Intelligent diagnosis report generation method and system based on electromyography and evoked potential data

Abstract

This invention provides a method and system for generating intelligent diagnostic reports based on electromyography (EMG) and evoked potential (EVP) data, belonging to the field of medical diagnostic technology. First, it acquires the subject's EMG and EVP data sets, performs difference compensation and association mapping on the two types of data, establishes dynamic association relationships, and obtains a cross-data type association feature set. Then, it calls a multi-round iterative intelligent diagnostic reasoning model to perform symptom association analysis and generate a preliminary diagnostic conclusion. Next, it integrates the preliminary diagnostic conclusion with clinical reference information to obtain an optimized diagnostic conclusion. Finally, it generates a structured intelligent diagnostic report based on the optimized diagnostic conclusion, including an electrophysiological evidence chain, association analysis details, clinical matching results, and phased suggestions. Each part can be viewed through an associated index, improving diagnostic accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical diagnostic technology, and more specifically, to an intelligent diagnostic report generation method and system based on electromyography and evoked potential data. Background Technology

[0002] In the field of medical diagnostics, accurate diagnosis of nervous system and related muscle diseases is directly related to the formulation of treatment plans and prognoses for patients. Currently, common diagnostic methods mainly rely on the detection and analysis of single physiological signals. For example, electromyography (EMG) can record the electrophysiological signals of muscles in different states. By analyzing these signals, the neuromuscular functional state of muscles can be understood, and it can be determined whether there are problems such as nerve damage or muscle lesions. Evoked potential detection records the electrophysiological response of the nervous system after being given specific stimuli (such as visual, auditory, or somatosensory stimuli), thereby assessing the functional integrity of neural conduction pathways.

[0003] However, existing diagnostic methods have significant limitations. On the one hand, the information obtained from a single test is relatively one-sided and cannot fully reflect complex physiological and pathological states. Different physiological signals are inherently related and mutually influential; analyzing only a single signal can easily miss key information, leading to inaccurate diagnostic results. For example, some neurological diseases may simultaneously affect muscle function and nerve conduction pathways; relying solely on electromyography or evoked potentials may not accurately determine the type and severity of the disease. On the other hand, existing diagnostic report generation methods lack systematicity and coherence. Diagnostic reports often simply list test data and preliminary diagnostic results without in-depth analysis of the correlation between different test data or effective integration of clinical reference information with test results. This forces doctors to spend a significant amount of time on comprehensive analysis when interpreting the reports, increasing the difficulty and time cost of diagnosis and making judgment bias more likely. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating intelligent diagnostic reports based on electromyography and evoked potential data, the method comprising: Acquire the subject's electromyography (EMG) data set and evoked potential (EVP) data set. The EMG data set includes continuous electrophysiological signal records of different muscle parts of the subject in the resting state, voluntary contraction state, and forced contraction state. The evoked potential data set includes multi-channel electrophysiological response records of the subject's nervous system under the action of visual stimulation, auditory stimulation, and somatosensory stimulation. The electromyography (EMG) dataset and the evoked potential (EVP) dataset are processed with data type difference compensation and dynamic association mapping to establish a dynamic association relationship between EMG features and EVP features under the same physiological state, resulting in a cross-data type association feature set including an association stability index. The pre-trained multi-round iterative intelligent diagnostic reasoning model is invoked to perform disease association analysis on the cross-data type association feature set, and a preliminary diagnostic conclusion containing multiple suspected diseases and their multi-dimensional association descriptions is generated. The preliminary diagnostic conclusion is fused with the subject's clinical reference information through a multi-source evidence chain to obtain an optimized diagnostic conclusion. Based on the optimized diagnostic conclusions, a structured intelligent diagnostic report is generated. This report includes the electrophysiological evidence chain for disease determination, details of cross-data type feature association analysis, clinical feature matching results, and phased clinical recommendations. Furthermore, each part of the report can be viewed via an associated index.

[0005] Furthermore, embodiments of the present invention also provide an intelligent diagnostic report generation system based on electromyography and evoked potential data, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described intelligent diagnostic report generation method based on electromyography and evoked potential data by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions stored in a computer-readable storage medium, a processor of an intelligent diagnostic report generation system based on electromyography and evoked potential data reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the intelligent diagnostic report generation system based on electromyography and evoked potential data to execute the above-described intelligent diagnostic report generation method based on electromyography and evoked potential data.

[0007] Based on the above, we first acquired electromyography (EMG) and evoked potential (EV) datasets from the subjects, encompassing rich electrophysiological information under different muscle states and various stimulation conditions. By performing data type difference compensation and dynamic correlation mapping on these two types of data, we established a dynamic correlation between EMG and EV features under the same physiological state, obtaining a cross-data type correlation feature set including a correlation stability index. This deeply explored the intrinsic connections between different physiological signals, reflecting the subject's physiological and pathological state. Then, we invoked a pre-trained multi-round iterative intelligent diagnostic reasoning model to perform disease correlation analysis on the cross-data type correlation feature set, generating preliminary diagnostic conclusions containing multiple suspected diseases and their multi-dimensional correlation descriptions. This multi-round iterative intelligent diagnostic reasoning model, after multiple rounds of training, can accurately identify potential diseases from complex correlation features and provide detailed correlation descriptions. It integrates the preliminary diagnostic conclusions with the subject's clinical reference information through multi-source evidence chain fusion, further optimizing the diagnostic conclusions. It fully considers the actual clinical situation, combining test data with the patient's clinical symptoms, medical history, and other information, avoiding misdiagnosis and missed diagnosis that might result from solely relying on test data, making the diagnostic results more consistent with clinical reality. The structured intelligent diagnostic report generated based on the optimized diagnostic conclusions not only includes key information such as the electrophysiological evidence chain for disease determination, details of cross-data type feature correlation analysis, and clinical feature matching results, but also provides phased clinical suggestions. Furthermore, the various sections can be accessed via related indexes, allowing doctors to quickly locate the information they need and improve diagnostic efficiency. Simultaneously, it helps doctors develop more scientific and reasonable treatment plans, thereby significantly improving the accuracy, comprehensiveness, and efficiency of diagnosing nervous system and related muscle diseases. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the intelligent diagnostic report generation method based on electromyography and evoked potential data provided in the embodiments of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the intelligent diagnostic report generation system based on electromyography and evoked potential data provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an intelligent diagnostic report generation method based on electromyography and evoked potential data according to an embodiment of the present invention. The following is a detailed description of the intelligent diagnostic report generation method based on electromyography and evoked potential data.

[0011] Step S110: Obtain the subject's electromyography (EMG) data set and evoked potential (EVP) data set. The EMG data set includes continuous electrophysiological signal records of different muscle parts of the subject in the resting state, voluntary contraction state, and forced contraction state. The evoked potential (EVP) data set includes multi-channel electrophysiological response records of the subject's nervous system under visual stimulation, auditory stimulation, and somatosensory stimulation.

[0012] In this embodiment, a subject suspected of having peripheral neuropathy is used as an example. First, electromyography (EMG) data sets of the subject are collected using an EMG device. Specifically, the upper limb muscle groups (such as biceps brachii, triceps brachii, wrist flexors, etc.) and lower limb muscle groups (such as quadriceps femoris, gastrocnemius, tibialis anterior, etc.) are selected as different muscle sites. For each muscle site, continuous electrophysiological signals are recorded in the resting state, voluntary contraction state, and forced contraction state. In the resting state, the subject must keep the muscles completely relaxed and not perform any active contraction movements, and the recording time is lasting for a certain duration; in the voluntary contraction state, the subject performs slight muscle contraction according to instructions, causing the muscle to produce a small force output, and the recording time is also lasting for a certain duration; in the forced contraction state, the subject contracts the muscle to the maximum extent, producing the maximum force output, and the recording time is correspondingly long. The above records form an EMG data set, which contains the electrophysiological signals of different muscle sites in the three activity states.

[0013] Next, the subject's evoked potential data set was collected. Visual, auditory, and somatosensory stimuli were applied using an evoked potential instrument. Visual stimulation used flashing light or pattern flipping stimulation, with the stimulation frequency and intensity set according to standard evoked potential testing procedures, and the electrophysiological response signals of the visual cortex were recorded. Auditory stimulation used short-sound stimulation, transmitted to the subject through headphones, and the electrophysiological response signals of the auditory cortex and related neural pathways were recorded. Somatosensory stimulation involved electrical stimulation of the subject's median nerve, ulnar nerve, tibial nerve, etc., and the electrophysiological response signals of the corresponding neural conduction pathways were recorded. The multi-channel electrophysiological response records of the nervous system under different stimuli constituted the evoked potential data set.

[0014] During data collection, sensitive data such as the subjects' physiological signals are involved. To protect the subjects' privacy, data encryption technology is used to encrypt the collected data, and secure transmission protocols, such as SSL / TLS, are used during data transmission to prevent data theft during transmission. Simultaneously, access control is implemented for the stored data; only authorized medical and technical personnel can access and process this data. Furthermore, data access and operations are logged for auditing and traceability, ensuring the privacy protection and prevention of leakage of sensitive data.

[0015] Step S120: Perform data type difference compensation and dynamic correlation mapping processing on the electromyography data set and the evoked potential data set to establish a dynamic correlation relationship between electromyography features and evoked potential features under the same physiological state, and obtain a cross-data type correlation feature set including the correlation stability index.

[0016] Step S121: Extract electrophysiological signals of each muscle part under different activity states from the electromyography data set, perform sampling frequency standardization and amplitude range normalization on the electrophysiological signals under each activity state, and convert electromyography signals with different sampling frequencies and amplitude ranges into electromyography standardized signals with a unified sampling frequency and a unified amplitude range.

[0017] For the electromyography (EMG) data set of the aforementioned subjects suspected of having peripheral neuropathy, the electrophysiological signals of each muscle group were first extracted under resting, voluntary, and forced contraction conditions. Since different muscle groups may have used different sampling frequencies for EMG signals (e.g., some muscle groups might have a set sampling frequency of a certain number of samples per second, while others might have a different frequency), sampling frequency standardization is necessary. Specifically, interpolation or downsampling methods are used to adjust all EMG signals with different sampling frequencies to a uniform sampling frequency, which is determined based on the commonly used standard frequency range for clinical EMG testing.

[0018] After standardizing the sampling frequency, the amplitude range of the electrophysiological signals for each activity state is normalized. The amplitude of electromyography (EMG) signals from different muscle sites may vary due to individual differences, muscle size, electrode placement, and other factors, resulting in differences in amplitude range. Normalization transforms the amplitude range of all EMG signals to a uniform interval, for example, to the range of -1 to 1. After sampling frequency standardization and amplitude range normalization, a standardized EMG signal is obtained.

[0019] Step S122: Extract the electrophysiological response signals corresponding to each stimulus type from the evoked potential data set, perform amplitude range normalization processing on the electrophysiological response signals under each stimulus type, and convert evoked potential signals with different amplitude ranges into evoked potential standardized signals with a uniform amplitude range.

[0020] For the subject's evoked potential data set, electrophysiological response signals corresponding to visual, auditory, and somatosensory stimuli were extracted. Since the amplitude range of evoked potential signals may differ between different stimulus types—for example, the amplitude of visual evoked potential signals may be relatively small, while the amplitude of somatosensory evoked potential signals may be larger—amplitude range normalization of the electrophysiological response signals for each stimulus type is necessary. The maximum-minimum normalization method is also used to obtain standardized evoked potential signals.

[0021] Step S123: Analyze the noise type and noise intensity of the electromyography (EMG) standardized signal, determine the noise suppression parameters of the EMG signal, and perform adaptive noise suppression processing on the EMG standardized signal based on the noise suppression parameters of the EMG signal to obtain the EMG denoised signal.

[0022] Noise analysis is performed on standardized electromyography (EMG) signals. Common noise types include power frequency noise (such as 50Hz or 60Hz AC power interference), EMG artifacts (interference from electrical activity in other muscles), and motion artifacts (noise caused by the subject's body movement). Spectral analysis is used to observe the power spectral density distribution of the standardized EMG signal, identifying the noise frequency components and thus determining the noise type. Simultaneously, the proportion of noise in the signal is calculated, or the noise intensity is determined through signal-to-noise ratio estimation.

[0023] Based on the noise type and intensity obtained from the analysis, the noise suppression parameters of the electromyography (EMG) signal are determined. For power frequency noise, a notch filter can be used for suppression, and the noise suppression parameters include the center frequency and bandwidth of the notch filter. For broadband noise such as EMG artifacts and motion artifacts, adaptive filtering algorithms, such as the least mean square adaptive filter, can be used. In this case, the noise suppression parameters include the filter order and convergence factor.

[0024] Based on the determined noise suppression parameters of the electromyography (EMG) signal, adaptive noise suppression processing is performed on the standardized EMG signal. For example, for power frequency noise, the center frequency of the notch filter is set to the noise frequency, and the bandwidth is adjusted to cover the noise frequency range for signal filtering. For other types of noise, the reference noise signal of the adaptive filter is set to the noise component separated from the standardized EMG signal, and the filter coefficients are continuously adjusted through an adaptive algorithm to achieve the best noise suppression effect. After processing, a denoised EMG signal is obtained, in which the noise component is effectively suppressed while retaining the useful information of the EMG signal.

[0025] Step S124: Analyze the noise type and noise intensity of the evoked potential normalized signal, determine the noise suppression parameters of the evoked potential signal, and perform adaptive noise suppression processing on the evoked potential normalized signal based on the noise suppression parameters of the evoked potential signal to obtain the evoked potential denoised signal.

