Local random fluctuations are eliminated, but the impedance change characteristics at the actual tissue interface are preserved. Impedance gradient values ​​are calculated based on the smooth impedance distribution field. The impedance gradient describes the rate of change of impedance in space, i.e., the change in impedance per unit distance. At each spatial location within the electrode region, the rate of change of impedance along the three orthogonal directions of the X, Y, and Z axes is calculated, yielding the impedance gradient vector at that location. The magnitude of the gradient vector is called the gradient amplitude, indicating the degree of impedance change at that location; the direction of the gradient vector is called the gradient direction, pointing in the direction of the fastest increase in impedance. The collection of gradient vectors from all spatial locations constitutes the impedance gradient field. In the impedance gradient field, the location of the tissue interface exhibits distinct characteristics: the gradient amplitude reaches its peak at the tissue interface because of the abrupt change in impedance values ​​between different tissue types on both sides of the interface; the gradient direction is perpendicular to the interface at the interface, pointing towards the side with higher impedance tissue. Impedance abrupt change regions at tissue interfaces are identified by analyzing the distribution characteristics of gradient magnitude and gradient direction in the impedance gradient field. Specifically, a gradient magnitude threshold is set, and spatial locations where the gradient magnitude exceeds the threshold are marked as potential abrupt change points. Then, the gradient directions of these abrupt change points are checked for consistency. If a group of adjacent abrupt change points have similar gradient directions, it indicates that they belong to the same tissue interface, thus identifying an impedance abrupt change region.

[0043] The method for obtaining the deviation between the induced potential amplitude and the personalized target amplitude, and analyzing the waveform characteristic parameters to obtain the first current adjustment step value, includes: The difference between the current evoked potential amplitude and the personalized target amplitude is calculated and normalized to obtain the amplitude deviation coefficient. The latency, waveform morphology coefficient, and signal-to-noise ratio (SNR) of the waveform characteristic parameters are calculated to deviate from their corresponding normal reference ranges to obtain latency deviation, morphology deviation, and SNR deviation, respectively. The absolute deviation value is obtained by subtracting the currently measured evoked potential amplitude from the personalized target amplitude. Since the personalized target amplitude may vary significantly among different patients, the absolute deviation value is normalized by dividing it by the personalized target amplitude to obtain the amplitude deviation coefficient, ensuring comparability between different patients. The amplitude deviation coefficient is a dimensionless relative value. A positive value indicates that the current evoked potential amplitude is lower than the personalized target amplitude, requiring an increase in the stimulation current; a negative value indicates that the current evoked potential amplitude exceeds the personalized target amplitude, requiring a decrease in the stimulation current. The absolute value reflects the degree of deviation from the personalized target amplitude. For example, if a patient's personalized target amplitude is 8 mV and the currently measured evoked potential amplitude is 5.6 mV, the absolute deviation is 2.4 mV, and the normalized amplitude deviation coefficient is 0.3, indicating that the current evoked potential amplitude is 30% lower than the personalized target amplitude. Besides focusing on whether the amplitude meets the target, it is also necessary to assess the overall quality of the evoked potential waveform. Even if the amplitude is close to the target value, if the waveform exhibits abnormal morphology, prolonged latency, or a decreased signal-to-noise ratio, it indicates an unsatisfactory stimulus state, which may be due to stimulus location deviation, inappropriate stimulus frequency, or changes in the patient's condition. The system assesses the degree of abnormality in the waveform characteristic parameters of latency, waveform morphology coefficient, and signal-to-noise ratio. The system internally stores the normal reference ranges for each of these three parameters, which are statistically derived from a large amount of normal evoked potential data. For latency, the normal range is usually determined based on the distance between the stimulus site and the recording site and the known nerve conduction velocity; for waveform morphology coefficient, the normal range reflects the shape characteristics of a typical evoked potential waveform; for signal-to-noise ratio, the normal range ensures sufficient signal strength and controllable noise interference. The system compares each currently measured parameter value with its normal range to calculate the degree of deviation. The calculation method for latency deviation is as follows: if the current latency is within the normal range, the deviation is zero; if it exceeds the range, the deviation is obtained by dividing the excess by the upper (or lower) limit of the normal range. Morphological deviation and signal-to-noise ratio deviation are calculated using similar logic.

