A method and system for quantitative detection of sensory nerves based on electroencephalogram features

Abstract

The application discloses a kind of sensation nerve quantitative detection method and system based on electroencephalogram characteristics, it is related to electroencephalogram measurement technical field, including: establishing and corresponding detection file of user;Respectively current stimulation module and electroencephalogram detection module are worn in the wrist of user and head;By current stimulation module, current stimulation is gradually applied to user, and gamma wave in the electroencephalogram signal of user is collected in real time by electroencephalogram detection module;Analysis gamma wave in the electroencephalogram signal of user, whether the current intensity that current stimulation module applies to user exceeds the physical stimulation perception threshold of subject is judged;Analysis sensation disorder severity classification.The application has the advantages that: detection is carried out by electroencephalogram signal, does not depend on the subjective feedback of patient, avoids the interference of emotion, psychology and other factors, even if special patient cannot carry out self-perception report, sensation nerve quantitative detection can also be completed by objective physiological response of electroencephalogram.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroencephalogram measurement, in particular to a sensory nerve quantitative detection method and system based on electroencephalogram characteristics. BACKGROUND

[0002] Quantitative sensory testing (QST) is a method of measuring the threshold of light touch (or pressure), vibration, temperature (cold and warm), and pain (cold and heat) in subjects using physical stimulation, and evaluating the perception function of peripheral nerves and central nervous system of the subjects, which is mainly used for detecting abnormal conditions of sensory functions such as pain, touch, temperature and vibration. The conventional quantitative sensory testing method mainly relies on standardized stimulation and subjective response of patients, and is used for diagnosing peripheral neuropathy, central nervous system diseases and pain management. This method has some defects that are difficult to overcome. First, the conventional method cannot accurately quantify the physical stimulation received by the patient, and the subjective report of the patient is not accurate and objective, and is easily affected by factors such as emotion, psychological state and pain tolerance, so it cannot accurately reflect the severity of sensory disturbance when evaluating the function of the nervous system. Secondly, the detection process is time-consuming, and the current intensity needs to be manually adjusted by medical staff step by step. This process usually needs to be tried several times to determine the sensory threshold of the patient. Each level of adjustment requires the medical staff to observe the patient's response, record the data, and the process is relatively cumbersome. Finally, the conventional quantitative sensory testing cannot be evaluated when facing special patients who may not be able to report self-perception due to consciousness disorders, language disorders or being too young.

[0003] Current perception threshold (CPT) is a method of evaluating sensory nerve function by electrical stimulation that has gradually become mainstream in recent years. Different frequencies of electrical signals can activate different types of nerve fibers, thereby evaluating the functional status of different nerve fibers. CPT is commonly used to evaluate the conduction ability of sensory nerves, which can help to identify nerve damage early, especially small fiber nerve damage, which is crucial for early intervention of chronic diseases such as diabetes or drug-induced nerve damage. This detection can also evaluate the pain sensitivity of individuals, especially in the treatment of pain diseases (such as fibromyalgia, chronic pain syndrome, etc.), which helps doctors to quantify the changes in pain perception. When performing current perception threshold measurement, the physical stimulation from the outside not only activates the peripheral nervous system, but also affects the cortical activity of the brain through nerve conduction. The brain's response to these stimuli is often manifested as changes in electroencephalogram (EEG), which provides additional available information for quantitative nerve detection, and helps to better understand the response state of the nervous system.

[0004] Conventional sensory nerve quantitative detection methods cannot accurately quantify the pain stimulus received by the patient, and the patient's subjective report is not accurate and objective, and is easily affected by factors such as mood, psychological state, and pain tolerance, and cannot accurately reflect the degree of nerve damage when evaluating nerve function. Secondly, the detection process is time-consuming and requires medical personnel to manually adjust the current intensity step by step, which usually requires multiple attempts to determine the patient's sensory threshold. Each level of adjustment requires medical personnel to observe the patient's response, record data, and the process is relatively cumbersome. Finally, the conventional sensory nerve quantitative detection method faces special patient groups, such as patients who may be unable to self-perceive due to consciousness disorders, language disorders, or being too young, resulting in the traditional method being unable to accurately assess their nerve function. SUMMARY