[0026] Similarly, noise analysis is performed on the evoked potential (EP) standardized signals. EP signals are typically weak and easily affected by various types of noise, including environmental noise, amplifier noise, and EEG background noise. By performing time-domain and frequency-domain analysis on the EP standardized signals, the noise type is identified. For example, environmental noise may have significant power spectral peaks within a specific frequency range, amplifier noise manifests as broadband noise, and EEG background noise is related to the subject's brain activity.

[0027] The noise suppression parameters for evoked potential signals are determined based on the noise type and intensity. Common noise suppression methods for evoked potential signals include averaging and stacking techniques, and wavelet transform denoising. If averaging is used, the noise suppression parameters include the number of stacks, which is determined based on the noise intensity; the stronger the noise, the more stacks are needed. If wavelet transform denoising is used, the noise suppression parameters include the selection of the wavelet basis function, the number of decomposition levels, the threshold function, and the threshold value.

[0028] Based on the determined noise suppression parameters of the evoked potential signal, adaptive noise suppression processing is performed on the evoked potential normalized signal. Taking the average superposition technique as an example, the evoked potential signals recorded under multiple stimuli are superimposed and averaged. Since the noise is random, the noise will cancel each other out during the superposition process, while the evoked potential signal, as a deterministic signal, will be enhanced, thereby improving the signal-to-noise ratio and obtaining the evoked potential denoised signal.

[0029] Step S125: Based on the signal bandwidth, amplitude fluctuation range, and signal duration of the electromyography noise reduction signal, and combined with the signal bandwidth, amplitude fluctuation range, and signal duration of the evoked potential noise reduction signal, a data type adaptation model is constructed. This data type adaptation model includes signal parameter conversion formulas and data type difference compensation coefficients.

[0030] Step S1251: Statistically analyze the signal bandwidth of the electromyography (EMG) noise-reduced signal under different activity states, calculate the average value and standard deviation of the signal bandwidth under all activity states, and obtain the EMG signal bandwidth characteristics.

[0031] For the electromyography (EMG) noise-reduced signals of the aforementioned subjects, the signal bandwidth was statistically analyzed in the resting, voluntary, and forced contraction states. The signal bandwidth was calculated by performing spectral analysis on the EMG noise-reduced signals to determine the frequency range where the signal energy was mainly concentrated; the difference between the upper and lower limits of this frequency range was the signal bandwidth. For each activity state, the signal bandwidth of multiple samples was calculated, and then the average and standard deviation of the signal bandwidth across all activity states were calculated. For example, the signal bandwidth of the EMG noise-reduced signal in the resting state may be relatively narrow, and the signal bandwidth in the voluntary and forced contraction states may vary. By statistically processing the signal bandwidth in these different activity states, the bandwidth characteristics of the EMG signal were obtained, which reflect the overall bandwidth characteristics of the EMG noise-reduced signal.

[0032] Step S1252: Statistically analyze the amplitude fluctuation range of the electromyography noise-reduced signal under different activity states, calculate the average value and standard deviation of the amplitude fluctuation range under all activity states, and obtain the amplitude characteristics of the electromyography signal.

[0033] Similarly, for the denoised electromyography (EMG) signal, its amplitude fluctuates within different ranges under different activity states. At rest, the amplitude fluctuation range is smaller; during voluntary and forceful contractions, the amplitude fluctuation range increases with the increase in muscle contraction intensity. By statistically analyzing the amplitude fluctuation range (i.e., the difference between the maximum and minimum signal values) of the denoised EMG signal for each activity state, and then calculating the average and standard deviation of the amplitude fluctuation range across all activity states, the amplitude characteristics of the EMG signal are obtained. These amplitude characteristics reflect the variation characteristics of the amplitude of the denoised EMG signal.

[0034] Step S1253: Statistically analyze the signal duration of the electromyography (EMG) noise-reduced signal under different activity states, calculate the average and standard deviation of the signal duration under all activity states, and obtain the temporal characteristics of the EMG signal.

[0035] The duration of the denoised electromyography (EMG) signal varies under different activity states. The signal duration at rest is the time from the start to the end of recording, while the signal duration during voluntary and forced contractions is determined based on the duration of muscle contraction. By statistically analyzing the signal duration of the denoised EMG signal for each activity state, and then calculating the mean and standard deviation of the signal duration across all activity states, the temporal characteristics of the EMG signal are obtained. These temporal characteristics reflect the persistence of the denoised EMG signal over time.

[0036] Step S1254: Using the same statistical method, obtain the bandwidth characteristics, amplitude characteristics, and temporal characteristics of the evoked potential noise reduction signal under different stimulus types.

[0037] For evoked potential (EV) noise-reduced signals, the same statistical methods as for electromyography (EMG) noise-reduced signals were used. For three different stimulus types—visual, auditory, and somatosensory—the signal bandwidth, amplitude fluctuation range, and signal duration of the evoked potential noise-reduced signals were statistically analyzed for each stimulus type. For example, the signal bandwidth of visual and auditory evoked potential signals may differ due to variations in stimulus frequency and neural conduction pathways, and the amplitude fluctuation range and signal duration of somatosensory evoked potential signals also have their own characteristics. By calculating the mean and standard deviation of these parameters for different stimulus types, the bandwidth characteristics, amplitude characteristics, and temporal characteristics of the evoked potential signals were obtained.

[0038] Step S1255: Calculate the difference between the bandwidth characteristics of the electromyography signal and the bandwidth characteristics of the evoked potential signal. Based on the difference, construct a bandwidth conversion formula. The bandwidth conversion formula is a linear conversion. The conversion coefficient of the linear conversion is determined by the difference and historical adaptation data.

[0039] The bandwidth characteristics of electromyography (EMG) signals and evoked potential (EVP) signals are compared, and the difference between them is calculated. This difference can be represented by the difference between their average values. A bandwidth conversion formula is constructed based on this difference. Since the bandwidth characteristics of EMG and EVP signals may have a linear relationship, the bandwidth conversion formula adopts a linear conversion form. Determining the conversion coefficients for the linear conversion requires combining the difference value with historical adaptation data. Historical adaptation data refers to the data accumulated from past type-matching of EMG and EVP data from a large number of similar subjects. By analyzing the historical adaptation data, the relationship between the difference value and the conversion coefficients is identified, thereby determining the current conversion coefficients. This allows the bandwidth conversion formula to transform the bandwidth of the EMG signal into a form that matches the bandwidth characteristics of the EVP signal.

[0040] Step S1256: Calculate the difference between the amplitude characteristics of the electromyography signal and the amplitude characteristics of the evoked potential signal. Based on this difference, construct an amplitude conversion formula. The amplitude conversion formula is a nonlinear conversion. The conversion parameters of the nonlinear conversion are obtained by fitting the amplitude distribution curves of the two signals.

[0041] The difference between the amplitude characteristics of electromyography (EMG) signals and evoked potential (EVP) signals can also be expressed as the difference in average values. Since the amplitude characteristics of EMG and EVP signals may exhibit complex nonlinear relationships, the amplitude conversion formula employs nonlinear conversion methods, such as exponential, logarithmic, or polynomial conversions. To determine the conversion parameters for the nonlinear conversion, amplitude distribution curves of the denoised EMG and evoked potential signals are first plotted. Then, using curve fitting, a nonlinear function that best fits the relationship between the two amplitude distribution curves is found. The parameters of this nonlinear function are the conversion parameters, thus constructing the amplitude conversion formula.

[0042] Step S1257: Calculate the difference between the temporal characteristics of electromyography signals and the temporal characteristics of evoked potential signals, and construct a temporal conversion formula based on the difference. The temporal conversion formula includes a time offset correction term, which is determined by the acquisition delay time of the two signals.

[0043] The difference between the temporal characteristics of electromyography (EMG) signals and evoked potential (EVP) signals is mainly reflected in the difference in the average signal duration. When constructing the temporal conversion formula, in addition to considering the conversion of signal duration, the synchronization of the two signals in time also needs to be considered. Therefore, the temporal conversion formula includes a time offset correction term. The time offset correction term is determined based on the acquisition delay time of the two signals, which refers to the time difference between the application of the stimulus or the start of muscle activity and the recording of the signal. By measuring and analyzing the acquisition delay time of EMG and evoked potential signals, this time offset correction term is incorporated into the temporal conversion formula to achieve accurate adaptation of the two signals in the temporal dimension.

[0044] Step S1258: Integrate the bandwidth conversion formula, the amplitude conversion formula, and the timing conversion formula to form the signal parameter conversion formula of the data type adaptation model.

[0045] The bandwidth conversion formula, amplitude conversion formula, and timing conversion formula constructed above convert the parameters of electromyography (EMG) signals and evoked potential (EVP) signals from three dimensions: bandwidth, amplitude, and timing, respectively, making the two different types of signals comparable and compatible at the parameter level. During the integration process, it is necessary to ensure the coordination and consistency between the conversion formulas. For example, the application of the timing conversion formula will not affect the conversion results of the bandwidth and amplitude conversion formulas. The conversion order and method of each formula are determined according to the signal characteristics and adaptation requirements.

[0046] Step S1259: Analyze the noise residual difference between electromyography signal and evoked potential signal, calculate the noise power spectral density difference between the two signals in the same frequency range, and determine the noise compensation coefficient based on the noise power spectral density difference.

[0047] Despite noise reduction processing, both electromyography (EMG) and evoked potential (EVP) noise-reduced signals may still retain some noise, and the degree of noise residue differs between the two signals. Analyzing this difference in noise residue, the power spectral density of the two signals is estimated within the same frequency range, and the noise power spectral density difference is calculated. This difference reflects the difference in noise energy between the two signals at the same frequency point. A noise compensation coefficient is determined based on this noise power spectral density difference. The magnitude of the noise compensation coefficient is directly proportional to the noise power spectral density difference; that is, the larger the difference, the larger the noise compensation coefficient, in order to compensate for the characteristic differences between the two signals caused by the difference in noise residue.

[0048] Step S12510: Analyze the difference in signal attenuation characteristics between electromyography (EMG) signals and evoked potential (EVP) signals, calculate the difference in amplitude attenuation rate between the two signals at the same transmission distance, and determine the attenuation compensation coefficient based on this difference in amplitude attenuation rate.

[0049] Electromyography (EMG) signals and evoked potential (EPP) signals experience signal attenuation during transmission due to factors such as the transmission medium and distance, and their attenuation characteristics may differ. To analyze the difference in signal attenuation characteristics, the amplitude attenuation of EMG and EEP signals is measured separately over the same transmission distance, and the difference in amplitude attenuation rates is calculated. The amplitude attenuation rate is the logarithmic form of the ratio of the signal amplitude before and after transmission. Based on this difference in amplitude attenuation rates, an attenuation compensation coefficient is determined. This coefficient is used to adjust for the amplitude difference caused by the different attenuation characteristics of the two signals, resulting in a better match in amplitude characteristics after compensation.

[0050] Step S12511: Analyze the difference in signal response speed between electromyography (EMG) signals and evoked potential (EVP) signals, calculate the difference in response delay time between the two signals under the same stimulus intensity, and determine the response compensation coefficient based on this difference in response delay time.

[0051] Electromyography (EMG) signals are electrical signals generated by muscle activity, while evoked potential (EPP) signals are responses of the nervous system to stimuli. The response speeds of these two signals differ. Under the same stimulus intensity, the time from stimulus application to a noticeable response in both EMG and EEPP signals is measured—the response delay time—and the difference in response delay time is calculated. Based on this difference, a response compensation coefficient is determined. This coefficient adjusts for the difference in time response between the two signals, ensuring accurate mapping of signal characteristics under the same physiological state during dynamic correlation mapping.

[0052] Step S12512: Integrate the noise compensation coefficient, the attenuation compensation coefficient, and the response compensation coefficient to form the data type difference compensation coefficient of the data type adaptation model. Each compensation coefficient contains sub-coefficients for different signal states. Electromyography signal states are divided according to activity state, and evoked potential signal states are divided according to stimulus type.

[0053] For example, step S125121: Divide the activity state of the electromyography signal into three categories: resting state, voluntary contraction state, and forced contraction state, and assign a unique electromyography state identifier to each category.

[0054] The activity states of electromyography (EMG) signals are clearly divided into three categories: resting state, voluntary contraction state, and forced contraction state. Each category is assigned a unique EMG state identifier; for example, the resting state is identified as J, the voluntary contraction state as S, and the forced contraction state as Y. This facilitates the differentiation and identification of EMG signals in different activity states during subsequent processing.

[0055] Step S125122: Divide the evoked potential signals into three categories: visual stimulation, auditory stimulation, and somatosensory stimulation, and assign a unique evoked potential state identifier to each category.

[0056] Similarly, the stimulation types of evoked potential signals are divided into three categories: visual stimulation, auditory stimulation, and somatosensory stimulation, and each is assigned a unique evoked potential state label, such as V for visual stimulation, A for auditory stimulation, and Q for somatosensory stimulation.

[0057] Step S125123: For the noise compensation coefficient, for each activity state of the electromyography signal, collect the noise power spectral density difference between the electromyography signal and the evoked potential signal in the same frequency range under that activity state, and calculate the noise compensation sub-coefficient corresponding to each electromyography activity state based on the difference.

[0058] For the noise compensation coefficient, the noise power spectral density difference between the electromyography (EMG) signal and the evoked potential (EPP) signal within the same frequency range is collected for each of the three activity states: resting state, voluntary contraction state, and forced contraction state. For example, in the resting state, the noise power spectral density difference between the noise-reduced EMG signal and the noise-reduced EEP signal under visual stimulation within a certain frequency range is collected, and then the noise compensation coefficient corresponding to the resting state is calculated based on this difference. A similar method is used to calculate the noise compensation coefficients corresponding to the voluntary contraction state and the forced contraction state.