[0044] The latency deviation, morphological deviation, and signal-to-noise ratio deviation are fused using a nonlinear weighted fusion method to obtain a comprehensive waveform anomaly index. The amplitude deviation coefficient is calculated using an adaptive step mapping function to obtain a first current adjustment step value, and the mapping slope of the adaptive step mapping function is adjusted according to the comprehensive waveform anomaly index. Nonlinear weighting is used because the impact of abnormalities in different deviation parameters on the overall waveform quality is not proportional, and there may be a synergistic effect when multiple deviation parameters are abnormal simultaneously. The fusion process first assigns weight coefficients to the three deviations. These weight coefficients are determined based on clinical experience and statistical analysis. Usually, latency abnormalities have a higher weight because they directly reflect the neurological functional state. Then, a nonlinear function is used to synthesize the weighted deviations. A common approach is to first perform a weighted summation to obtain a preliminary index, and then perform a nonlinear transformation using a sigmoid function or an exponential function, so that the index increases slowly when the abnormality is mild and increases rapidly when the abnormality is severe. The final comprehensive waveform anomaly index is a value between 0 and 1, where 0 indicates a completely normal waveform and 1 indicates a severely abnormal waveform.

[0045] The method of calculating the first current adjustment step value by means of the amplitude deviation coefficient through an adaptive step mapping function, and adjusting the mapping slope of the adaptive step mapping function according to the comprehensive waveform anomaly index includes: The adaptive step mapping function employs a piecewise linear mapping relationship, dividing the amplitude deviation coefficient range into a rapid adjustment segment and a fine adjustment segment. The mapping slope within each segment represents the increment of the current adjustment step value corresponding to a unit amplitude deviation coefficient. The piecewise linear mapping relationship is based on the consideration of different adjustment strategies required at different stages of the stimulation current adjustment process. When the amplitude deviation coefficient is large, it indicates that the current stimulation current is far from the ideal value, requiring rapid adjustment to quickly approach the target; this region is called the rapid adjustment segment. When the amplitude deviation coefficient is small, it indicates that the stimulation current is already close to the ideal value, requiring fine adjustment to avoid overshoot; this region is called the fine adjustment segment. The amplitude deviation coefficient is divided into these two segments based on its absolute value. The threshold for division is usually set between 0.1 and 0.15, meaning that an absolute value of the deviation coefficient greater than 0.15 belongs to the rapid adjustment segment, and less than 0.15 belongs to the fine adjustment segment. A linear mapping relationship is used within each segment, meaning the current step value is proportional to the amplitude deviation coefficient, and the proportionality coefficient is the mapping slope of that segment. The fast adjustment section has a larger mapping slope, which allows a larger deviation to generate a larger current adjustment step value; the fine adjustment section has a smaller mapping slope, which keeps the current adjustment step value within a small range even if the deviation is not zero, thus avoiding repeated oscillations.

[0046] An initial mapping slope baseline value is obtained for the current patient, and this baseline value is assigned to both the rapid adjustment segment and the fine adjustment segment as the base mapping slope. The slope attenuation coefficient is obtained by normalizing the comprehensive waveform abnormality index and then transforming it using an inverse proportional function. The initial mapping slope baseline value for the current patient is not a fixed value applicable to all patients, but is calculated based on the patient's individual characteristics. Due to differences in nerve sensitivity and tissue conduction characteristics, the same change in current may produce different degrees of evoked potential amplitude changes in different patients, thus requiring different mapping slopes. Based on the patient's age, gender, body mass index, and other characteristic parameters, combined with the adjustment experience of similar patients in historical data, a suitable initial mapping slope baseline value is estimated for the patient. Typically, the base mapping slope of the rapid adjustment segment is 3-5 times that of the fine adjustment segment. For example, for a 60-year-old female patient, the estimated base mapping slope for the rapid adjustment segment is 4.5 mA / unit deviation coefficient, and for the fine adjustment segment, it is 1.2 mA / unit deviation coefficient. The composite waveform anomaly index is normalized to a preset adjustment range. The composite waveform anomaly index is originally a value between 0 and 1, and the preset adjustment range is usually set to 0.1-1.0. This is to ensure that even when the waveform is completely normal (anomaly index of 0), the slope adjustment is not too aggressive, maintaining a certain degree of conservatism. Normalization maps the anomaly index to the 0.1-1.0 range through a linear transformation. The inverse proportional function takes the form: slope attenuation coefficient = constant / normalized anomaly index. The characteristic of the inverse proportional function is that when the anomaly index is small (good waveform quality), the attenuation coefficient is large, with little impact on the slope; when the anomaly index is large (poor waveform quality), the attenuation coefficient is small, significantly reducing the slope. The selection of the constant ensures a reasonable range for the attenuation coefficient, typically set to vary between 0.3 and 1.0. For example, when the normalized composite waveform anomaly index is 0.2, the slope attenuation coefficient is 1.0, indicating no attenuation; when the composite waveform anomaly index is 0.8, the slope attenuation coefficient may be 0.4, indicating that the slope attenuation coefficient needs to be reduced to 40% of the original value.