[0005] To solve the above technical problems, a sensory nerve quantitative detection method and system based on electroencephalogram features are provided, and a low-granularity nonlinear automatic adjustment current intensity strategy is proposed, which adjusts the current intensity nonlinearly, uses the degree of electroencephalogram activity as the termination basis, and quickly measures the electroencephalogram signal under different intensity current stimulation, so that the electroencephalogram signal acquisition process of the patient can be quickly and automatically completed.

[0006] To achieve the above purpose, the technical scheme adopted by the present application is:

[0007] A sensory nerve quantitative detection method based on electroencephalogram features, comprising:

[0008] Establish a detection file corresponding to each user, which at least includes: user name, gender, age, contact information, medical history record, detection date and detection time;

[0009] The current stimulation module and the electroencephalogram detection module are respectively worn on the wrist and head of the user;

[0010] The current stimulation module applies current stimulation to the user step by step, and the electroencephalogram detection module acquires gamma waves in the user's electroencephalogram signal in real time;

[0011] Analyze the gamma waves in the user's electroencephalogram signal to determine whether the current intensity applied to the user by the current stimulation module exceeds the physical stimulation perception threshold of the subject, if yes, terminate the detection, and output the electroencephalogram data of the user, if not, continue the detection;

[0012] Process the electroencephalogram data of the user to obtain high-quality electroencephalogram signals in the target frequency band, and analyze the severity of sensory disorders based on the high-quality electroencephalogram signals.

[0013] As a preferred embodiment of the present application, the current stimulation module applies current stimulation to the user step by step specifically includes:

[0014] Set the initial current intensity and the current intensity adjustment interval to 0.05 seconds;

[0015] The current intensity is adjusted based on a low-granularity nonlinear automatic adjustment strategy. The specific current intensity adjustment steps are as follows:

[0016] Let the increase in the initial current be denoted as . ;

[0017] The magnitude of each subsequent increase in current is as follows:

[0018]

[0019] in, Let k be the current change amplitude at time step t, k be the deceleration rate, and t be the time step.

[0020] As a preferred embodiment of this solution, the specific steps for analyzing the gamma waves in the user's electroencephalogram (EEG) signal and determining whether the current intensity applied by the current stimulation module to the user exceeds the subject's physical stimulation perception threshold are as follows:

[0021] Using ten cycles as the grouping standard, amplitude data of ten consecutive cycles of Gamma wave were collected and formed into an amplitude array;

[0022] Calculate the ratio of the sum of amplitudes in two adjacent amplitude arrays, and determine whether the ratio is greater than the amplitude change threshold. If it is, the current intensity is determined to have exceeded the physical stimulus perception threshold of the subject, and the experiment is terminated. If not, no response is made.

[0023] Furthermore, a sensory nerve quantitative detection system based on EEG characteristics is proposed to implement the aforementioned sensory nerve quantitative detection method based on EEG characteristics. This system includes: a user detection profile construction module, a current stimulation module, an EEG signal acquisition module, a stimulation threshold determination module, an EEG signal processing module, and a neural quantitative analysis module, wherein:

[0024] The user test profile building module is used to create a digital profile containing the user's name, gender, age, contact information, medical history, test date, and test time.

[0025] The current stimulation module includes a wearable electrode array configured to apply stepwise increasing current stimulation to the subject according to a preset nonlinear adjustment strategy.

[0026] The EEG signal acquisition module integrates a high-precision Gamma wave sensor array to capture the subject’s head electrophysiological signals in real time and extract the time-frequency features of the Gamma band.

[0027] The stimulation threshold determination module dynamically determines whether the physical stimulation perception threshold is exceeded by analyzing the continuous period Gamma amplitude change rate, and triggers a detection termination instruction.