[0059] Step S125124: For each stimulus type of evoked potential signal, collect the noise power spectral density difference between the evoked potential signal and the electromyography signal in the same frequency range under that stimulus type, and calculate the noise compensator coefficient corresponding to each evoked potential stimulus type based on the difference.

[0060] For the three stimulus types of evoked potential signals—visual, auditory, and somatosensory—the noise power spectral density difference between the evoked potential signal and the electromyography (EMG) signal within the same frequency range was collected for each stimulus type. For example, under visual stimulation, the noise power spectral density difference between the denoised evoked potential signal and the denoised EMG signal at rest was collected within the same frequency range. Based on this difference, the noise compensator coefficient corresponding to the visual stimulus type was calculated, and so on, to obtain the noise compensator coefficients corresponding to other stimulus types.

[0061] Step S125: Integrate the noise compensator coefficients corresponding to all electromyography activity states and evoked potential stimulation types to form a noise compensation coefficient matrix. The rows of the noise compensation coefficient matrix represent the electromyography activity states, the columns of the noise compensation coefficient matrix represent the evoked potential stimulation types, and the elements of the noise compensation coefficient matrix are the noise compensator coefficients of the corresponding state combinations.

[0062] The noise compensator coefficients corresponding to each activity state of the electromyography (EMG) and each stimulus type of the evoked potential (EVP) are integrated. A matrix is ​​constructed with EMG activity states as rows and EVP stimulus types as columns. Each element in the matrix represents the noise compensator coefficient for the corresponding combination of EMG activity state and EVP stimulus type, forming a noise compensation coefficient matrix. For example, the element in row J and column V of the matrix represents the noise compensation coefficient for the combination of the resting EMG signal and the visual stimulus evoked potential signal.

[0063] Step S125126: For the attenuation compensation coefficient, for each activity state of the electromyography signal, collect the difference in amplitude attenuation rate between the electromyography signal and the evoked potential signal under the same transmission distance in that activity state, and calculate the attenuation compensation sub-coefficient corresponding to each electromyography activity state based on the difference.

[0064] Following a method similar to that used for noise compensator coefficients, for each activity state of the electromyography (EMG) signal, the difference in amplitude attenuation rate between the EMG signal and the evoked potential (EPP) signal over the same transmission distance is collected for the attenuation compensation coefficient. For example, in a voluntary contraction state, the amplitude attenuation rates of the EMG signal and the EEP signal over the same transmission distance are measured, and the difference between the two is calculated. Based on this difference, the attenuation compensator coefficient corresponding to the voluntary contraction state is determined.

[0065] Step S125127: For each stimulus type of evoked potential signal, collect the difference in amplitude attenuation rate between the evoked potential signal and the electromyography signal under the same transmission distance for that stimulus type, and calculate the attenuation compensator coefficient corresponding to each evoked potential stimulus type based on the difference.

[0066] For each stimulus type of evoked potential signal, the difference in amplitude attenuation rate between the evoked potential signal and the electromyography signal under that stimulus type over the same transmission distance is collected, and then the attenuation compensator coefficient corresponding to each evoked potential stimulus type is calculated. For example, under auditory stimulation, the difference in amplitude attenuation rate between the evoked potential signal and the electromyography signal over the same transmission distance is calculated to determine the attenuation compensator coefficient for that stimulus type.

[0067] Step S125128: Integrate the attenuation compensator coefficients corresponding to all electromyography activity states and evoked potential stimulation types to form an attenuation compensator coefficient matrix. The rows of the attenuation compensator coefficient matrix represent the electromyography activity states, the columns of the attenuation compensator coefficient matrix represent the evoked potential stimulation types, and the elements of the attenuation compensator coefficient matrix are the attenuation compensator coefficients of the corresponding state combinations.

[0068] The attenuation compensator coefficients corresponding to each activity state of electromyography and each stimulation type of evoked potential are integrated into an attenuation compensator coefficient matrix. The rows of the matrix represent the activity state of electromyography, the columns represent the stimulation type of evoked potential, and the matrix elements are the attenuation compensator coefficients of the corresponding combination.

[0069] Step S125129: For the response compensation coefficient, for each activity state of the electromyography signal, collect the response delay time difference between the electromyography signal and the evoked potential signal under the same stimulus intensity in that activity state, and calculate the response compensation coefficient corresponding to each electromyography activity state based on the difference.

[0070] For the response compensation coefficient, similarly for each activity state of the electromyography (EMG) signal, the difference in response delay time between the EMG signal and the evoked potential signal under the same stimulus intensity is collected for that activity state. For example, under forceful contraction, the response delay time of the EMG signal and the evoked potential signal under the same stimulus intensity is measured, the difference is calculated, and the response compensation coefficient corresponding to the forceful contraction state is obtained based on the difference.

[0071] Step S1251210: For each stimulus type of evoked potential signal, collect the difference in response delay time between the evoked potential signal and the electromyography signal under the same stimulus intensity for that stimulus type, and calculate the response compensator coefficient corresponding to each evoked potential stimulus type based on the difference.

[0072] For each stimulus type of evoked potential signal, the difference in response delay between the evoked potential signal and the electromyography (EMG) signal under the same stimulus intensity is collected, and the corresponding response compensator coefficient is calculated. For example, under somatosensory stimulation, the difference in response delay between the evoked potential signal and the EMG signal is determined, and then the response compensator coefficient for that stimulus type is obtained.

[0073] Step S1251211: Integrate the response compensator coefficients corresponding to all electromyographic activity states and evoked potential stimulation types to form a response compensator coefficient matrix. The rows of the response compensator coefficient matrix represent the electromyographic activity states, the columns of the response compensator coefficient matrix represent the evoked potential stimulation types, and the elements of the response compensator coefficient matrix are the response compensator coefficients of the corresponding state combinations.

[0074] The response compensator coefficients corresponding to each activity state of electromyography and each stimulus type of evoked potential are integrated into a response compensator coefficient matrix. The matrix structure is the same as the noise compensation coefficient matrix and the attenuation compensation coefficient matrix. The rows correspond to the activity state of electromyography, the columns correspond to the stimulus type of evoked potential, and the elements are the response compensator coefficients of the corresponding combination.

[0075] Step S1251212: Integrate the noise compensation coefficient matrix, the attenuation compensation coefficient matrix, and the response compensation coefficient matrix to form the data type difference compensation coefficient of the data type adaptation model. The data type difference compensation coefficient contains three sub-matrices, which correspond to the difference compensation of the three dimensions of noise, attenuation, and response, respectively. Each sub-matrix covers all combinations of electromyographic activity states and evoked potential stimulation types.

[0076] Finally, the noise compensation coefficient matrix, attenuation compensation coefficient matrix, and response compensation coefficient matrix are integrated together to form the data type difference compensation coefficient of the data type adaptation model. This data type difference compensation coefficient contains three sub-matrices, which compensate for the differences between electromyography (EMG) signals and evoked potential (EPP) signals from the three dimensions of noise, attenuation, and response, respectively. Each sub-matrix covers all combinations of EMG activity states and EEPP stimulation types, ensuring accurate difference compensation under different state combinations.

[0077] Step S126: Input the electromyography (EMG) noise-reduced signal into the data type adaptation model, adjust the signal dimension of the EMG noise-reduced signal through the signal parameter conversion formula, and then compensate for the signal characteristic differences between EMG and evoked potentials through the data type difference compensation coefficient to obtain the EMG adaptation signal.

[0078] The denoised electromyography (EMG) signal is input into the constructed data type adaptation model. First, the signal parameter transformation formula in the model is applied to transform the signal bandwidth, amplitude fluctuation range, and signal duration of the denoised EMG signal, adjusting its signal dimensions to achieve compatibility with evoked potential (EPP) signals at the parameter level. Then, based on the data type difference compensation coefficient, the denoised EMG signal is compensated for differences in three dimensions: noise, attenuation, and response. Specifically, based on the activity state corresponding to the denoised EMG signal and the stimulation type of the evoked potential signal to be adapted, corresponding compensation coefficients are selected from the three sub-matrices of the data type difference compensation coefficient to compensate for noise residue, amplitude attenuation, and response speed of the denoised EMG signal, respectively, ultimately obtaining the adapted EMG signal. This adapted EMG signal is closer to the EEPP signal in signal characteristics, facilitating subsequent feature correlation analysis.

[0079] Step S127: Input the evoked potential noise reduction signal into the data type adaptation model, adjust the signal dimension of the evoked potential noise reduction signal through the signal parameter conversion formula, and then compensate for the signal characteristic difference between evoked potential and electromyography through the data type difference compensation coefficient to obtain the evoked potential adaptation signal.

[0080] Similar to electromyography (EMG) signal denoising, the evoked potential (EPP) denoised signal is input into a data type adaptation model. For example, firstly, signal dimensions such as bandwidth, amplitude fluctuation range, and signal duration of the EPP denoised signal are adjusted using signal parameter conversion formulas to match the dimensions of the EMG signal. Then, based on the data type difference compensation coefficients, corresponding compensation coefficients are selected from three sub-matrices to compensate for the noise residue, amplitude attenuation, and response speed of the EPP denoised signal, based on the stimulus type corresponding to the EPP denoised signal and the activity state of the adapted EMG signal. This results in an adapted EPP signal.

[0081] Step S128: Extract the discharge frequency features, discharge amplitude features, and discharge interval features from the electromyography (EMG) adaptation signal to form an EMG adaptation feature subset; extract the latency features, amplitude features, and waveform morphology features from the evoked potential (EPP) adaptation signal to form an EPP adaptation feature subset.

[0082] For electromyography (EMG) adaptation signals, features are extracted to form an EMG adaptation feature subset. The discharge frequency feature refers to the number of motor unit potentials fired per unit time in the EMG signal, determined by performing time-domain analysis on the EMG adaptation signal and counting the number of discharges within a set time period. The discharge amplitude feature refers to the magnitude of the motor unit potential, which can be obtained by measuring the peak value or peak-to-peak value of the signal. The discharge interval feature refers to the time interval between two adjacent discharges, obtained by calculating the difference between adjacent discharge times. These extracted features are combined to form the EMG adaptation feature subset, which contains characteristic information of the EMG adaptation signal in terms of discharge frequency, amplitude, and interval.

[0083] For evoked potential (EP) adaptation signals, latency, amplitude, and waveform morphology features are extracted to form an EP adaptation feature subset. Latency refers to the time from stimulus application to the appearance of a specific peak in the EP waveform; amplitude refers to the amplitude difference between a specific peak and trough in the EP waveform; waveform morphology features are obtained by analyzing the shape of the EP waveform, including the rising slope, falling slope, and the presence of specific waveform components. Integrating these features constitutes the EP adaptation feature subset.

[0084] Step S129: Perform time axis alignment processing on the electromyography adaptation feature subset and the evoked potential adaptation feature subset to determine the corresponding electromyography adaptation features and evoked potential adaptation features within the same time interval.

[0085] Since electromyography (EMG) and evoked potential (EPP) adaptation signals are acquired under different activity states and stimulus types, time axis alignment is required to establish their dynamic correlation. First, a common time reference frame is determined, such as using the moment of stimulus application or the start of muscle activity as the time zero point. Then, the features in the EMG and EEPP feature subsets are arranged chronologically, and their time coordinates are adjusted according to the time reference frame. By comparing the time coordinates of both, the corresponding EMG and EEPP adaptation features within the same time interval are determined. For example, within a certain time interval, the discharge frequency feature in the EMG feature subset corresponds to the latency feature in the EEPP feature subset, thus establishing a temporal correspondence.

[0086] Step S1210: Calculate the correlation coefficient between each pair of corresponding features, determine the feature association weights based on the correlation coefficients, and construct the feature association mapping matrix.

[0087] For each pair of corresponding EMG-matching features and evoked potential-matching features determined after time-axis alignment, the correlation coefficient between them is calculated. The correlation coefficient measures the degree of linear correlation between two features, and its value ranges from -1 to 1. The correlation coefficient can be calculated using methods such as Pearson correlation coefficient or Spearman rank correlation coefficient, with the appropriate calculation method chosen based on the type and distribution characteristics of the feature data. Based on the calculated correlation coefficient, the feature association weights are determined. The larger the absolute value of the correlation coefficient, the higher the degree of correlation between the two features, and the greater the corresponding feature association weight. The feature association weights of all feature pairs are arranged in a predetermined order to construct a matrix. The rows of the matrix represent EMG-matching features, the columns represent evoked potential-matching features, and the elements in the matrix are the feature association weights of the corresponding feature pairs, forming a feature association mapping matrix.

[0088] Step S1211: Calculate the fluctuation range of each element in the feature association mapping matrix. If the fluctuation range exceeds the preset range, adjust the difference compensation coefficient of the data type adaptation model and regenerate the feature association mapping matrix until the fluctuation range meets the preset range.