[0047] The adjusted mapping slope is obtained by adjusting the base mapping slope according to the slope attenuation coefficient. The corresponding adjusted mapping slope is selected based on the adjustment segment to which the current amplitude deviation coefficient belongs. The amplitude deviation coefficient and the adjusted mapping slope are linearly calculated to obtain the first current adjustment step value. The absolute value of the current amplitude deviation coefficient is determined to belong to which adjustment segment? If the absolute value of the deviation coefficient is greater than the segment division threshold, the adjusted mapping slope of the fast adjustment segment is selected; otherwise, the adjusted mapping slope of the fine adjustment segment is selected. The amplitude deviation coefficient is multiplied by the selected adjusted mapping slope to obtain the first current adjustment step value. For example, at a certain moment, the amplitude deviation coefficient is 0.25 (belonging to the fast adjustment segment), the base mapping slope of the fast adjustment segment is 4.5 mA / unit deviation coefficient, the current comprehensive waveform anomaly index, after normalization and inverse proportional transformation, yields a slope attenuation coefficient of 0.7, and the adjusted mapping slope is 3.15 mA / unit. The final calculated first current adjustment step value is 0.25 × 3.15 = 0.79 mA.

[0048] The method for extracting the temporal, frequency, and morphological features of the evoked potential waveform and obtaining a comprehensive quality score through a convolutional neural network includes: Temporal features are extracted from the evoked potential waveform through temporal domain analysis. These features include the peak occurrence time, the time interval between adjacent peaks, the rising slope of the waveform, and the falling slope of the waveform. A fast Fourier transform (FFT) is performed on the evoked potential waveform to obtain a frequency domain spectrum. Frequency domain features are extracted from this spectrum, including the dominant frequency component, the proportion of spectral energy within a preset physiological frequency band, and the concentration of spectral energy distribution. Morphological features are extracted from the evoked potential waveform through morphological analysis, including waveform left-right symmetry indices, waveform envelope smoothness indices, and baseline drift amplitude. The waveform is analyzed from three different perspectives to extract complementary feature information, which is then fused and evaluated using a deep learning model. Temporal domain analysis directly extracts features from the waveform on the time axis, identifying the positions of each peak in the evoked potential waveform and recording the precise occurrence time of each major peak, reflecting the response timing of different nerve fiber components after neural activation. The time interval between adjacent peaks is calculated; this time interval is related to the repolarization time of the nerve and the subsequent propagation process of excitation. The rising slope of a waveform refers to the rate at which the voltage changes over time as the waveform rises from the baseline to its peak value. A larger slope indicates faster neural activation. The falling slope describes the speed at which the waveform falls back from its peak value to the baseline, reflecting the speed of neural recovery. These four time-domain characteristics characterize the dynamic changes of evoked potentials over time from different perspectives. For example, in a typical median nerve evoked potential waveform, the main peak appears 3.5 milliseconds after stimulation, with an interval of 1.2 milliseconds between the previous small peak. The rising slope is 2.8 mV / ms, and the falling slope is -1.9 mV / ms. Frequency domain analysis reveals the frequency components of the waveform through Fourier transform. Performing a Fast Fourier Transform on the evoked potential waveform converts the time-domain signal to the frequency domain, obtaining the frequency spectrum. The horizontal axis of the frequency spectrum is frequency, and the vertical axis is the energy intensity of that frequency component. From the frequency spectrum, we can see which frequency components the waveform contains and the relative strengths of each component. The system extracts three key frequency domain features. The dominant frequency component refers to the frequency with the strongest energy in the frequency domain spectrum, representing the most important oscillatory characteristic of the waveform and related to the inherent frequency characteristics of the nerve. The proportion of spectral energy within the preset physiological frequency band refers to the ratio of the sum of spectral energy within a physiologically meaningful frequency range (usually 10-500 Hz) to the total spectral energy. A high proportion indicates that the main components of the waveform are concentrated in the physiological frequency band, while a low proportion may indicate more high-frequency noise or low-frequency drift. The concentration of spectral energy distribution describes the degree of dispersion of spectral energy along the frequency axis. High concentration indicates that the energy is mainly concentrated on a few frequency components, and the waveform is relatively regular; low concentration indicates that the energy is dispersed across multiple frequencies, and the waveform is more complex or contains interference. For example, the Fourier transform of a certain evoked potential waveform shows that the dominant frequency component is 85 Hz, the energy in the 10-500 Hz frequency band accounts for 92% of the total energy, and the concentration index of spectral energy is 0.68.Morphological analysis focuses on the geometric shape characteristics of the waveform. The waveform's left-right symmetry index is calculated by comparing the similarity of the shapes on the left and right sides of the waveform peak. Normal evoked potential waveforms typically exhibit a certain degree of symmetry; asymmetry may indicate an abnormality. The system calculates parameters such as the area and slope of the left and right halves of the waveform, and obtains the symmetry index through correlation analysis or overlap calculation. A value closer to 1 indicates greater symmetry. The waveform envelope smoothness index reflects the smoothness of the waveform's outer contour. The system first extracts the upper and lower envelopes of the waveform, then calculates the second derivative or curvature change of the envelopes. Envelopes with high smoothness have small and continuous curvature changes, while envelopes with low smoothness exhibit more jagged edges or abrupt changes. Baseline drift amplitude measures the degree of fluctuation in the waveform within the baseline region where it should remain horizontal. Ideally, the baseline should be a stable zero-potential line. In actual measurements, the baseline may drift slowly due to electrode polarization, patient movement, or other factors. The system extracts the voltage values ​​in non-signal regions of the waveform (before stimulation and after waveform completion) and calculates their standard deviation or peak-to-peak value as the baseline drift amplitude. For example, a certain waveform has a symmetry index of 0.82, an envelope smoothness index of 0.91, and a baseline drift amplitude of 0.15 millivolts.