[0028] The electroencephalogram signal processing module processes the electroencephalogram data of the user to obtain high-quality electroencephalogram signals in a target frequency band, and analyzes the severity grading of the sensory disorder based on the high-quality electroencephalogram signals.

[0029] Compared with the prior art, the present application has the following beneficial effects:

[0030] The present application provides a sensory nerve quantitative detection method and system based on electroencephalogram features, and proposes a low-granularity nonlinear automatic adjustment current intensity strategy. The current intensity is adjusted nonlinearly to realize fast acquisition without losing the accuracy of the sensitive current intensity range. The active degree of the electroencephalogram wave is used as the termination measurement basis. The electroencephalogram wave signals and corresponding intensity information under different intensity current stimulation are collected for subsequent quantitative detection and classification. The present application detects through the electroencephalogram (EEG) signal, does not rely on the subjective feedback of the patient, avoids the interference of factors such as emotions and psychology, and even special patients who cannot self-perceive and report can complete the sensory nerve quantitative detection through the objective physiological response of the electroencephalogram. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 A flowchart of the sensory nerve quantitative detection method based on electroencephalogram features is provided for embodiment one.

[0032] Figure 2 A flowchart of the method for giving an optimized implementation scheme for setting the initial current intensity is provided for embodiment two. DETAILED DESCRIPTION

[0033] The following description is used to disclose the present application so that those skilled in the art can implement the present application. In the following description EMBODIMENTS

[0034] Referring to Figure 1 As shown in the figure, a sensory nerve quantitative detection method based on electroencephalogram features comprises the following steps:

[0035] A detection file corresponding to the user is established, and the detection file at least includes the user's name, gender, age, contact information, medical history record, detection date and detection time;

[0036] The current stimulation module and the electroencephalogram detection module are respectively worn on the wrist and head of the user;

[0037] The current stimulation module gradually applies current stimulation to the user, and the electroencephalogram detection module real-time acquires the gamma wave in the electroencephalogram signal of the user;

[0038] In this step, the specific implementation means are: setting the initial current intensity as 0.01 mA, and setting the current intensity adjustment interval as 0.05 seconds;

[0039] The current intensity is adjusted based on a low-granularity nonlinear automatic adjustment current intensity strategy, and the specific current intensity adjustment steps are:

[0040] An initial current increase change amplitude is set, denoted as , According to experimental data analysis, half of the minimum value between the minimum current intensity of the electric stimulation perception in the historical test data and the initial current intensity is taken;

[0041] Each subsequent current increase change amplitude is:

[0042] ;

[0043] wherein, is the current change amplitude at time step t, k is the decreasing rate, and t is the time step;

[0044] The gamma wave in the user's electroencephalogram signal is analyzed to determine whether the current intensity applied to the user by the current stimulation module exceeds the physical stimulation perception threshold of the subject, if yes, the detection is terminated, and the electroencephalogram data of the user is output, if not, the detection is continued;

[0045] Specifically, in this step, ten cycles are taken as the grouping standard, the amplitude data of ten consecutive cycles of the gamma wave are collected, and an amplitude array is formed;

[0046] The ratio of the sum of the amplitudes in two adjacent amplitude arrays is calculated, and it is determined whether the ratio is greater than the amplitude change threshold, if yes, it is determined that the current intensity has exceeded the physical stimulation perception threshold of the subject, the experiment is terminated, if not, no response is made, and the detection is continued;

[0047] It can be understood that the gamma wave in the electroencephalogram signal is highly related to the pain perception of the human body, and the amplitude of the gamma wave significantly increases when the human body is subjected to pain stimulation. Therefore, the amplitude change of the gamma wave is taken as the experiment termination basis in the present application: wherein, is the amplitude array of the gamma wave in ten consecutive cycles, is the i-th amplitude array, For the i-1th amplitude array, C is the threshold of the amplitude change of the gamma wave, when the ratio of the sum of two adjacent amplitude arrays of the gamma wave is greater than C, it is determined that the current intensity has exceeded the physical stimulation perception threshold of the subject, the experiment is terminated, and the determination of the threshold of the amplitude change needs to collect data through an induction experiment, analyze the data time curve after normalization, and find the normalized value corresponding to the point with the maximum slope of the curve.