[0089] The fluctuation amplitude of each element in the feature association mapping matrix is ​​calculated. The fluctuation amplitude can be measured by calculating the standard deviation or variance of the matrix elements, reflecting the dispersion of the feature association weights. The calculated fluctuation amplitude is compared with a preset range, which is a reasonable fluctuation interval determined based on historical data and clinical experience. If the fluctuation amplitude exceeds the preset range, it indicates that the difference compensation coefficient of the current data type adaptation model may not be accurate enough, resulting in poor stability of the feature association weights. In this case, it is necessary to adjust the data type difference compensation coefficient of the data type adaptation model, for example, by fine-tuning relevant elements in the noise compensation coefficient matrix, attenuation compensation coefficient matrix, or response compensation coefficient matrix. After adjustment, the electromyography denoised signal and evoked potential denoised signal are re-adapted, features are extracted, and a new feature association mapping matrix is ​​constructed. The fluctuation amplitude of each element is calculated again, and this process is repeated until the fluctuation amplitude of each element in the feature association mapping matrix meets the preset range, ensuring that the feature association weights have good stability.

[0090] Step S1212: Based on the stabilized feature association mapping matrix, the electromyography-adapted feature subset and the evoked potential-adapted feature subset are recombined to generate a cross-data type association feature set containing the association stability index of each pair of associated features. The association stability index is calculated by the fluctuation range of the feature association weight and the correlation coefficient.

[0091] After obtaining the stabilized feature association mapping matrix, the electromyography (EMG) and evoked potential (EVP) feature subsets are recombined based on this matrix. According to the feature association weights of each element in the feature association mapping matrix, feature pairs with high correlation between the EMG and EVP features are combined to form cross-data type associated features. For each pair of associated features, its association stability index is calculated. The calculation of the association stability index comprehensively considers the fluctuation range of the feature association weights and the correlation coefficient. Specifically, the correlation coefficient is first standardized to ensure it is on the same dimension as the fluctuation range. Then, a predetermined weighting method is used to combine the two. For example, the association stability index can be expressed as the product of the standardized correlation coefficient and (1 minus the standardized fluctuation range). This considers both the strength and stability of the association between features. Each pair of associated features and its association stability index are integrated to generate a cross-data type associated feature set.

[0092] Step S130: Call the pre-trained multi-round iterative intelligent diagnostic reasoning model to perform disease association analysis on the cross-data type association feature set, and generate a preliminary diagnostic conclusion containing multiple suspected diseases and their multi-dimensional association descriptions.

[0093] Step S131: Input the cross-data type associated feature set into the feature preprocessing layer of the multi-round iterative intelligent diagnostic reasoning model, perform outlier detection on each associated feature in the cross-data type associated feature set, and if an outlier exists, replace the outlier based on the historical normal data distribution of the associated feature to obtain the preprocessed cross-data type associated feature set.

[0094] The cross-data type association feature set obtained above is input into the feature preprocessing layer of the multi-round iterative intelligent diagnostic inference model. This multi-round iterative intelligent diagnostic inference model is pre-trained and capable of performing disease association analysis on electrophysiological features. In the feature preprocessing layer, outlier detection is first performed on each associated feature in the cross-data type association feature set. Outlier detection can employ statistical methods, such as the Z-score method or the quartile method. For each associated feature, a normal range is determined based on its historical normal data distribution. If a feature value exceeds this normal range, it is determined to be an outlier. When an outlier is detected, outlier replacement is required. The replacement method is based on the historical normal data distribution of the associated feature; for example, the mean, median, or predicted value generated by a distribution model can be used to replace the outlier. After outlier detection and replacement processing, a preprocessed cross-data type association feature set is obtained. The feature values ​​in this cross-data type association feature set are more consistent with the normal data distribution, reducing the impact of outliers on subsequent diagnostic analysis.

[0095] Step S132: Perform feature dimension compression on the preprocessed cross-data type associated feature set, and use a feature selection algorithm to filter associated features that contribute more than a preset threshold to disease identification, forming a core cross-data type associated feature set.

[0096] The preprocessed cross-data type associated feature set may contain many feature dimensions. To improve the efficiency and accuracy of diagnostic reasoning, feature dimension compression is necessary. Feature selection algorithms, such as ReliefF, information gain, or variance inflation factor, are used to calculate the contribution of each associated feature to disease identification. The contribution is calculated based on the correlation between the feature and the disease or the classification information contained in the feature. A preset threshold is set, and associated features with a contribution exceeding the threshold are selected and combined to form the core cross-data type associated feature set.

[0097] Step S133: Input the core cross-data type association feature set into the initial diagnosis layer of the multi-round iterative intelligent diagnosis reasoning model, and call the disease feature template library built into the multi-round iterative intelligent diagnosis reasoning model. The disease feature template library contains typical cross-data type association feature templates corresponding to various preset diseases, and each typical cross-data type association feature template is labeled with feature weight distribution.

[0098] The core set of cross-data type association features is input into the initial diagnostic layer of the multi-round iterative intelligent diagnostic reasoning model. The initial diagnostic layer is responsible for preliminary symptom matching. In this layer, the model's built-in symptom feature template library is invoked. This library stores typical cross-data type association feature templates corresponding to various preset symptom (such as peripheral neuropathy, myopathy, motor neuron disease, etc.). Each typical cross-data type association feature template is derived from a large amount of clinical case data, encompassing the typical performance of that symptom in cross-data type association features. Each template is also labeled with a feature weight distribution, which indicates the importance of each associated feature in the template for the diagnosis of that symptom.

[0099] Step S134: Calculate the similarity between the core cross-data type association feature set and each typical cross-data type association feature template. The similarity calculation between the core cross-data type association feature set and each typical cross-data type association feature template is performed from three dimensions: feature numerical matching degree, association stability index matching degree, and feature temporal change trend matching degree.

[0100] Step S1341: For each associated feature in the core cross-data type associated feature set, extract the feature value of the associated feature, and at the same time extract the feature value range of the same type of associated feature in the corresponding typical cross-data type associated feature template.

[0101] For each associated feature in the core cross-data type association feature set, its feature value is extracted one by one. For example, a certain association feature is "electromyography discharge frequency-evoked potential latency association feature", and its feature value is the specific value after processing. At the same time, the feature value range of the same type of association feature is extracted from the typical cross-data type association feature template. This feature value range is the normal or typical value range of the disease for this association feature.

[0102] Step S1342: Calculate the degree of overlap between the feature value of the associated feature and the feature value range. If the feature value of the associated feature is within the feature value range, the degree of overlap is a specific value. If the feature value of the associated feature is outside the feature value range, calculate the distance between the feature value of the associated feature and the boundary of the feature value range. Calculate the degree of overlap based on this distance to obtain the numerical matching degree of the associated feature.

[0103] For each associated feature, the degree of overlap between its feature value and the corresponding feature value range in the template is calculated. If the feature value falls within the range, the overlap is 1 (or another specific value indicating complete overlap); if the feature value is outside the range, the distance between the feature value and the range boundary is calculated. The closer the distance, the higher the overlap; the farther the distance, the lower the overlap. In other words, the overlap is inversely proportional to the distance. The numerical matching degree of each associated feature is calculated in this way, reflecting the degree of matching between the feature value and the template value range.

[0104] Step S1343: Perform a weighted average of the numerical matching degree of all associated features in the core cross-data type associated feature set. The weighting coefficient of this weighted average is the feature weight of each associated feature in the typical cross-data type associated feature template, thus obtaining the feature numerical matching degree.

[0105] The numerical matching degree of all associated features in the core cross-data type associated feature set is weighted and averaged. The weighting coefficient of the weighted average is the feature weight of each associated feature in the typical cross-data type associated feature template. The higher the feature weight of the associated feature, the greater the proportion of its numerical matching degree in the weighted average. The feature numerical matching degree is obtained by calculating the feature numerical matching degree through weighted averaging. This feature numerical matching degree comprehensively reflects the overall matching degree between the core cross-data type associated feature set and the typical template at the feature numerical level.

[0106] Step S1344: Extract the association stability index of each association feature in the core cross-data type association feature set, and at the same time extract the association stability index range of the same type of association features in the corresponding typical cross-data type association feature template.

[0107] Similar to the calculation of feature numerical matching degree, the association stability index of each associated feature in the core cross-data type association feature set is extracted. This association stability index was determined in the previous process of generating the cross-data type association feature set. At the same time, the association stability index range of the same type of associated features is extracted from the typical cross-data type association feature template. This association stability index range is the normal or typical range of the association stability index of the associated features corresponding to the disease.

[0108] Step S1345: Calculate the deviation rate between the correlation stability index of the correlation feature and the index range. The deviation rate is the difference between the actual index and the midpoint of the index range divided by the absolute value of the half-width of the index range.

[0109] Calculate the deviation rate between the association stability index of each associated feature and the index range in the template. The formula for calculating the deviation rate is: Deviation rate = |(Actual index - Midpoint of index range)| / Half width of index range. Where the midpoint of the index range is the average of the upper and lower limits of the index range, and the half width of the index range is (Upper limit - Lower limit) / 2. The deviation rate reflects the degree of deviation between the actual association stability index and the index range in the template.

[0110] Step S1346: Calculate the stable matching sub-degree based on the deviation rate. The calculation of the stable matching sub-degree adopts a preset mapping relationship between the deviation rate and the sub-degree.

[0111] Based on a pre-defined mapping relationship between deviation rate and stable match degree, the calculated deviation rate is converted into a stable match degree. This mapping relationship is usually pre-defined; the smaller the deviation rate, the higher the stable match degree; the larger the deviation rate, the lower the stable match degree. For example, when the deviation rate is 0, the stable match degree is 1; as the deviation rate increases, the stable match degree gradually decreases.

[0112] Step S1347: Calculate the weighted average of the stable matching degree of all associated features in the core cross-data type associated feature set. The weighting coefficient of the weighted average is the feature weight of each associated feature in the typical cross-data type associated feature template, and obtain the association stability index matching degree.

[0113] A weighted average of the stability matching degrees of all associated features is calculated, with the weighting coefficients also representing the feature weights of the associated features in the typical template. The weighted average yields the association stability index matching degree, which reflects the degree of matching between the core cross-data type associated feature set and the typical template in terms of association stability.

[0114] Step S1348: For each associated feature in the core cross-data type associated feature set, arrange the feature values ​​of the associated feature in chronological order to form a time sequence of feature values.

[0115] For each associated feature in the core cross-data type associated feature set, its feature values ​​at different time points are arranged in chronological order to form a time-series sequence of feature values. This sequence reflects the changes of the associated feature over time.

[0116] Step S1349: Extract the time-series change trend curve of the feature values ​​of the same type of related features in the corresponding typical cross-data type related feature template. The time-series change trend curve of the feature values ​​includes the division of rising segment, steady segment, and falling segment and the range of change rate of each segment.

[0117] This study extracts the time-series trend curves of characteristic values ​​for the same type of associated features from typical cross-data type association feature templates. These trend curves are derived from historical data of the disease and describe typical time-series change patterns of associated features during disease development or under different physiological states. The curves include rising, stable, and falling segments, as well as the range of change rates for each segment. The range of change rates indicates the degree to which the characteristic values ​​rise or fall at the corresponding stage.

[0118] Step S13410: Use the dynamic time warping algorithm to align the feature numerical time series sequence with the feature numerical time series change trend curve, and calculate the Euclidean distance between the aligned sequence and the curve.

[0119] To compare the similarity between the characteristic numerical time series and the characteristic numerical time series trend curve, a dynamic time warping algorithm is used to align them. The dynamic time warping algorithm can handle cases with different sequence lengths or inconsistent time axes by finding the optimal time alignment path to match the two sequences. After alignment, the Euclidean distance between the aligned characteristic numerical time series and the characteristic numerical time series trend curve is calculated. Euclidean distance is a commonly used metric to measure the difference between corresponding points in two sequences; the smaller the distance, the more similar the two sequences are.

[0120] Step S13411: Calculate the trend matching sub-degree based on Euclidean distance. The calculation of the trend matching sub-degree adopts a preset mapping relationship between distance and sub-degree.

[0121] Based on a predefined mapping relationship between Euclidean distance and trend matching sub-degree, the calculated Euclidean distance is converted into trend matching sub-degree. This mapping relationship ensures that the smaller the Euclidean distance, the higher the trend matching sub-degree; and the larger the Euclidean distance, the lower the trend matching sub-degree. For example, when the Euclidean distance is 0, the trend matching sub-degree is 1; as the distance increases, the sub-degree gradually decreases.

[0122] Step S13412: Calculate the weighted average of the trend matching degree of all related features in the core cross-data type related feature set. The weighting coefficient of the weighted average is the feature weight of each related feature in the typical cross-data type related feature template, and obtain the feature time series change trend matching degree.

[0123] The trend matching degree of all associated features is weighted and averaged, with the weighting coefficient being the feature weight of the associated feature in the typical template. This yields the feature temporal change trend matching degree, which reflects the degree of matching between the core cross-data type associated feature set and the typical template in terms of temporal change trend.

[0124] Step S135: Based on the similarity calculation results of the three dimensions, the initial correlation degree of each preset disease is obtained by weighted summation. The weighting coefficient of this weighted summation method is obtained by training the multi-round iterative intelligent diagnostic reasoning model through historical diagnostic data.

[0125] After obtaining the similarity calculation results for the three dimensions—feature numerical matching degree, association stability index matching degree, and feature temporal change trend matching degree—an initial association degree for each preset disease is calculated using a weighted summation method. The formula for the weighted summation is: Initial Association Degree = Feature Numerical Matching Degree × Weight 1 + Association Stability Index Matching Degree × Weight 2 + Feature Temporal Change Trend Matching Degree × Weight 3. Here, Weight 1, Weight 2, and Weight 3 are the weighting coefficients for the three dimensions of similarity. These weighting coefficients are obtained by training a multi-round iterative intelligent diagnostic reasoning model on a large amount of historical diagnostic data, reflecting the importance of different dimensions of similarity in disease diagnosis. Through the above weighted summation method, the matching situation of the three dimensions is comprehensively considered to obtain the initial association degree for each preset disease. The higher the initial association degree, the more similar the core cross-data type association feature set is to the typical feature template of the disease.