[0049] After normalizing the temporal, frequency, and morphological features, they are organized into three feature vectors according to feature type. These three feature vectors are then mapped to three independent feature channels to construct a multi-channel feature matrix. A fusion feature pattern across feature channels is extracted using convolutional and pooling layers of a convolutional neural network. Finally, a fully connected layer maps the fusion feature pattern to a scoring space to obtain a comprehensive quality score. Since different types of features have different numerical ranges and dimensions, directly using them would lead to features with larger numerical values ​​dominating the analysis process. Normalization maps each feature value to a uniform numerical range (usually 0-1), making different features comparable. After normalization, features of the same type are organized into a single feature vector. For example, temporal features contain four values ​​forming a 4-dimensional vector, frequency features contain three values ​​forming a 3-dimensional vector, and morphological features contain three values ​​forming a 3-dimensional vector. These three feature vectors are then mapped to three independent feature channels to construct a multi-channel feature matrix. This multi-channel feature matrix is ​​input into a pre-trained convolutional neural network containing multiple convolutional and pooling layers. Convolutional layers perform convolution operations by sliding convolution kernels across the feature matrix, extracting local feature patterns. Since the input is multi-channel, convolution operations can simultaneously consider information from different feature types, extracting fused feature patterns across feature channels. For example, a convolution kernel might learn a combination of patterns such as "steep rising edge in the time domain + moderate dominant frequency in the frequency domain + good morphological symmetry," corresponding to a high-quality waveform. Pooling layers downsample the output of the convolutional layers, preserving the most significant features while reducing computational cost. After multiple layers of convolution and pooling, the convolutional neural network extracts high-level abstract feature representations. Fully connected layers map multidimensional features to a scoring space. The output layer is a single neuron whose output value is the overall quality score. The overall quality score is a value from 0 to 100, where 100 represents perfect waveform quality and 0 represents extremely poor waveform quality. The overall quality score is learned by the convolutional neural network through training on a large amount of labeled data. In the training data, each evoked potential waveform is labeled with a quality score by clinical experts. The convolutional neural network establishes a mapping relationship from multidimensional features to quality scores by learning from these samples. For example, the current evoked potential waveform, after being processed by a convolutional neural network, receives a comprehensive quality score of 78, indicating that the waveform quality is good but there is still room for improvement.

[0050] The method for dynamically calculating the second current adjustment step value based on the comprehensive quality score, and adjusting the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value includes: The quality improvement space is obtained by calculating the difference between the overall quality score and the preset quality score threshold. This quality improvement space is then transformed using a logarithmic function to obtain a nonlinear quality gain coefficient. The quality improvement space is the difference between the current overall quality score and the preset quality score threshold. The preset quality score threshold is the standard for judging whether the waveform quality is acceptable, typically set between 85 and 90 points. If the current score is 78 points and the threshold is 85 points, the quality improvement space is 7 points, representing how much improvement is needed to reach the ideal state. The quality improvement space is transformed using a logarithmic function to obtain the nonlinear quality gain coefficient. The characteristic of a logarithmic function is that the output changes rapidly when the input value is small and the output changes slowly when the input value is large. This nonlinear transformation is designed so that when the quality improvement space is large (the score is far from the threshold), a relatively aggressive adjustment strategy can be adopted; when the quality improvement space is small (the score is already close to the threshold), careful adjustment should be made to avoid over-adjustment. The nonlinear quality gain coefficient after logarithmic transformation serves as an adjustment factor for subsequent step value calculations. A larger coefficient allows for larger current adjustment step values, while a smaller coefficient limits the current adjustment step values ​​to maintain small adjustments.