[0048] Based on the electroencephalogram data of the user, high-quality electroencephalogram signals in a target frequency band are obtained, and the severity of the sensory disorder is analyzed based on the high-quality electroencephalogram signals.

[0049] Embodiment two:

[0050] Referring to Figure 2 The embodiment further optimizes the setting of the initial current intensity based on the embodiment one, and specifically includes:

[0051] Based on gender, age, and medical history records, historical test data are classified to obtain a plurality of data class sets;

[0052] Based on the collected detection records of the user, a data class set corresponding to the user is determined, denoted as the corresponding data class set;

[0053] The current intensity corresponding to the minimum current intensity of the electrical stimulation perception of each historical test data in the corresponding data class set is determined, denoted as the sample current intensity;

[0054] The average value and the standard deviation of all sample current intensities corresponding to the corresponding data class set are calculated, and the value obtained by subtracting 6 times the standard deviation from the average value is used as the initial current intensity;

[0055] It can be understood that at the initial stage of system operation, due to the lack of sufficient detection data as sample analysis, in order to ensure full coverage during detection, the initial current intensity needs to be as small as possible to reduce the missed detection rate. Therefore, the initial current intensity of 0.01 mA is given in the embodiment one. However, due to the differences between different test individuals, the test duration of users with high physical stimulation perception threshold may be longer. Therefore, based on the accumulation of sufficient detection data as samples by the system, a more optimized scheme for determining the initial current intensity is proposed. The detection objects corresponding to the historical detection data are classified by features, for example, the age range + gender + whether suffering from rheumatic diseases are used as classification standards. The historical data are classified based on the standards. During the measurement test, the more similar class group is selected based on the user's condition, and the data in the class group are used to determine the optimized initial current intensity.

[0056] In the scheme, the initial current intensity based on data analysis adopts the 6 sigma criterion, that is, the initial current intensity is determined in the manner of average value minus 6 standard deviations, based on the normal distribution criterion, the probability that the minimum current intensity of the user's electric stimulation perception is above the initial current intensity is 99.99%, based on the Chebyshev lower limit calculation, the probability that the minimum current intensity of the user's electric stimulation perception is above the initial current intensity is 98.61%, by using the initial current intensity, the applied current intensity can reach the minimum current intensity of the user's electric stimulation perception as soon as possible, thereby the test time can be effectively shortened and the detection efficiency can be improved.

[0057] Embodiment three:

[0058] In the embodiment, based on the basis of embodiment one, it is further pointed out that: based on the electroencephalogram data of the user, high-quality electroencephalogram signals of the target frequency band are obtained, and the analysis of the severity grading of the sensory disorder based on the high-quality electroencephalogram signals specifically includes:

[0059] A sensory nerve quantitative detection model is constructed, the sensory nerve quantitative detection model takes the electroencephalogram signals collected at different time steps as input, and takes the probability that the electroencephalogram signals at the time step belong to the perception of electric stimulation as output;

[0060] The sensory nerve quantitative detection model specifically includes:

[0061] The time convolution network module is composed of three causal convolution layers with filter sizes of 128, 256 and 128 respectively, each layer integrates an extended convolution mechanism and a residual connection, and is used for extracting neural electric signal time sequence features;

[0062] The contrast learning module pre-trains the features output by the time convolution network through a contrast loss function, and the contrast loss is calculated based on the Euclidean distance of positive and negative sample pairs;

[0063] The classification module includes a fully connected layer and a Softmax classifier, and is used for a downstream classification task;

[0064] The model training adopts a two-stage training method: in the pre-training stage, only the time convolution network and the contrast learning module are optimized, and in the downstream training stage, the time convolution network parameters are frozen and the classification module is optimized;

[0065] In the time convolution network module:

[0066] The causal convolution satisfies that the output y(t) satisfies, wherein x(t) is an input sequence, w(i) is a convolution kernel, and k is the size of the convolution kernel, so as to ensure that the output y(t) of the causal convolution depends only on the current and historical inputs;