[0126] Step S136: Filter the preset symptoms whose initial correlation exceeds the preset initial threshold to form an initial suspected symptom list. The initial suspected symptom list includes the initial correlation of each initial suspected symptom and the similarity details in three dimensions.

[0127] A preset initial threshold is set, determined based on clinical diagnostic experience and model training results, to determine the likelihood of a symptom being suspected. The initial correlation score of each preset symptom is compared with the preset initial threshold, and symptoms with an initial correlation score exceeding the threshold are selected; these symptoms are considered potential suspected symptoms. The selected symptoms are then combined to form an initial list of suspected symptoms. This list includes not only the initial correlation score value for each suspected symptom but also detailed similarity information across three dimensions: feature numerical matching degree, correlation stability index matching degree, and feature temporal change trend matching degree, for subsequent more detailed analysis and evaluation.

[0128] Step S137: Input the initial list of suspected symptoms into the symptom feature feedback module of the multi-round iterative intelligent diagnostic reasoning model. The symptom feature feedback module calls the symptom association rule library built into the multi-round iterative intelligent diagnostic reasoning model to extract the association feature constraints corresponding to each initial suspected symptom.

[0129] The initial list of suspected symptoms is input into the symptom feature feedback module of the multi-round iterative intelligent diagnostic reasoning model. The role of this module is to further check the constraints of the initial suspected symptoms. This module calls the model's built-in symptom association rule library, which stores the constraint relationships between various symptoms and associated features. For each symptom in the initial list of suspected symptoms, the corresponding association feature constraints are extracted from the symptom association rule library. Association feature constraints refer to specific conditions that the symptom must meet regarding certain associated features. For example, a certain symptom might require that the feature value of a specific associated feature must be greater than a certain value, or that the association stability index must be within a certain range, or that the temporal change trend of the feature must include specific stages, etc.

[0130] Step S138: Analyze whether the core cross-data type association feature set satisfies the association feature constraints of each initial suspected disease. If not, calculate the degree of non-satisfaction of the association feature constraints and reduce the initial association degree of the corresponding initial suspected disease based on the degree of non-satisfaction.

[0131] For each initial suspected symptom, the associated features in the core cross-data type associated feature set are compared and analyzed with the associated feature constraints corresponding to that symptom to determine whether the constraints are met. If an associated feature meets the constraints, it does not affect the initial association degree; if it does not meet the constraints, the degree of non-compliance needs to be calculated. The calculation of the degree of non-compliance is determined based on the type of constraint and the degree to which the feature value deviates from the constraint range. For example, for numerical constraints, the degree of non-compliance can be measured by calculating the deviation of the feature value from the constraint range. Based on the calculated degree of non-compliance, the initial association degree of the initial suspected symptom is adjusted to be reduced; the higher the degree of non-compliance, the greater the reduction in the initial association degree.

[0132] Step S139: If the initial correlation of the initial suspected disease is lower than the preset initial threshold after the initial correlation is reduced, the disease is removed from the list of initial suspected diseases; if the initial correlation of the initial suspected disease is still higher than the preset initial threshold after the initial correlation is reduced, the disease is retained in the list of initial suspected diseases, and the details of the non-compliance of the correlation feature constraints are recorded.

[0133] After adjusting the initial correlation of the initial suspected symptoms to lower the initial correlation, the adjusted correlation is compared again with the preset initial threshold. If the adjusted correlation is lower than the preset initial threshold, it indicates that the likelihood of the symptom being suspected is low, and the symptom is removed from the initial suspected symptom list; if the adjusted correlation is still higher than the preset initial threshold, the symptom is retained in the initial suspected symptom list. Simultaneously, for the retained symptom, a detailed record of any unmet correlation feature constraints is made, including the specific content of the unmet constraints, the corresponding correlation features, and the degree of non-compliance. This information is helpful for subsequent iterative optimization analysis.

[0134] Step S1310: Input the adjusted initial list of suspected symptoms into the iterative optimization layer of the multi-round iterative intelligent diagnostic reasoning model. Based on the satisfaction of the associated feature constraints of each suspected symptom, extract the feature subset related to the associated feature constraints from the core cross-data type associated feature set.

[0135] The initial list of suspected symptoms, after the aforementioned adjustments, is input into the iterative optimization layer of the multi-round iterative intelligent diagnostic reasoning model. The purpose of the iterative optimization layer is to further optimize the correlation of suspected symptoms through multiple rounds of iteration. For each suspected symptom in the list, based on the satisfaction of its correlation feature constraints, a subset of features related to the correlation feature constraints is extracted from the core cross-data type correlation feature set. This subset of features contains correlation features that play a key role in the satisfaction of the constraints for that symptom. Through in-depth analysis of this subset of features, the degree of suspicion of the symptom can be assessed more accurately.

[0136] Step S1311: Perform feature enhancement processing on the feature subset, increase the weight of key features in the feature subset, and generate an enhanced feature subset.

[0137] Feature enhancement processing is performed on the extracted feature subset. Key features refer to those that play a major role in the associated feature constraints or are valuable for disease diagnosis. By increasing the weight of these key features, they are given greater weight in subsequent similarity calculations, thus highlighting their impact on disease assessment. The magnitude of the increase in feature weights can be determined based on the importance of the features and the degree to which constraints are not met. For example, for key features that do not meet constraints, their weights can be appropriately increased to focus more on the matching of these features in subsequent iterations. After feature enhancement processing, an enhanced feature subset is generated.

[0138] Step S1312: Recalculate the similarity between the enhanced feature subset and the corresponding typical cross-data type association feature template of the suspected disease, update the association degree of each suspected disease, and obtain the iterative association degree.

[0139] Using the enhanced feature subset, the similarity between it and the typical cross-data type association feature template of the corresponding suspected disease is recalculated. The similarity calculation is still performed from three dimensions: feature numerical matching degree, association stability index matching degree, and feature temporal change trend matching degree, but the weights of key features in the feature subset have been enhanced. Based on the new similarity calculation results, the association degree of each suspected disease is updated according to the same weighted summation method as in step S135, resulting in the iterative association degree.

[0140] Step S1313: Repeat the operations of feature subset extraction, feature enhancement, similarity recalculation, and correlation update until the number of iterations reaches the preset number of iterations or the change in the correlation of suspected diseases is lower than the preset change threshold.

[0141] Set a preset number of iterations and a preset change threshold. The preset number of iterations is the maximum number of iterations set in advance to prevent the iteration process from running indefinitely; the preset change threshold is a standard to measure whether the change in correlation is small enough. Repeat steps S1310 to S1312, i.e., extract feature subsets, perform feature enhancement, recalculate similarity, and update correlation. After each iteration, check whether the number of iterations has reached the preset number of iterations, or whether the change in the correlation of suspected symptoms is lower than the preset change threshold. If either condition is met, stop the iteration process. Through multiple iterations, continuously optimize the weights of associated features and similarity calculations to make the correlation of suspected symptoms more accurate and stable.

[0142] Step S1314: For each suspected symptom after iteration, generate a multi-dimensional correlation description based on the matching degree of feature numerical values, the matching degree of correlation stability index, the matching degree of feature temporal change trend, and the satisfaction degree of constraint conditions. The multi-dimensional correlation description includes specific calculation results and level classification.

[0143] After the iteration process, a multi-dimensional correlation description is generated for each suspected symptom. This description includes feature numerical matching degree, correlation stability index matching degree, feature temporal trend matching degree, and satisfaction degree of newly added constraints. The constraint satisfaction degree is calculated based on the satisfaction of constraints on the associated features during the iteration process, reflecting the overall degree to which the associated features of the symptom satisfy the constraints. Each dimension's correlation description not only includes the specific calculated numerical results but also provides a grading system; for example, the matching degree is divided into high, medium, and low levels, with the grading criteria determined based on model training and clinical experience. Through this multi-dimensional correlation description, the correlation between suspected symptom and the core cross-data type association feature set can be comprehensively and in detail reflected.

[0144] Step S1315: Integrate all iterated suspected symptoms and their multi-dimensional correlation descriptions to generate a preliminary diagnostic conclusion.

[0145] All iteratively identified suspected symptoms and their corresponding multi-dimensional correlation descriptions are integrated to form a preliminary diagnostic conclusion. The preliminary diagnostic conclusion lists various suspected symptoms and details the correlation of each symptom across multiple dimensions.

[0146] Step S140: The preliminary diagnostic conclusion is fused with the patient's clinical reference information using a multi-source evidence chain to obtain an optimized diagnostic conclusion.

[0147] Step S141: Obtain the subject's clinical reference information, which includes the subject's medical history, physical examination information and other auxiliary examination information. The medical history includes the subject's history of neurological diseases, muscle diseases and medication history. The physical examination information includes the results of muscle strength test, nerve reflex test and muscle atrophy test. Other auxiliary examination information includes the results of blood biochemistry test and imaging test.

[0148] Clinical reference information was collected from the subjects suspected of having peripheral neuropathy. Medical history information was obtained by reviewing the subjects' medical records, including their history of neurological diseases (such as multiple sclerosis, Parkinson's disease, etc.), muscle diseases (such as muscular dystrophy, myasthenia gravis, etc.), and medication history (such as long-term use of certain drugs that may affect neuromuscular function). Physical examination information was obtained by medical staff through physical examination of the subjects, including muscle strength test results (usually assessed using muscle strength grading standards), nerve reflex test results (such as whether knee and ankle reflexes are normal, hyperactive, or weakened), and muscle atrophy test results (observing whether muscle volume has decreased, and the degree of symmetry). Other auxiliary examination information included blood biochemistry test results (such as creatine kinase, blood glucose, vitamin levels, etc.) and imaging examination results (such as changes in nerve or muscle structure shown by neuro-ultrasound, MRI, etc.).

[0149] Step S142: Perform text semantic analysis on the medical history record information to extract key information related to the nervous and muscular systems, forming key features of the medical history. The key features of the medical history include disease type, duration of onset, frequency of symptom manifestation, and description of the effects of medication.

[0150] Textual semantic analysis was performed on the collected medical history records. Natural language processing (NLP) techniques were used to segment, tag, and understand the text content of the medical history records. Key information related to the nervous and muscular systems was identified and extracted. For example, the disease type was extracted from past medical history to determine whether it was a nervous system disease or a muscular system disease; the duration of onset was extracted from the description of the onset of symptoms, i.e., the time span from the appearance of symptoms to the visit to the doctor; the frequency of symptom manifestation was extracted from the symptom description, such as whether the symptoms were persistent, intermittent, or occasional; and the description of the effects of medication was extracted from the medication history, i.e., whether the medication might affect nerve or muscle function and the nature and extent of the effect. The extracted key information was integrated to form the key features of the medical history.

[0151] Step S143: The physical examination information is structured and converted from non-standardized physical examination results into standardized feature descriptions to form key physical examination features. The key physical examination features include descriptions of muscle strength level, descriptions of nerve reflex response type, and descriptions of muscle atrophy degree.

[0152] Physical examination information is often presented in a non-standardized textual description format, requiring structuring. For example, a muscle strength test result might be described as "slightly weak left upper limb muscles," which needs to be converted into a standardized muscle strength grading description, such as "left upper limb muscle strength grade 4" (using a 0-5 grade muscle strength grading standard); a nerve reflex test result described as "hyperactive knee reflex" should be converted into a standardized description of the nerve reflex response type, such as "hyperactive knee reflex (+++)"; and a muscle atrophy test result described as "mild muscle atrophy in the right lower limb" should be converted into a standardized description of the degree of muscle atrophy, such as "mild muscle atrophy in the right lower limb." Through this structuring process, non-standardized physical examination results are transformed into unified and standardized feature representations, forming key features of the physical examination.

[0153] Step S144: Perform data format unification processing on the other auxiliary examination information, convert the results of different types of auxiliary examinations into directly comparable numerical or classification features to form key auxiliary examination features, which include descriptions of biochemical indicator numerical ranges and descriptions of abnormal imaging areas.

[0154] Other auxiliary examination information includes different types of test results, which require standardized data format processing. For blood biochemistry test results, the values ​​of each biochemical indicator are converted into numerical range descriptions. For example, "creatine kinase: 200 U / L" is converted into "creatine kinase value is within the normal range (reference range: 25-200 U / L)" or "creatine kinase value is higher than the normal range," etc. For imaging examination results, descriptive imaging findings are converted into categorical features, such as "neuroultrasound shows thickening of the median nerve" is converted into "abnormal imaging area exists in the median nerve," with a specific description of the location and characteristics of the abnormal area. Through standardized data format processing, different types of auxiliary examination results are converted into directly comparable numerical or categorical features, forming key features of auxiliary examinations.

[0155] Step S145: Integrate the key features of the medical history, the key features of the physical examination, and the key features of the auxiliary examination to form a set of key clinical features, and label the feature credibility of each key clinical feature.

[0156] Key features from medical history, physical examination, and auxiliary examinations are integrated to form a set of clinical key features. This set includes all key features extracted from clinical reference information. Simultaneously, the reliability of each clinical key feature is assigned. The determination of feature reliability is based on the source and reliability of the information. For example, key features from auxiliary examinations obtained through objective testing may have higher reliability, while the reliability of patient-reported medical history may be relatively lower. Alternatively, the reliability may be assessed based on the clarity and consistency of the information. Reliability can be expressed using a grading system, such as high, medium, and low.