[0051] Based on the optimized stimulus current value, positive and negative micro-perturbation currents are applied respectively. The positive micro-perturbation current is the optimized stimulus current value plus a preset micro-perturbation amplitude, and the negative micro-perturbation current is the optimized stimulus current value minus the preset micro-perturbation amplitude. To determine which direction the current should be adjusted to improve the quality score, a trial-and-error micro-perturbation method is used. Based on the current optimized stimulus current value, positive and negative micro-perturbation currents are applied respectively. The positive micro-perturbation current is the optimized stimulus current value plus a preset micro-perturbation amplitude, which is usually set to 2-5% of the optimized stimulus current value. For example, if the optimized stimulus current value is 10 mA and the micro-perturbation amplitude is 0.3 mA, then the positive micro-perturbation current is 10.3 mA. The negative micro-perturbation current is the optimized stimulus current value minus the same micro-perturbation amplitude, i.e., 9.7 mA. These two micro-perturbation current values ​​represent two possible adjustment directions of current increase and decrease, respectively.

[0052] The evoked potential waveforms of the positive and negative micro-perturbation currents are acquired, and the quality improvement direction and effective quality gradient are determined. A second current adjustment step value is calculated based on the effective quality gradient and the nonlinear quality gain coefficient. The second current adjustment step value is applied to the optimized stimulation current value according to the determined quality improvement direction to obtain a fine-tuned stimulation current value. The effective quality gradient reflects the quality improvement brought about by a unit current change, while the nonlinear quality gain coefficient reflects the distance to the target and the appropriate speed of approach. The current adjustment step value is inversely proportional to the effective quality gradient; a large gradient indicates high adjustment sensitivity requiring only small steps, while a small gradient indicates low adjustment sensitivity requiring large steps. The current adjustment step value is directly proportional to the nonlinear quality gain coefficient; a large coefficient allows for larger steps to accelerate convergence, while a small coefficient requires small steps for cautious adjustment. The second current adjustment step value calculated in this way can adaptively balance adjustment speed and stability. If the quality improvement direction is current increase, the current adjustment step value is added to the optimized stimulation current value; if the quality improvement direction is current decrease, the current adjustment step value is subtracted from the optimized stimulation current value. Stimulation is applied using this fine-tuned stimulation current value, and evoked potential waveforms are collected to calculate a new comprehensive quality score. If the new comprehensive quality score exceeds a preset quality score threshold, the fine-tuning process ends, and the current fine-tuned stimulation current value becomes the final stimulation current value. If the threshold has not yet been reached, the above process is repeated to continue fine-tuning. Each iteration is based on the latest stimulation current value to perform micro-perturbation detection and step value calculation, gradually approaching the optimal parameters until the quality score meets the requirements.

[0053] The method for obtaining the induced potential waveforms of the positive and negative micro-perturbation currents and determining the quality improvement direction and effective quality gradient includes: Waveforms of evoked potentials corresponding to positive and negative micro-perturbation currents were acquired, and their time-domain, frequency-domain, and morphological features were extracted. Convolutional neural networks were used to obtain quality scores for both positive and negative micro-perturbation. The difference between the positive and negative micro-perturbation quality scores was calculated as the positive quality gradient, and the difference between the negative and negative micro-perturbation quality scores was calculated as the negative quality gradient. When the positive quality gradient was positive, the direction of current increase was considered the direction of quality improvement; when the negative quality gradient was positive, the direction of current decrease was considered the direction of quality improvement. The quality gradient corresponding to the direction of quality improvement was selected as the effective quality gradient. The target nerve was stimulated using a positive micro-perturbation current. Stimulation parameters remained the same as when the stimulation current was optimized, except for the current intensity, including pulse width, frequency, and stimulation duration. After stimulation, evoked potential waveforms were acquired using recording electrodes; these waveforms were recorded as positive micro-perturbation evoked potentials. Negative micro-perturbation evoked potentials were acquired by applying a negative micro-perturbation current (optimized stimulation current value minus preset micro-perturbation amplitude). Sufficient time must be allowed between these two micro-perturbation stimuli and the previous stimulation with the optimized current to ensure that the neural tissue fully recovers to a resting state and to avoid residual effects from the previous stimulation. This interval is typically several seconds to tens of seconds. Complete feature extraction processes are performed on both positive and negative micro-perturbation evoked potentials, and the results are input into the same convolutional neural network model. A positive micro-perturbation quality score is output for the positive micro-perturbation waveform, and a negative micro-perturbation quality score is output for the negative micro-perturbation waveform. Both quality scores are values ​​from 0 to 100, falling within the same scoring system as the previously calculated comprehensive quality score, allowing for direct numerical comparison. The positive quality gradient equals the positive micro-perturbation quality score minus the comprehensive quality score (i.e., the quality score corresponding to the optimized stimulation current), reflecting the change in quality score after increasing the micro-perturbation amplitude of the current. If the positive micro-perturbation quality score is higher than the comprehensive quality score (positive difference), it indicates that increasing the current improves quality; if the positive micro-perturbation quality score is lower than the comprehensive quality score (negative difference), it indicates that increasing the current decreases quality. The negative quality gradient equals the negative micro-perturbation quality score minus the overall quality score, and has a similar meaning, reflecting the impact of reducing current on quality. For example, if the overall quality score corresponding to the optimized stimulus current is 78, and the score after a positive micro-perturbation is 80, the positive quality gradient is +2; if the score after a negative micro-perturbation is 76, the negative quality gradient is -2. The rule for determining the direction based on the signs of the two quality gradients is straightforward: when the positive quality gradient is positive, the direction of increasing current is considered the direction of quality improvement, because a positive value indicates that increasing current leads to quality improvement. When the negative quality gradient is positive, the direction of decreasing current is considered the direction of quality improvement, because a positive value indicates that decreasing current leads to quality improvement. In the example above, the positive quality gradient +2 is positive, and the negative quality gradient -2 is negative; therefore, the direction of increasing current is considered the direction of quality improvement.The quality gradient corresponding to the direction of quality improvement is selected as the effective quality gradient. If the direction of current increase is determined to be the direction of quality improvement, the effective quality gradient equals the positive quality gradient; if the direction of current decrease is determined to be the direction of quality improvement, the effective quality gradient equals the absolute value of the negative quality gradient. The magnitude of the effective quality gradient represents the rate of improvement in the quality score when adjusting along the direction of improvement. By determining the adjustment direction and gradient using this method based on actual detection, the relationship between current adjustment and quality change under the current state can be accurately grasped, achieving precise fine-tuning control.