[0067] The extended convolution function is: wherein d is an inflation factor, k is the size of the convolution kernel, and the extended convolution expands the receptive field by introducing gaps in the convolution kernel, thereby being able to capture longer temporal dependencies without increasing the computational cost;

[0068] The residual connection is to directly pass the input to the output for addition, improving the stability during model training and the liquidity of gradients during back propagation;

[0069] In the contrast learning module:

[0070] The contrast loss function adopts a formula based on Euclidean distance, wherein, represents that the sample pair is a positive sample pair or a negative sample pair, 1 is a positive sample pair, and 0 is a negative sample pair, represents the Euclidean distance between the sample features output by the temporal convolution network module, and m is a preset boundary distance, is a model loss value;

[0071] wherein the positive sample pair is electroencephalogram data perceived to the electric stimulation at two different time windows, and the negative sample pair is electroencephalogram data perceived to the electric stimulation and not perceived to the electric stimulation;

[0072] The model is pre-trained through contrast learning, and the goal is to learn high-quality feature representation, rather than directly classifying. The loss function of the model makes the feature distance of the positive sample pair as close as possible, and the feature distance of the negative sample pair as far as possible. The high-quality feature embedding space learned by the model in the pre-training stage can be used for downstream classification tasks. In the pre-training stage, the classification module does not participate in the training;

[0073] The classification module:

[0074] The loss function is, wherein N is the number of samples, is the true label of the i-th sample, has a value of 0 or 1, is the prediction probability that the i-th sample is the true label 1, and L is a model loss value;

[0075] The downstream task training stage is to use the pre-trained representation for the classification module. At this time, only the classification module of the model will be optimized. The features and labels of the training data are input into the classifier module, the loss is calculated, and the classifier is optimized.

[0076] The electroencephalogram signals collected by the user at different time steps are input into the sensory nerve quantitative detection model as a same batch, the probability that each time step corresponds to the electroencephalogram signal is perceived to the electric stimulation is obtained, it is judged whether the probability is greater than the probability threshold, if yes, the segment of the electroencephalogram signal is marked as “true”, that is, the electric stimulation corresponding to the segment of the electroencephalogram signal has been perceived, if not, no response is made;

[0077] wherein the probability threshold is set to 0.6, and the sensory nerve quantitative detection model output value greater than 0.6 is judged as "true", that is, the electrical stimulation corresponding to the brain electrical signal is perceived;

[0078] wherein the collected brain electrical signal is subjected to data noise separation, low frequency drift and high frequency noise are removed by filtering algorithm, and signals in specific frequency bands are extracted, and independent component analysis (ICA) is combined to separate and remove muscle electrical, eye movement and other artifact interference. The filtering algorithm used includes high-pass filtering, low-pass filtering, band-pass filtering and notch filtering. The EEG data after noise separation is cut according to time window, and high-quality brain electrical signals in the target frequency band are retained. The high-quality brain electrical signals in the target frequency band are input into the sensory nerve quantitative detection model;

[0079] The current intensity corresponding to all brain electrical signals marked as "true" is output, and the current intensity when the classification result is "true" for the first time is selected as the minimum current intensity of electrical stimulation perception. According to the minimum current intensity of electrical stimulation perception, the severity of sensory disorder of the experimental personnel is classified.

[0080] Example Four:

[0081] Based on the sensory nerve quantitative detection method based on brain electrical characteristics proposed in examples one to three, the present embodiment proposes a sensory nerve quantitative detection system based on brain electrical characteristics, which comprises a user detection file construction module, a current stimulation module, a brain electrical signal acquisition module, a stimulation threshold determination module, a brain electrical signal processing module and a neural quantitative analysis module, wherein:

[0082] The user detection file construction module is used to create a digital file containing the user's name, gender, age, contact information, medical history record, detection date and detection time;