[0157] Step S146: Extract the multi-dimensional correlation description of each suspected disease from the preliminary diagnostic conclusion, determine the core cross-data type correlation features corresponding to each suspected disease, and form a set of electrophysiological features of suspected diseases.

[0158] Multi-dimensional correlation descriptions of each suspected symptom are extracted from the preliminary diagnostic conclusions. Based on these descriptions, core cross-data type correlation features corresponding to each suspected symptom are determined. Core cross-data type correlation features refer to cross-data type correlation features that play a key role in the diagnosis of the symptom, and these features have high correlation and importance in the multi-dimensional correlation descriptions. Each suspected symptom and its corresponding core cross-data type correlation features are integrated to form a set of electrophysiological features for suspected symptom symptoms. This set of electrophysiological features establishes a correspondence between suspected symptom symptoms and electrophysiological features.

[0159] Step S147: Construct an evidence association network of electrophysiological features and clinical features. The nodes of the evidence association network of electrophysiological features and clinical features include each electrophysiological feature in the set of electrophysiological features of suspected diseases and each clinical feature in the set of key clinical features. The edges of the evidence association network of electrophysiological features and clinical features represent the association between two nodes. The association is calculated by the frequency of co-occurrence of the two features in historical diagnostic data.

[0160] Step S1471: Extract all electrophysiological features from the set of electrophysiological features of suspected diseases, assign a unique electrophysiological node identifier to each electrophysiological feature, and form a set of electrophysiological nodes.

[0161] Extract all electrophysiological features from the suspected disease electrophysiological feature set, ensuring that each electrophysiological feature is unique. Assign a unique electrophysiological node identifier to each extracted electrophysiological feature; the identifier can be letters, numbers, or a combination thereof, used to uniquely identify the electrophysiological feature in the network. Combine all identified electrophysiological features together to form an electrophysiological node set.

[0162] Step S1472: Extract all clinical features from the set of key clinical features, assign a unique clinical node identifier to each clinical feature, and form a set of clinical nodes.

[0163] Similarly, all clinical features are extracted from the set of key clinical features, and each clinical feature is unique. A unique clinical node identifier, similar to an electrophysiological node identifier, is assigned to each clinical feature to identify it within the network. All identified clinical features are then integrated to form a set of clinical nodes.

[0164] Step S1473: Integrate the electrophysiological node set and the clinical node set to form a node set of the evidence association network of electrophysiological features and clinical features. Each node includes a node type, feature description and corresponding suspected disease identifier. The node types are divided into two categories: electrophysiological and clinical.

[0165] The electrophysiological node set and the clinical node set are integrated to form a node set of an evidence association network between electrophysiological and clinical features. Each node, in addition to a unique identifier, includes a node type (electrophysiological or clinical), a feature description (a detailed explanation of the feature), and a corresponding suspected disease identifier (the suspected disease to which the feature belongs). For example, an electrophysiological node might be categorized as electrophysiological, its feature description as "electromyography discharge frequency-evoked potential latency association feature," and its corresponding suspected disease identifier as a code for peripheral neuropathy.

[0166] Step S1474: Obtain the historical diagnostic database built into the multi-round iterative intelligent diagnostic reasoning model. This historical diagnostic database contains electrophysiological characteristic data, clinical characteristic data, and final diagnostic results of multiple historical subjects.

[0167] The system utilizes the historical diagnostic database built into the multi-round iterative intelligent diagnostic reasoning model. This database stores a large amount of relevant data on historical examinees, including each examinee's electrophysiological characteristics (such as electromyography and evoked potential-related characteristics), clinical characteristics (such as medical history, physical examination, and auxiliary examination characteristics), and the final diagnostic results (confirmed symptoms).

[0168] Step S1475: For each suspected symptom, filter out historical records from the historical diagnosis database that contain the final diagnosis result of the suspected symptom, forming a subset of historical association records for the suspected symptom.

[0169] For each suspected symptom in the list, historical records from the historical diagnostic database that contain the symptom in the final diagnosis are filtered out. For example, for the suspected symptom "peripheral neuropathy," all historical patient records with a final diagnosis of peripheral neuropathy are filtered out; these records constitute a subset of historically relevant records for this suspected symptom. This filtering allows focus to be placed on historical data related to the current suspected symptom.

[0170] Step S1476: Extract the electrophysiological and clinical features from each record in the subset of historical records related to the suspected disease, and count the number of times each pair of electrophysiological and clinical features co-occurs.

[0171] For each historical record in the subset of historical association records for each suspected condition, electrophysiological and clinical features are extracted. Then, the frequency of co-occurrence of each pair of electrophysiological and clinical features is counted; that is, the number of times a certain electrophysiological feature and a certain clinical feature appear simultaneously in the same historical record. For example, the frequency of co-occurrence of the electrophysiological feature "abnormal electromyography discharge frequency" and the clinical feature "muscle weakness" in the subset of historical association records for peripheral neuropathy is counted.

[0172] Step S1477: Calculate the ratio of the number of times each pair of electrophysiological features and clinical features co-occurs to the total number of records in the subset of historical records associated with the suspected disease, to obtain the co-occurrence frequency of the pair of electrophysiological features and clinical features.

[0173] For each pair of electrophysiological and clinical features, the ratio of their co-occurrence frequency to the total number of records in the historical associated subset of the suspected case is calculated. This ratio is the co-occurrence frequency. The co-occurrence frequency reflects the probability that a pair of electrophysiological and clinical features will co-occur in the historical cases of the suspected case. For example, if the co-occurrence frequency is 50 and the total number of records in the historical associated subset is 100, then the co-occurrence frequency is 0.5.

[0174] Step S1478: If the co-occurrence frequency exceeds a preset frequency threshold, then establish an edge in the evidence association network of electrophysiological features and clinical features between the corresponding electrophysiological node and clinical node, and the weight of the edge in the evidence association network of electrophysiological features and clinical features is the co-occurrence frequency.

[0175] A preset frequency threshold is established, determined based on the statistical characteristics of historical data and clinical experience, to determine whether a significant correlation exists between electrophysiological and clinical features. If the co-occurrence frequency of a pair of electrophysiological and clinical features exceeds the preset frequency threshold, a correlation is considered to exist between the two features, and an edge is established between the corresponding electrophysiological and clinical nodes in the evidence association network. The weight of this edge represents the co-occurrence frequency of the pair of features; a higher weight indicates a stronger correlation between the two features.

[0176] Step S1479: If the co-occurrence frequency does not exceed the preset frequency threshold, then the edges of the evidence association network between electrophysiological features and clinical features are not established.

[0177] If the frequency of co-occurrence does not exceed the preset frequency threshold, it is considered that the correlation between the electrophysiological features and clinical features is not significant, and no edge is established between the corresponding nodes.

[0178] Step S14710: For all electrophysiological nodes and clinical nodes of the same suspected disease, repeat the operations of calculating the common occurrence frequency and establishing the edge of the evidence association network of electrophysiological features and clinical features to form a local evidence association network for the suspected disease.

[0179] For all combinations of electrophysiological and clinical nodes for the same suspected symptom, steps S1476 to S1479 are repeated, namely, calculating the common frequency of each pair of nodes, determining whether to establish an edge based on the frequency threshold, and finally forming a local evidence association network for the suspected symptom. This local evidence association network only includes the electrophysiological nodes, clinical nodes, and the association edges between them related to the suspected symptom.

[0180] Step S14711: Integrate the local evidence association networks of all suspected symptoms, remove duplicate nodes and edges of the evidence association networks of electrophysiological features and clinical features, and take the average weight of the edges of the evidence association networks of duplicate electrophysiological features and clinical features as the average weight in each local network to obtain the global evidence association network of electrophysiological features and clinical features.

[0181] The local evidence networks for all suspected symptoms are integrated. During the integration process, for nodes that appear repeatedly in different local networks (i.e., nodes with the same electrophysiological or clinical characteristics), only one node is retained; for edges that appear repeatedly (i.e., edges connecting the same pair of nodes), their weights are averaged to obtain the weight of the integrated edge. Through the above integration method, a global evidence network of electrophysiological and clinical characteristics containing all nodes and edges related to suspected symptoms is obtained.

[0182] Step S14712: For each edge in the evidence association network of global electrophysiological features and clinical features, label the suspected disease identifier corresponding to that edge.

[0183] Step S147121: Assign a unique disease identification code to each suspected disease. The disease identification code includes a disease category number and a disease sequence number. The disease category number is determined according to the system to which the disease belongs.

[0184] Each suspected symptom in the list of suspected symptom cases is assigned a unique symptom identification code. The symptom identification code consists of a symptom category number and a symptom sequence number. The symptom category number is based on the system to which the symptom belongs; for example, the category number for neurological symptom cases is 01, and the category number for muscular symptom cases is 02. The symptom sequence number is an ordered number within the same category. For example, if the sequence number for peripheral neuropathy in neurological symptom cases is 001, then its symptom identification code is 01001.

[0185] Step S147122: Traverse each edge in the evidence association network of global electrophysiological features and clinical features, and obtain the source of the edge's establishment.

[0186] Traverse each edge in the global evidence association network to determine the source of each edge, i.e., from which local evidence association networks of suspected diseases the edge was integrated.

[0187] Step S147123: If the edge comes from only a local evidence association network of a suspected disease, then add the disease identification code of the suspected disease to the attribute information of the edge, and mark it as a single source plus disease identification code.

[0188] If an edge originates from only one local evidence network associated with a suspected disease, then the disease identification code of that suspected disease is added to the edge's attribute information, and it is marked as a single source. For example, if the edge's attribute information is marked "Single Source: 01001", it means that the edge comes from the local network of the suspected disease with disease identification code 01001.

[0189] Step S147124: If the edge comes from a local evidence association network of multiple suspected diseases, add the disease identification codes of all relevant suspected diseases to the attribute information of the edge. Each disease identification code is separated by a delimiter and marked as multiple sources plus multiple disease identification codes.

[0190] If an edge originates from a local evidence network of multiple suspected symptoms, then the symptom identification codes of all relevant suspected symptoms should be added to the attribute information of that edge. These codes should be separated by a delimiter (such as a comma) and marked as originating from multiple sources. For example, annotated as "Multiple Sources: 01001, 01002" indicates that the edge originates from the local networks of two suspected symptoms with symptom identification codes 01001 and 01002.

[0191] Step S147125: For the labeled edges, further add the common occurrence frequency corresponding to each suspected disease from each source. The common occurrence frequency corresponds one-to-one with the disease identification code, forming a key-value pair format of disease identification code plus frequency, and add it to the attribute information of the edge.

[0192] After labeling the symptom identification codes, each edge is further augmented with the common frequency of each suspected symptom from each source. The common frequency corresponds one-to-one with the symptom identification code, represented in key-value pair format. For example, "01001:0.6,01002:0.5" indicates that the common frequency for symptom identification code 01001 is 0.6, and the common frequency for symptom identification code 01002 is 0.5. This information is added to the edge's attribute information, enriching the edge's attribute description.

[0193] Step S147126: After completing the annotation of all edges, generate an attribute list of the evidence association network of global electrophysiological features and clinical features. The attribute list includes the identifier of each edge, the connected node pairs, the weight, the disease identification code and the corresponding frequency.

[0194] After labeling all edges, an attribute list of the global evidence association network is generated. This attribute list records the relevant attributes of each edge in detail, including the edge's unique identifier, the connected electrophysiological and clinical node pairs, the edge's weight, the disease identification code, and the corresponding co-occurrence frequency.

[0195] Step S147127: Bind and store the list of attributes to the evidence association network of global electrophysiological and clinical features.

[0196] The generated attribute list is stored together with the evidence association network of global electrophysiological and clinical features so that the attribute information of the edges can be quickly queried and used in subsequent network analysis and applications.

[0197] Step S148: Assign initial weights to each node in the evidence association network of the electrophysiological feature and the clinical feature. The initial weights of the electrophysiological feature nodes are determined by the contribution of the electrophysiological feature in the multi-dimensional association description, and the initial weights of the clinical feature nodes are determined by the credibility of the clinical feature.

[0198] Initial weights are assigned to each node in the evidence association network of global electrophysiological and clinical features. For electrophysiological feature nodes, the initial weight is determined based on the contribution of the electrophysiological feature to the multi-dimensional association description; the higher the contribution, the greater the initial weight. For clinical feature nodes, the initial weight is determined based on the credibility of the clinical feature; the higher the credibility level, the greater the initial weight.

[0199] Step S149: Based on the correlation strength of the edges in the evidence association network of the electrophysiological feature and the clinical feature, the weight of each node in the evidence association network of the electrophysiological feature and the clinical feature is updated using the node weight propagation algorithm, so that the closely related nodes pass weights to each other until the weight change of all nodes in the evidence association network of the electrophysiological feature and the clinical feature is lower than the preset weight threshold.

[0200] A node weight propagation algorithm is employed, which updates node weights based on the strength of edge relationships (i.e., edge weights) within the network. Edges with stronger relationships have a greater impact on the weight transfer between connected nodes. During propagation, each node's weight is updated based on the weights of its connected nodes and the weights of its edges. This propagation process is iterated until the weight changes of all nodes in the network are below a preset weight threshold, at which point the node weights reach a stable state. Through weight propagation, the weights of closely related nodes reinforce each other, better reflecting the true degree of correlation between features.