[0054] Example 2: Based on the same inventive concept, such as Figure 2 As shown, this embodiment also provides a current regulation system for the nerve electrical stimulation process. The system includes: a tissue impedance distribution data acquisition and analysis module, an initial stimulation parameter generation module, a waveform feature parameter extraction module, a coarse-tuning stimulation current analysis module, and a fine-tuning stimulation current analysis module, with each module connected in sequence via communication. The tissue impedance distribution data acquisition and analysis module is used to acquire patient characteristic parameters, including age, gender, body mass index, and subcutaneous fat thickness; apply microcurrent at the electrode placement location to perform tissue impedance scanning to obtain tissue impedance distribution data along the area surrounding the electrode; and identify the location and depth of the target nerve based on the tissue impedance distribution data.

[0055] The initial stimulation parameter generation module is used to obtain recommended initial stimulation current values ​​and personalized target amplitudes from the patient characteristic parameters, target nerve location, and target nerve depth through a deep neural network model; the deep neural network model is trained using historical patient data.

[0056] The waveform feature parameter extraction module is used to acquire evoked potential waveforms and extract waveform feature parameters after applying electrical stimulation to the target nerve location according to the recommended initial stimulation current value. The waveform feature parameters include evoked potential amplitude, latency, waveform morphology coefficient, and signal-to-noise ratio.

[0057] The coarse-adjustment stimulation current analysis module is used to obtain the deviation between the evoked potential amplitude and the personalized target amplitude, and to obtain a first current adjustment step value by analyzing the waveform feature parameters; the recommended initial stimulation current value is adjusted according to the first current adjustment step value to obtain the coarse-adjustment stimulation current until the evoked potential amplitude exceeds the personalized target amplitude, then the current coarse-adjustment stimulation current value is recorded as the optimized stimulation current value, and the time domain features, frequency domain features and morphological features of the evoked potential waveform are extracted and a comprehensive quality score is obtained through a convolutional neural network.

[0058] The fine-tuning stimulation current analysis module is used to dynamically calculate the second current adjustment step value based on the comprehensive quality score, and adjust the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value until the comprehensive quality score of the fine-tuned stimulation current value is greater than the preset quality score threshold. Then, the current fine-tuned stimulation current value is taken as the final stimulation current value.

[0059] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0060] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for regulating the current during a nerve electrical stimulation process, characterized in that, The method includes: The patient's characteristic parameters are acquired, including age, gender, body mass index, and subcutaneous fat thickness; a microcurrent is applied at the electrode placement location to perform tissue impedance scanning to obtain tissue impedance distribution data along the area surrounding the electrode; the location and depth of the target nerve are identified based on the tissue impedance distribution data. The patient's characteristic parameters, target nerve location, and target nerve depth are used to obtain recommended initial stimulation current values ​​and personalized target amplitudes through a deep neural network model; the deep neural network model is trained using historical patient data. After applying electrical stimulation to the target nerve location according to the recommended initial stimulation current value, the evoked potential waveform is acquired and the waveform feature parameters are extracted. The waveform feature parameters include the evoked potential amplitude, latency, waveform morphology coefficient and signal-to-noise ratio. The deviation between the evoked potential amplitude and the personalized target amplitude is obtained, and a first current adjustment step value is obtained by analyzing the waveform feature parameters. The recommended initial stimulation current value is adjusted according to the first current adjustment step value to obtain the coarse-tuned stimulation current until the evoked potential amplitude exceeds the personalized target amplitude. Then, the current coarse-tuned stimulation current value is recorded as the optimized stimulation current value, and the time domain features, frequency domain features and morphological features of the evoked potential waveform are extracted and a comprehensive quality score is obtained through a convolutional neural network. The second current adjustment step value is dynamically calculated based on the comprehensive quality score. The optimized stimulation current value is adjusted according to the second current adjustment step value to obtain the fine-tuned stimulation current value. When the comprehensive quality score of the fine-tuned stimulation current value is greater than the preset quality score threshold, the current fine-tuned stimulation current value is taken as the final stimulation current value.