[0083] The current stimulation module comprises a wearable electrode array configured to apply a stepwise increasing current stimulation to the subject according to a preset nonlinear adjustment strategy. The current stimulation module can directly set the initial current intensity to 0.01 mA, or determine the initial current intensity by the initial current intensity determination method proposed in example two;

[0084] The brain electrical signal acquisition module integrates a high-precision Gamma wave sensor array to capture the subject's head electrical physiological signals in real time and extract Gamma wave band time-frequency features;

[0085] The stimulation threshold determination module dynamically determines whether the physical stimulation perception threshold is exceeded by analyzing the continuous cycle Gamma wave amplitude change rate, and triggers the detection termination instruction;

[0086] The electroencephalogram signal processing module processes the electroencephalogram data of the user to obtain high-quality electroencephalogram signals in a target frequency band, and analyzes the severity grading of the sensory disorder based on the high-quality electroencephalogram signals:

[0087] Specifically, the electroencephalogram signal processing module integrates a sensory nerve quantitative detection model, which includes a time convolution network module, a contrast learning module, and a classification module. The model includes three time convolution networks with filter sizes of 128, 256, and 128. The time convolution network is based on convolution operation and uses causal convolution, extended convolution, and residual connection mechanisms to capture long-term dependencies in time series data. The main goal of causal convolution is to ensure that the convolution operation only uses data from the current time and previous time points, and does not use future data. For causal convolution, the output y(t) only depends on the current and previous inputs: where x(t) is the input sequence, w(i) is the convolution kernel, and k is the size of the convolution kernel. Extended convolution expands the receptive field by introducing gaps in the convolution kernel, allowing it to capture longer temporal dependencies without increasing computational complexity. The extended convolution function is: where d is the dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability of model training and the flow of gradients during backpropagation. The contrast learning module uses contrastive loss to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs. The contrastive loss function uses the Euclidean distance-based formula: where d is the dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability of model training and the flow of gradients during backpropagation. The contrast learning module uses contrastive loss to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs. The contrastive loss function uses the Euclidean distance-based formula: where d is the dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability of model training and the flow of gradients during backpropagation. The contrast learning module uses contrastive loss to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs. The contrastive loss function uses the Euclidean distance-based formula: where d is the dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability of model training and the flow of gradients during backpropagation. The contrast learning module uses contrastive loss to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs. The contrastive loss function uses the Euclidean distance-based formula: where d is the dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability of model training and the flow of gradients during backpropagation. The contrast learning module uses contrastive loss to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs. The contrastive loss function uses the Euclidean distance-based formula: where d is the dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability of model training and the flow of gradients during backpropagation. The contrast learning module uses contrastive loss to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs. The contrastive loss function uses the Euclidean distance-based formula: is the prediction probability of the i-th sample being the real label 1. Model training is divided into two stages: pre-training stage and downstream task training stage. Model pre-training is carried out through contrast learning, and the goal is to learn high-quality feature representation instead of direct classification. The loss function of the model makes the feature distance of positive sample pairs as close as possible, and the feature distance of negative sample pairs as far as possible. The high-quality feature embedding space learned by the model in the pre-training stage can be used for downstream classification tasks. In the pre-training stage, the classification module does not participate in the training. The downstream task training stage is to use the pre-trained representation for the classification module. At this time, only the classification module of the model will be optimized. The features and labels of the training data are input into the classifier module, the loss is calculated and the classifier is optimized;

[0088] The EEG signal processing module inputs the EEG signals collected by the experimenters at different time steps into the above model as the same batch, the model simultaneously classifies and predicts the EEG signals at all time steps, the value (range 0 to 1) output by the model represents the probability that a certain EEG signal belongs to the perception of electrical stimulation, and a threshold value of 0.6 is set as the classification standard, and the output value greater than 0.6 is judged as 'true', that is, the electrical stimulation corresponding to the EEG signal has been perceived. The current intensity corresponding to the EEG signal with a classification result of 'true' is recorded, and the current intensity when the classification result is first 'true' is selected as the minimum current intensity of electrical stimulation perception. According to the minimum current intensity of electrical stimulation perception, the experimenters are classified according to the severity of sensory disorder.