[0201] Step S1410: For each suspected disease, calculate the final weight sum of the electrophysiological feature nodes corresponding to the suspected disease and the final weight sum of the clinical feature nodes related to the suspected disease. Use a weighted fusion method to obtain the clinical fusion correlation degree of the suspected disease. The weighting ratio of the weighted fusion method is determined by the disease type. For nervous system diseases, the weight of electrophysiological features is emphasized, and for muscular system diseases, the weight of clinical features is emphasized.

[0202] For each suspected symptom, the final sum of weights for its corresponding electrophysiological feature nodes and the final sum of weights for its related clinical feature nodes are calculated separately. The final sum of weights for electrophysiological feature nodes refers to the sum of the weights obtained after weight propagation for all electrophysiological feature nodes corresponding to the suspected symptom; the final sum of weights for related clinical feature nodes refers to the sum of the final weights of all clinical feature nodes related to the suspected symptom. Then, a weighted fusion method is used to calculate the clinical fusion correlation. The weighting ratio is determined according to the symptom type. For neurological symptom conditions, the weight of electrophysiological features is relatively high, for example, 0.6 for electrophysiological features and 0.4 for clinical features; for muscular symptom conditions, the weight of clinical features is relatively high, for example, 0.6 for clinical features and 0.4 for electrophysiological features. Through the above weighted fusion, the evidence from electrophysiological features and clinical features is combined to obtain the clinical fusion correlation for each suspected symptom.

[0203] Step S1411: Adjust the multi-dimensional correlation description of the corresponding suspected disease according to the clinical fusion correlation, improve the description level of suspected diseases with high clinical fusion correlation in the feature time-series change trend matching degree dimension, and reduce the description level of suspected diseases with low clinical fusion correlation in the feature time-series change trend matching degree dimension.

[0204] Based on the calculated clinical fusion correlation, the multi-dimensional correlation descriptions of corresponding suspected diseases are adjusted. Specifically, the adjustments primarily target the feature temporal change trend matching dimension. For suspected diseases with high clinical fusion correlation, it indicates a close integration of clinical and electrophysiological evidence, thus increasing the descriptive level of their feature temporal change trend matching. Conversely, for suspected diseases with low clinical fusion correlation, the descriptive level of their feature temporal change trend matching is decreased. These adjustments enable the multi-dimensional correlation descriptions to better reflect the overall evidence from both clinical and electrophysiological perspectives.

[0205] Step S1412: Screen suspected diseases with clinical fusion correlation exceeding the preset fusion threshold, and sort the screened suspected diseases from high to low according to clinical fusion correlation.

[0206] A preset fusion threshold is set to determine whether a suspected symptom has a high clinical fusion correlation. Suspected symptom whose clinical fusion correlation exceeds the threshold is selected, and these symptoms are sorted from high to low according to their clinical fusion correlation to form an ordered list of symptom.

[0207] Step S1413: Integrate the sorted suspected symptoms and their adjusted multi-dimensional correlation descriptions to obtain the optimized diagnostic conclusion.

[0208] The ranked suspected symptoms and their adjusted multi-dimensional correlation descriptions are integrated to form an optimized diagnostic conclusion. This optimized conclusion takes into account electrophysiological characteristics and clinical reference information more comprehensively, resulting in higher accuracy.

[0209] Step S150: Generate a structured intelligent diagnostic report based on the optimized diagnostic conclusion. The structured intelligent diagnostic report includes the electrophysiological evidence chain for disease determination, details of cross-data type feature association analysis, clinical feature matching results, and phased clinical suggestions. The content of each part of the structured intelligent diagnostic report can be viewed by jumping to related indexes.

[0210] A structured intelligent diagnostic report is generated based on the optimized diagnostic conclusions. The report consists of several parts: an electrophysiological evidence chain section detailing the electrophysiological features supporting each suspected symptom and their correlations; a cross-data type feature correlation analysis section explaining the dynamic correlation mapping process and results between electromyography and evoked potential features; a clinical feature matching results section showcasing the matching status between clinical reference information and suspected symptoms; and a phased clinical recommendation section providing clinical management suggestions for different stages based on the diagnostic conclusions, such as further examinations, treatment plans, and rehabilitation training recommendations. The various sections of the report can be navigated to via a linked index. For example, clicking on a suspected symptom will take you to the electrophysiological evidence chain and clinical feature matching results section for that symptom, allowing medical staff to quickly access relevant information.

[0211] In one exemplary embodiment, an intelligent diagnostic report generation system based on electromyography (EMG) and evoked potential (EVP) data is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, this intelligent diagnostic report generation system based on electromyography (EMG) and evoked potential (EVP) data includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and EMP are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the EMP. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The EMP interface is used for information exchange between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When executed by a processor, this computer program implements a method for generating intelligent diagnostic reports based on electromyography (EMG) and evoked potential (EVP) data. The display unit of this intelligent diagnostic report generation system, used to form a visually visible image, can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device for this intelligent diagnostic report generation system can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the system's casing, or an external keyboard, touchpad, or mouse.

[0212] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating intelligent diagnostic reports based on electromyography and evoked potential data, characterized in that, The method includes: Acquire the subject's electromyography (EMG) data set and evoked potential (EVP) data set. The EMG data set includes continuous electrophysiological signal records of different muscle parts of the subject in the resting state, voluntary contraction state, and forced contraction state. The evoked potential data set includes multi-channel electrophysiological response records of the subject's nervous system under the action of visual stimulation, auditory stimulation, and somatosensory stimulation. The electromyography (EMG) dataset and the evoked potential (EVP) dataset are processed with data type difference compensation and dynamic association mapping to establish a dynamic association relationship between EMG features and EVP features under the same physiological state, resulting in a cross-data type association feature set including an association stability index. The pre-trained multi-round iterative intelligent diagnostic reasoning model is invoked to perform disease association analysis on the cross-data type association feature set, and a preliminary diagnostic conclusion containing multiple suspected diseases and their multi-dimensional association descriptions is generated. The preliminary diagnostic conclusion is fused with the subject's clinical reference information through a multi-source evidence chain to obtain an optimized diagnostic conclusion. Based on the optimized diagnostic conclusions, a structured intelligent diagnostic report is generated. This report includes the electrophysiological evidence chain for disease determination, details of cross-data type feature association analysis, clinical feature matching results, and phased clinical recommendations. Furthermore, each part of the report can be viewed via an associated index.

2. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 1, characterized in that, The process involves performing data type difference compensation and dynamic correlation mapping on the electromyography (EMG) dataset and the evoked potential (EVP) dataset to establish a dynamic correlation between EMG features and EVP features under the same physiological state, resulting in a cross-data type correlation feature set including a correlation stability index, comprising: Electrophysiological signals of each muscle part under different activity states are extracted from the electromyography data set. The sampling frequency is standardized and the amplitude range is normalized for each electrophysiological signal under each activity state. Electromyography signals with different sampling frequencies and amplitude ranges are converted into electromyography standardized signals with a unified sampling frequency and a unified amplitude range. The electrophysiological response signals corresponding to each stimulus type are extracted from the evoked potential data set. The amplitude range of the electrophysiological response signals under each stimulus type is normalized, and the evoked potential signals with different amplitude ranges are converted into evoked potential standardized signals with a uniform amplitude range. The noise type and intensity of the electromyography (EMG) standardized signal are analyzed to determine the noise suppression parameters of the EMG signal. Based on these noise suppression parameters, adaptive noise suppression processing is performed on the EMG standardized signal to obtain the EMG denoised signal. The noise type and noise intensity of the evoked potential normalized signal are analyzed to determine the noise suppression parameters of the evoked potential signal. Based on the noise suppression parameters of the evoked potential signal, adaptive noise suppression processing is performed on the evoked potential normalized signal to obtain the evoked potential denoised signal. Based on the signal bandwidth, amplitude fluctuation range, and signal duration of the electromyography noise-reduced signal, and combined with the signal bandwidth, amplitude fluctuation range, and signal duration of the evoked potential noise-reduced signal, a data type adaptation model is constructed. This data type adaptation model includes signal parameter conversion formulas and data type difference compensation coefficients. The electromyography (EMG) noise-reduced signal is input into the data type adaptation model. The signal dimension of the EMG noise-reduced signal is adjusted by the signal parameter conversion formula. Then, the signal characteristic difference between EMG and evoked potential is compensated by the data type difference compensation coefficient to obtain the EMG adaptation signal. The evoked potential noise reduction signal is input into the data type adaptation model. The signal dimension of the evoked potential noise reduction signal is adjusted by the signal parameter conversion formula. Then, the difference in signal characteristics between evoked potential and electromyography is compensated by the data type difference compensation coefficient to obtain the evoked potential adaptation signal. The discharge frequency, discharge amplitude, and discharge interval features of the electromyography (EMG) adaptation signal are extracted to form an EMG adaptation feature subset; the latency, amplitude, and waveform morphology features of the evoked potential (EVP) adaptation signal are extracted to form an EVP adaptation feature subset. The electromyography (EMG) fitting feature subset and the evoked potential (EVP) fitting feature subset are aligned on the time axis to determine the corresponding EMG fitting features and EVP fitting features within the same time interval. Calculate the correlation coefficient between each pair of corresponding features, determine the feature association weights based on the correlation coefficients, and construct a feature association mapping matrix; Calculate the fluctuation range of each element in the feature association mapping matrix. If the fluctuation range exceeds the preset range, adjust the difference compensation coefficient of the data type adaptation model and regenerate the feature association mapping matrix until the fluctuation range meets the preset range. Based on the stabilized feature association mapping matrix, the electromyography-adapted feature subset and the evoked potential-adapted feature subset are recombined to generate a cross-data type association feature set containing the association stability index of each pair of associated features. The association stability index is calculated by the fluctuation range of the feature association weights and the correlation coefficient.

3. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 1, characterized in that, The pre-trained multi-round iterative intelligent diagnostic reasoning model is invoked to perform symptom association analysis on the cross-data type associated feature set, generating preliminary diagnostic conclusions containing multiple suspected symptoms and their multi-dimensional association descriptions, including: The cross-data type associated feature set is input into the feature preprocessing layer of the multi-round iterative intelligent diagnostic reasoning model. Outlier detection is performed on each associated feature in the cross-data type associated feature set. If an outlier exists, the outlier is replaced based on the historical normal data distribution of the associated feature to obtain the preprocessed cross-data type associated feature set. The preprocessed cross-data type associated feature set is subjected to feature dimension compression, and a feature selection algorithm is used to filter the associated features that contribute more than a preset threshold to disease identification, thus forming a core cross-data type associated feature set. The core cross-data type association feature set is input into the initial diagnosis layer of the multi-round iterative intelligent diagnosis reasoning model, and the disease feature template library built into the multi-round iterative intelligent diagnosis reasoning model is called. The disease feature template library contains typical cross-data type association feature templates corresponding to a variety of preset diseases, and each typical cross-data type association feature template is labeled with feature weight distribution. The similarity between the core cross-data type association feature set and each typical cross-data type association feature template is calculated. The similarity calculation between the core cross-data type association feature set and each typical cross-data type association feature template is performed from three dimensions: feature numerical matching degree, association stability index matching degree, and feature temporal change trend matching degree. Based on the similarity calculation results of the three dimensions, the initial correlation degree of each preset disease is obtained by weighted summation. The weighting coefficient of this weighted summation method is obtained by training the multi-round iterative intelligent diagnostic reasoning model through historical diagnostic data. Filter out the preset symptoms whose initial correlation exceeds the preset initial threshold to form an initial suspected symptom list. This initial suspected symptom list includes the initial correlation of each initial suspected symptom and the similarity details in three dimensions. The initial list of suspected symptoms is input into the symptom feature feedback module of the multi-round iterative intelligent diagnostic reasoning model. The symptom feature feedback module calls the symptom association rule library built into the multi-round iterative intelligent diagnostic reasoning model to extract the association feature constraints corresponding to each initial suspected symptom. Analyze whether the core cross-data type association feature set satisfies the association feature constraints of each initial suspected disease. If not, calculate the degree of non-satisfaction of the association feature constraints and reduce the initial association degree of the corresponding initial suspected disease based on the degree of non-satisfaction. If the initial correlation of an initial suspected disease decreases below the preset initial threshold, the disease is removed from the initial suspected disease list; if the initial correlation of an initial suspected disease decreases but remains above the preset initial threshold, the disease is retained in the initial suspected disease list, and details of the unmet correlation feature constraints are recorded. The adjusted initial list of suspected symptoms is input into the iterative optimization layer of the multi-round iterative intelligent diagnostic reasoning model. Based on the satisfaction of the associated feature constraints for each suspected symptom, a subset of features related to the associated feature constraints is extracted from the core cross-data type associated feature set. The feature subset is subjected to feature enhancement processing to increase the weight of key features in the feature subset, thereby generating an enhanced feature subset; The similarity between the enhanced feature subset and the corresponding typical cross-data type association feature template of the suspected disease is recalculated, and the association degree of each suspected disease is updated to obtain the iterative association degree. Repeat the operations of feature subset extraction, feature enhancement, similarity recalculation, and correlation update until the number of iterations reaches the preset number of iterations or the change in the correlation of suspected symptoms is lower than the preset change threshold. For each suspected symptom after iteration, a multi-dimensional correlation description is generated based on the matching degree of feature numerical values, the matching degree of correlation stability index, the matching degree of feature temporal change trend, and the satisfaction degree of constraint conditions. The multi-dimensional correlation description includes specific calculation results and level classification. Integrate all iterative suspected symptoms and their multi-dimensional correlation descriptions to generate preliminary diagnostic conclusions.

4. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 1, characterized in that, The process of fusing the preliminary diagnostic conclusion with the examinee's clinical reference information using a multi-source evidence chain to obtain an optimized diagnostic conclusion includes: Obtain the clinical reference information of the examinee, which includes the examinee's medical history record information, physical examination information and other auxiliary examination information. The medical history record information includes the history of previous neurological diseases, muscle diseases and medication history. The physical examination information includes the results of muscle strength test, nerve reflex test and muscle atrophy test. Other auxiliary examination information includes the results of blood biochemistry test and imaging test. The medical history record information is subjected to text semantic analysis to extract key information related to the nervous and muscular systems, forming key features of the medical history. These key features include disease type, duration of onset, frequency of symptom manifestation, and description of the effects of medication. The physical examination information is structured to convert non-standardized physical examination results into standardized feature descriptions, forming key physical examination features. These key physical examination features include descriptions of muscle strength level, nerve reflex response type, and degree of muscle atrophy. The data format of the other auxiliary examination information is uniformly processed, and the results of different types of auxiliary examinations are converted into directly comparable numerical or classification features to form key features of auxiliary examinations. The key features of auxiliary examinations include descriptions of the range of biochemical indicators and descriptions of abnormal imaging areas. Integrate the key features of the medical history, the key features of the physical examination, and the key features of the auxiliary examinations to form a set of key clinical features, and label the feature credibility of each key clinical feature. Extract the multi-dimensional correlation description of each suspected disease from the preliminary diagnostic conclusion, determine the core cross-data type correlation features corresponding to each suspected disease, and form a set of electrophysiological features of suspected diseases. An evidence association network of electrophysiological features and clinical features is constructed. The nodes of the evidence association network of electrophysiological features and clinical features contain each electrophysiological feature in the set of electrophysiological features of suspected diseases and each clinical feature in the set of key clinical features. The edges of the evidence association network of electrophysiological features and clinical features represent the association between two nodes. The association is calculated by the frequency of co-occurrence of the two features in historical diagnostic data. An initial weight is assigned to each node in the evidence association network between the electrophysiological feature and the clinical feature. The initial weight of the electrophysiological feature node is determined by the contribution of the electrophysiological feature in the multi-dimensional association description, and the initial weight of the clinical feature node is determined by the credibility of the clinical feature. Based on the strength of the edge association in the evidence association network of the electrophysiological feature and the clinical feature, the node weight propagation algorithm is used to update the weight of each node in the evidence association network of the electrophysiological feature and the clinical feature, so that the closely related nodes pass weights to each other until the weight change of all nodes in the evidence association network of the electrophysiological feature and the clinical feature is lower than the preset weight threshold. For each suspected disease, the final weight sum of the electrophysiological feature nodes corresponding to the suspected disease and the final weight sum of the clinical feature nodes related to the suspected disease are calculated. The clinical fusion correlation degree of the suspected disease is obtained by weighted fusion. The weighting ratio of the weighted fusion method is determined by the disease type. Neurological diseases focus on the weight of electrophysiological features, while muscular diseases focus on the weight of clinical features. Based on the clinical fusion correlation, the multi-dimensional correlation description of the corresponding suspected disease is adjusted, thereby increasing the description level of suspected diseases with high clinical fusion correlation in the feature time-series change trend matching dimension and decreasing the description level of suspected diseases with low clinical fusion correlation in the feature time-series change trend matching dimension. Screen suspected diseases whose clinical fusion correlation exceeds a preset fusion threshold, and sort the screened suspected diseases from high to low according to their clinical fusion correlation. By integrating and ranking the suspected symptoms and their adjusted multi-dimensional correlation descriptions, an optimized diagnostic conclusion is obtained.

5. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 2, characterized in that, The step involves constructing a data type adaptation model based on the signal bandwidth, amplitude fluctuation range, and signal duration of the electromyography (EMG) noise-reduced signal, combined with the signal bandwidth, amplitude fluctuation range, and signal duration of the evoked potential (EVP) noise-reduced signal. This includes: The signal bandwidth of the electromyography (EMG) noise-reduced signal under different activity states is statistically analyzed, and the average value and standard deviation of the signal bandwidth under all activity states are calculated to obtain the EMG signal bandwidth characteristics. The amplitude fluctuation range of the electromyography (EMG) noise-reduced signal under different activity states was statistically analyzed, and the average value and standard deviation of the amplitude fluctuation range under all activity states were calculated to obtain the amplitude characteristics of the EMG signal. The duration of the electromyography (EMG) noise-reduced signal under different activity states was statistically analyzed, and the average and standard deviation of the signal duration under all activity states were calculated to obtain the temporal characteristics of the EMG signal. Using the same statistical method, the bandwidth characteristics, amplitude characteristics, and temporal characteristics of the evoked potential noise reduction signal under different stimulus types were obtained respectively. The difference between the bandwidth characteristics of electromyography signals and the bandwidth characteristics of evoked potential signals is calculated. Based on this difference, a bandwidth conversion formula is constructed. The bandwidth conversion formula is a linear conversion, and the conversion coefficient of the linear conversion is determined by the difference and historical adaptation data. The difference between the amplitude characteristics of electromyography (EMG) signals and the amplitude characteristics of evoked potential (EVP) signals is calculated. An amplitude conversion formula is constructed based on this difference. The amplitude conversion formula is a nonlinear conversion, and the conversion parameters of the nonlinear conversion are obtained by fitting the amplitude distribution curves of the two signals. The difference between the temporal characteristics of electromyography (EMG) signals and the temporal characteristics of evoked potential (EVP) signals is calculated. A temporal conversion formula is constructed based on this difference. The temporal conversion formula includes a time offset correction term, which is determined by the acquisition delay time of the two signals. By integrating the bandwidth conversion formula, the amplitude conversion formula, and the timing conversion formula, a signal parameter conversion formula for the data type adaptation model is formed. The noise residual difference between electromyography (EMG) signals and evoked potential (EVP) signals was analyzed, and the noise power spectral density difference between the two signals in the same frequency range was calculated. Based on this noise power spectral density difference, the noise compensation coefficient was determined. The differences in signal attenuation characteristics between electromyography (EMG) signals and evoked potential (EVP) signals were analyzed. The difference in amplitude attenuation rate between the two signals at the same transmission distance was calculated, and the attenuation compensation coefficient was determined based on this difference in amplitude attenuation rate. The difference in signal response speed between electromyography (EMG) signals and evoked potential (EVP) signals was analyzed, and the difference in response delay time between the two signals under the same stimulus intensity was calculated. Based on this difference in response delay time, the response compensation coefficient was determined. The noise compensation coefficient, the attenuation compensation coefficient, and the response compensation coefficient are integrated to form the data type difference compensation coefficient of the data type adaptation model. Each compensation coefficient contains sub-coefficients for different signal states. Electromyography signal states are divided according to activity state, and evoked potential signal states are divided according to stimulus type.

6. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 3, characterized in that, The calculation of the similarity between the core cross-data type association feature set and each typical cross-data type association feature template includes: For each associated feature in the core cross-data type associated feature set, extract the feature value of the associated feature, and at the same time extract the feature value range of the same type of associated feature in the corresponding typical cross-data type associated feature template; Calculate the degree of overlap between the feature value of the associated feature and the feature value range. If the feature value of the associated feature is within the feature value range, the degree of overlap is a specific value. If the feature value of the associated feature is outside the feature value range, calculate the distance between the feature value of the associated feature and the boundary of the feature value range. Calculate the degree of overlap inversely based on this distance to obtain the numerical matching degree of the associated feature. The numerical matching degree of all associated features in the core cross-data type associated feature set is weighted and averaged. The weighting coefficient of the weighted average is the feature weight of each associated feature in the typical cross-data type associated feature template, and the feature numerical matching degree is obtained. Extract the association stability index of each association feature in the core cross-data type association feature set, and at the same time extract the association stability index range of the same type of association features in the corresponding typical cross-data type association feature template; The deviation rate between the correlation stability index and the index range of the correlation feature is calculated. The deviation rate is the difference between the actual index and the midpoint of the index range divided by the absolute value of the half-width of the index range. The stable matching sub-degree is calculated based on the deviation rate, and the calculation of the stable matching sub-degree adopts a preset mapping relationship between the deviation rate and the sub-degree. The stability matching degree of all associated features in the core cross-data type associated feature set is weighted and averaged. The weighting coefficient of the weighted average is the feature weight of each associated feature in the typical cross-data type associated feature template, and the association stability index matching degree is obtained. For each associated feature in the core cross-data type associated feature set, the feature values ​​of the associated feature are arranged in chronological order to form a time sequence of feature values; Extract the time-series trend curve of the same type of associated features in the corresponding typical cross-data type associated feature template. The time-series trend curve of the feature values ​​includes the division of rising segment, steady segment, and falling segment and the range of the rate of change of each segment. The dynamic time warping algorithm is used to align the feature numerical time series with the feature numerical time series change trend curve, and the Euclidean distance between the aligned sequence and the curve is calculated. Trend matching sub-degree is calculated based on Euclidean distance, and the calculation of trend matching sub-degree adopts a preset mapping relationship between distance and sub-degree; The trend matching degree of all related features in the core cross-data type related feature set is weighted and averaged. The weighting coefficient of the weighted average is the feature weight of each related feature in the typical cross-data type related feature template, so as to obtain the feature time series change trend matching degree.

7. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 4, characterized in that, The construction of the evidence association network between electrophysiological features and clinical features includes: Extract all electrophysiological features from the set of electrophysiological features of suspected diseases, assign a unique electrophysiological node identifier to each electrophysiological feature, and form an electrophysiological node set; Extract all clinical features from the set of key clinical features, assign a unique clinical node identifier to each clinical feature, and form a set of clinical nodes. The electrophysiological node set and the clinical node set are integrated to form a node set of an evidence association network of electrophysiological features and clinical features. Each node includes a node type, feature description and corresponding suspected disease identifier. The node types are divided into two categories: electrophysiological and clinical. Obtain the historical diagnostic database built into the multi-round iterative intelligent diagnostic reasoning model. This historical diagnostic database contains electrophysiological characteristic data, clinical characteristic data, and final diagnostic results of multiple historical subjects. For each suspected symptom, historical records in which the final diagnosis result contains the suspected symptom are filtered from the historical diagnosis database to form a subset of historical related records for the suspected symptom; For each record in the subset of historical records related to the suspected disease, extract the electrophysiological and clinical features, and count the number of times each pair of electrophysiological and clinical features co-occurs. The frequency of co-occurrence of each pair of electrophysiological and clinical features is obtained by calculating the ratio of the number of times each pair of electrophysiological and clinical features co-occurs to the total number of records in the subset of historical records associated with the suspected disease. If the co-occurrence frequency exceeds a preset frequency threshold, then an edge is established between the corresponding electrophysiological node and the clinical node to form an evidence association network of electrophysiological features and clinical features. The weight of the edge of the evidence association network of electrophysiological features and clinical features is the co-occurrence frequency. If the frequency of co-occurrence does not exceed a preset frequency threshold, then no edge is established in the evidence association network between electrophysiological features and clinical features; For all electrophysiological and clinical nodes of the same suspected disease, the operations of calculating the common occurrence frequency and establishing the edges of the evidence association network of electrophysiological and clinical features are repeatedly performed to form a local evidence association network for the suspected disease. Integrate the local evidence association networks of all suspected symptoms, remove duplicate nodes and edges of the evidence association networks of electrophysiological features and clinical features, and take the average weight of the edges of the evidence association networks of duplicate electrophysiological features and clinical features as the average weight in each local network to obtain the global evidence association network of electrophysiological features and clinical features. For each edge in the evidence association network of global electrophysiological features and clinical features, label the suspected disease identifier corresponding to that edge.

8. The intelligent diagnostic report generation method based on electromyography and evoked potential data according to claim 7, characterized in that, For each edge in the evidence association network of global electrophysiological features and clinical features, the suspected symptom identifier corresponding to that edge is labeled, including: Each suspected symptom is assigned a unique symptom identification code, which includes a symptom category number and a symptom sequence number. The symptom category number is divided according to the system to which the symptom belongs, and the system to which the symptom belongs is divided into two categories: nervous system and muscular system. Traverse each edge in the evidence association network of global electrophysiological and clinical features to obtain the source of that edge's establishment; If the edge comes from only a local evidence network of a suspected disease, then add the disease identification code of the suspected disease to the attribute information of the edge, marking it as a single source plus disease identification code; If the edge comes from a local evidence network of multiple suspected diseases, then add the disease identification codes of all relevant suspected diseases to the attribute information of the edge, with each disease identification code separated by a delimiter, and marked as multiple sources plus multiple disease identification codes; For the labeled edges, add the common frequency of each suspected disease from each source. This common frequency corresponds one-to-one with the disease identification code, forming a key-value pair format of disease identification code plus frequency, and add it to the attribute information of the edge. After all edges are labeled, an attribute list of the evidence association network of global electrophysiological and clinical features is generated. This attribute list includes the identifier of each edge, the connected node pairs, the weight, the disease identification code, and the corresponding frequency. This list of attributes is bound and stored with an evidence network linking global electrophysiological and clinical characteristics.

9. An intelligent diagnostic report generation system based on electromyography and evoked potential data, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the intelligent diagnostic report generation method based on electromyography and evoked potential data according to any one of claims 1 to 8 by executing the machine-executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. The processor of the intelligent diagnostic report generation system based on electromyography and evoked potential data reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the intelligent diagnostic report generation system based on electromyography and evoked potential data to perform the intelligent diagnostic report generation method based on electromyography and evoked potential data as described in any one of claims 1 to 8.