2. The current regulation method for a nerve electrical stimulation process according to claim 1, characterized in that, The method for identifying the location and depth of a target nerve based on the tissue impedance distribution data includes: Based on the tissue impedance distribution data, impedance abrupt change regions at the tissue interface are identified; connectivity analysis and morphological feature extraction are performed on the impedance abrupt change regions to obtain multiple candidate neural regions, the morphological features including the spatial extension direction of the impedance abrupt change region, the region aspect ratio, and impedance contrast. The morphological features of the candidate neural regions are matched with the pre-stored standard impedance feature templates of neural tissue to obtain the matching confidence of each candidate neural region. Based on the matching confidence, the neural region with the highest matching confidence is selected as the target neural location. The local impedance distribution curve at the target neural location is obtained and the impedance attenuation characteristics are extracted. The impedance attenuation characteristics are mapped to the conduction path length from the electrode surface to the target neural location. The target neural depth is obtained by solving the conduction path length through inversion calculation.

3. The current regulation method for a nerve electrical stimulation process according to claim 2, characterized in that, The method for identifying impedance abrupt change regions at tissue interfaces based on the tissue impedance distribution data includes: The tissue impedance distribution data is filtered to eliminate measurement noise and artifact interference to obtain a smooth impedance distribution field; the impedance gradient field is constructed by calculating the impedance gradient values ​​in each spatial direction in the region around the electrode; and the impedance abrupt change region of the tissue interface is identified by the gradient magnitude and gradient direction distribution characteristics in the impedance gradient field.

4. The current regulation method for a nerve electrical stimulation process according to claim 1, characterized in that, The method for obtaining the deviation between the induced potential amplitude and the personalized target amplitude, and analyzing the waveform characteristic parameters to obtain the first current adjustment step value, includes: The difference between the current evoked potential amplitude and the personalized target amplitude is calculated and normalized to obtain the amplitude deviation coefficient; the latency, waveform morphology coefficient and signal-to-noise ratio in the waveform characteristic parameters are respectively calculated with respect to their corresponding normal reference ranges to obtain the latency deviation, morphology deviation and signal-to-noise ratio deviation. The latency deviation, morphological deviation, and signal-to-noise ratio deviation are fused by nonlinear weighting to obtain a comprehensive waveform anomaly index; the amplitude deviation coefficient is calculated by an adaptive step mapping function to obtain a first current adjustment step value, and the mapping slope of the adaptive step mapping function is adjusted according to the comprehensive waveform anomaly index.

5. The current regulation method for the nerve electrical stimulation process according to claim 4, characterized in that, The method of calculating the first current adjustment step value by means of the amplitude deviation coefficient through an adaptive step mapping function, and adjusting the mapping slope of the adaptive step mapping function according to the comprehensive waveform anomaly index includes: The adaptive step mapping function adopts a piecewise linear mapping relationship, dividing the range of amplitude deviation coefficient into a fast adjustment segment and a fine adjustment segment. The mapping slope in each adjustment segment represents the increment of the current adjustment step value corresponding to a unit amplitude deviation coefficient. Obtain the initial mapping slope baseline value for the current patient, and assign the initial mapping slope baseline value to the fast adjustment segment and the fine adjustment segment respectively as the base mapping slope; after normalizing the comprehensive waveform abnormality index, obtain the slope attenuation coefficient through inverse proportional function transformation; The base mapping slope is adjusted according to the slope attenuation coefficient to obtain the adjusted mapping slope; the corresponding adjusted mapping slope is selected according to the adjustment segment to which the current amplitude deviation coefficient belongs, and the amplitude deviation coefficient and the adjusted mapping slope are linearly calculated to obtain the first current adjustment step value.