[0089] In summary, the advantages of the present application are: a low granularity nonlinear automatic adjustment current intensity strategy is proposed, the current intensity is adjusted through nonlinear adjustment, the accuracy of the sensitive current intensity interval is not lost, fast acquisition is realized, and the EEG signal and the corresponding intensity information under different intensity current stimulation are collected for subsequent quantitative detection and classification according to the termination measurement basis of the EEG activity level; the present application proposes a sensory nerve quantitative detection model, which is based on the connection between external stimulation and EEG signal, and performs sensory nerve quantitative detection on the EEG signal data characteristics under different current intensities, the model uses time convolution network and contrast learning to capture the long-time dependence of the input sequence, extracts EEG features in different states, uses a softmax classifier to detect and classify the extracted high-quality time sequence features, and finally classifies the severity of sensory disorder of the experimenters.

[0090] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for quantitative detection of sensory nerves based on electroencephalographic features, characterized by, The method comprises the following steps: establishing a detection file corresponding to each user, the detection file comprising at least the user's name, gender, age, contact information, medical history record, detection date and detection time; wearing the current stimulation module and the electroencephalogram detection module on the wrist and head of the user respectively; applying current stimulation to the user step by step through the current stimulation module, and collecting gamma waves in the electroencephalogram signal of the user in real time through the electroencephalogram detection module; analyzing the gamma waves in the electroencephalogram signal of the user to determine whether the current intensity applied to the user by the current stimulation module exceeds the physical stimulation perception threshold of the subject, and if so, terminating the detection and outputting the electroencephalogram data of the user, and if not, continuing the detection; processing the electroencephalogram data of the user to obtain high-quality electroencephalogram signals in a target frequency band, and analyzing the severity classification of sensory disorders based on the high-quality electroencephalogram signals; The method of analyzing the severity classification of sensory disorders based on the high-quality electroencephalogram signals comprises the following steps: constructing a sensory nerve quantitative detection model, wherein the sensory nerve quantitative detection model takes the electroencephalogram signals collected at different time steps as input, and takes the probability that the electroencephalogram signals at the time step belong to the time when the electrical stimulation is perceived as output; inputting the electroencephalogram signals collected at different time steps of the user into the sensory nerve quantitative detection model as a same batch, obtaining the probability that the corresponding electroencephalogram signals at each time step belong to the time when the electrical stimulation is perceived, and determining whether the probability is greater than a probability threshold, if so, marking the segment of electroencephalogram signals as "true", that is, the electrical stimulation corresponding to the segment of electroencephalogram signals has been perceived, and if not, taking no response; outputting the current intensity corresponding to all the electroencephalogram signals marked as "true", selecting the current intensity when the classification result is "true" for the first time as the minimum current intensity for electrical stimulation perception, and classifying the severity of sensory disorders of the experimental personnel according to the minimum current intensity for electrical stimulation perception; The sensory nerve quantitative detection model comprises the following steps: a time convolution network module composed of three causal convolution layers with filter sizes of 128, 256 and 128 respectively, each layer integrating an extended convolution mechanism and a residual connection, for extracting neural electrical signal time sequence features; a contrast learning module for pre-training the features output by the time convolution network through a contrast loss function, wherein the contrast loss is calculated based on the Euclidean distance of positive and negative sample pairs; a classification module containing a fully connected layer and a Softmax classifier, for downstream classification tasks; The model training adopts a two-stage training method: in the pre-training stage, only the time convolution network and the contrast learning module are optimized, and in the downstream training stage, the parameters of the time convolution network are frozen and the classification module is optimized.

2. The method according to claim 1, wherein, The method of applying current stimulation to the user step by step through the current stimulation module comprises the following steps: setting the initial current intensity as 0.01 mA and the current intensity adjustment interval as 0.05 seconds; adjusting the current intensity based on a low-granularity nonlinear automatic adjustment current intensity strategy, and the specific current intensity adjustment steps are as follows: An initial current increase variation amplitude is set, denoted as ; the current intensity is increased by 0.1 mA every 0.05 seconds; ； wherein, is the amplitude of the current change at time step t, k is the rate of decay, and t is the time step.

3. The method according to claim 2, wherein, The initial current intensity can also be determined by the following method: classifying historical test data based on gender, age and medical history record respectively to obtain a plurality of data class sets; Based on the collected user detection archives, determine the corresponding data class set of the user, denoted as the corresponding data class set; Determine the minimum current intensity of the electrical stimulation perception of each historical test data in the corresponding data class set, denoted as the sample current intensity; Calculate the average value and standard deviation of all sample current intensities corresponding to the corresponding data class set, and take the value of the average value minus 6 times the standard deviation as the initial current intensity.

4. The method of claim 1, wherein the method is characterized by, The specific steps of analyzing the gamma wave in the user's electroencephalogram signal to determine whether the current intensity applied to the user by the current stimulation module exceeds the physical stimulation perception threshold of the subject are: Group ten cycles as the grouping standard, collect the amplitude data of ten consecutive cycles of Gamma wave, and form an amplitude array; Calculate the ratio of the sum of amplitudes in the adjacent two amplitude arrays, and determine whether the ratio is greater than the amplitude change threshold. If yes, it is determined that the current intensity has exceeded the physical stimulation perception threshold of the subject, and the experiment is terminated. If not, no response is made.

5. The method of claim 1, wherein the method is characterized by, In the time convolution network module: A causal convolution satisfies that the output y(t) satisfies, where x(t) is the input sequence, w(i) is the convolution kernel, and k is the size of the convolution kernel, ensuring that the output y(t) satisfying the causal convolution only depends on the current and historical inputs. The extended convolution function is: where d is a dilation factor and k is the size of the convolution kernel. Residual connection is to pass the input directly to the output for addition, which improves the stability during model training and the liquidity of gradient during back propagation.

6. The method of claim 1, wherein the method is characterized by, In the contrast learning module: The contrast loss function adopts a formula based on Euclidean distance, wherein, represents that the sample pair is a positive sample pair or a negative sample pair, 1 is a positive sample pair, and 0 is a negative sample pair, represents the Euclidean distance between sample features output by the time convolution network module, m is a preset boundary distance, is a model loss value; Among them, the positive sample pair is the electroencephalogram data perceived to the electrical stimulation at two different time windows, and the negative sample pair is the electroencephalogram data perceived to the electrical stimulation and not perceived to the electrical stimulation.

7. The method according to claim 1, wherein the method is characterized by, In the classification module: The loss function is, where N is the number of samples, is the true label of the i-th sample, , the value of which is 0 or 1, is the predicted probability that the i-th sample is the true label 1, is the model loss value.

8. A sensory nerve quantitative detection system based on electroencephalogram features, for realizing the sensory nerve quantitative detection method based on electroencephalogram features as claimed in any one of claims 1-7, comprising a user detection archive construction module, a current stimulation module, an electroencephalogram signal acquisition module, a stimulation threshold determination module, an electroencephalogram signal processing module, and a nerve quantitative analysis module, wherein: The user detection archive construction module is used to create a digital archive containing the user's name, gender, age, contact information, medical history record, detection date, and detection time; The current stimulation module includes a wearable electrode array configured to apply a stepwise increasing current stimulation to the subject according to a preset nonlinear adjustment strategy; The electroencephalogram signal acquisition module integrates a high-precision Gamma wave sensor array to capture the subject's head electrophysiological signals in real time and extract Gamma wave band time-frequency features; The stimulation threshold determination module dynamically determines whether the physical stimulation perception threshold is exceeded by analyzing the continuous cycle Gamma wave amplitude change rate and triggers the detection termination instruction; The electroencephalogram signal processing module processes the electroencephalogram data of the user to obtain high-quality electroencephalogram signals in the target frequency band, and analyzes the severity grading of sensory disorders based on the high-quality electroencephalogram signals.

Images