6. The current regulation method for a nerve electrical stimulation process according to claim 1, characterized in that, The method for extracting the temporal, frequency, and morphological features of the evoked potential waveform and obtaining a comprehensive quality score through a convolutional neural network includes: The evoked potential waveform is subjected to time-domain analysis to extract time-domain features, including the peak occurrence time, the time interval between adjacent peaks, the rising slope of the waveform, and the falling slope of the waveform. A fast Fourier transform is performed on the evoked potential waveform to obtain frequency-domain spectra, from which frequency-domain features are extracted. These features include the dominant frequency component, the proportion of spectral energy within a preset physiological frequency band, and the concentration of spectral energy distribution. Morphological analysis is also performed on the evoked potential waveform to extract morphological features, including waveform left-right symmetry indices, waveform envelope smoothness indices, and baseline drift amplitude. After normalizing the time-domain features, frequency-domain features, and morphological features, they are organized into three feature vectors according to feature type. The three feature vectors are then mapped to three independent feature channels to construct a multi-channel feature matrix. The fusion feature pattern across feature channels is extracted through the convolutional and pooling layers of a convolutional neural network. Finally, the fusion feature pattern is mapped to the scoring space through a fully connected layer to obtain a comprehensive quality score.

7. The current regulation method for a nerve electrical stimulation process according to claim 1, characterized in that, The method for dynamically calculating the second current adjustment step value based on the comprehensive quality score, and adjusting the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value includes: The difference between the overall quality score and the preset quality score threshold is calculated to obtain the quality improvement space. The quality improvement space is then transformed by a logarithmic function to obtain a nonlinear quality gain coefficient. Based on the optimized stimulation current value, positive micro-perturbation current and negative micro-perturbation current are applied respectively. The positive micro-perturbation current is the optimized stimulation current value plus a preset micro-perturbation amplitude, and the negative micro-perturbation current is the optimized stimulation current value minus the preset micro-perturbation amplitude. The evoked potential waveforms of the positive and negative micro-perturbation currents are obtained, and the quality improvement direction and effective quality gradient are determined. The second current adjustment step value is calculated based on the effective quality gradient and the nonlinear quality gain coefficient. The second current adjustment step value is applied to the optimized stimulation current value according to the determined quality improvement direction to obtain the fine-tuned stimulation current value.

8. The current regulation method for a nerve electrical stimulation process according to claim 7, characterized in that, The method for obtaining the induced potential waveforms of the positive and negative micro-perturbation currents and determining the quality improvement direction and effective quality gradient includes: The evoked potential waveforms corresponding to positive and negative micro-perturbation currents were collected separately, and their time-domain, frequency-domain, and morphological features were extracted. A convolutional neural network was used to obtain the quality scores for the positive and negative micro-perturbation currents, respectively. The difference between the positive micro-perturbation quality score and the overall quality score was calculated as the positive quality gradient, and the difference between the negative micro-perturbation quality score and the overall quality score was calculated as the negative quality gradient. When the positive quality gradient was positive, the direction of current increase was determined as the quality improvement direction; when the negative quality gradient was positive, the direction of current decrease was determined as the quality improvement direction. The quality gradient corresponding to the quality improvement direction was selected as the effective quality gradient.

9. A current regulation system for a nerve electrical stimulation process, characterized in that, The system includes: a tissue impedance distribution data acquisition and analysis module, an initial stimulation parameter generation module, a waveform feature parameter extraction module, a coarse-tuning stimulation current analysis module, and a fine-tuning stimulation current analysis module, with each module connected in sequence via communication. The tissue impedance distribution data acquisition and analysis module is used to acquire patient characteristic parameters, including age, gender, body mass index, and subcutaneous fat thickness; apply a microcurrent at the electrode placement location to perform tissue impedance scanning to obtain tissue impedance distribution data along the area surrounding the electrode; and identify the location and depth of the target nerve based on the tissue impedance distribution data. The initial stimulation parameter generation module is used to obtain recommended initial stimulation current values ​​and personalized target amplitudes by using the patient characteristic parameters, target nerve location, and target nerve depth through a deep neural network model; the deep neural network model is trained using historical patient data. The waveform feature parameter extraction module is used to acquire evoked potential waveforms and extract waveform feature parameters after applying electrical stimulation to the target nerve location according to the recommended initial stimulation current value. The waveform feature parameters include evoked potential amplitude, latency, waveform morphology coefficient and signal-to-noise ratio. The coarse-adjustment stimulation current analysis module is used to obtain the deviation between the evoked potential amplitude and the personalized target amplitude, and to obtain a first current adjustment step value by analyzing the waveform feature parameters; the recommended initial stimulation current value is adjusted according to the first current adjustment step value to obtain the coarse-adjustment stimulation current until the evoked potential amplitude exceeds the personalized target amplitude, then the current coarse-adjustment stimulation current value is recorded as the optimized stimulation current value, and the time domain features, frequency domain features and morphological features of the evoked potential waveform are extracted and a comprehensive quality score is obtained through a convolutional neural network; The fine-tuning stimulation current analysis module is used to dynamically calculate the second current adjustment step value based on the comprehensive quality score, and adjust the optimized stimulation current value according to the second current adjustment step value to obtain the fine-tuned stimulation current value until the comprehensive quality score of the fine-tuned stimulation current value is greater than the preset quality score threshold. Then, the current fine-tuned stimulation current value is taken as the final stimulation current value.

Images