Neuropathic pain detection and analysis system based on electroencephalogram signals
By combining a 32-channel EEG electrode array and an improved adaptive filtering algorithm with a deep learning model, the objectivity and accuracy of neuropathic pain detection have been solved, enabling accurate detection and type differentiation of neuropathic pain, adapting to the differences in EEG signals among different individuals, and simplifying the operation process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTH CHINA UNIVERSITY OF SCIENCE & TECHNOLOGY AFFILIATED HOSPITAL
- Filing Date
- 2026-04-17
- Publication Date
- 2026-05-19
AI Technical Summary
Current technologies for detecting neuropathic pain rely on patients' subjective descriptions, lack objective quantitative indicators, are susceptible to interference when collecting EEG signals, have insufficient feature extraction, lack specificity in detection models, and lack dynamic calibration, leading to misdiagnosis, missed diagnosis, and complex operation.
A 32-channel EEG electrode array combined with an improved adaptive filtering algorithm is used to filter out interference. Pain type differentiation is achieved through multi-dimensional feature extraction and an improved CNN-LSTM-AM deep learning model. A dynamic calibration module is also provided to enable the system to achieve accurate detection and adaptive optimization.
It enables objective and accurate detection of neuropathic pain, avoids misdiagnosis and missed diagnosis, improves the accuracy of pain type differentiation, has dynamic calibration function, is easy to operate, and adapts to the differences in EEG signals of different individuals.
Smart Images
Figure CN122056566A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical testing technology, specifically a neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals. Background Technology
[0002] Neuropathic pain is chronic pain caused by damage or dysfunction of the nervous system. Its pathogenesis is complex and its clinical manifestations are diverse, mainly including peripheral neuropathic pain, central neuropathic pain, and mixed neuropathic pain. Currently, clinical detection of neuropathic pain relies primarily on patients' subjective descriptions combined with doctors' clinical experience. The lack of objective, quantifiable testing indicators leads to significant subjectivity in the results, making misdiagnosis and missed diagnosis common.
[0003] With the development of EEG signal detection technology, some pain detection methods based on EEG signals have emerged in existing technologies. However, these methods generally have the following drawbacks: First, during the EEG signal acquisition process, they are easily affected by power frequency interference, electromyographic interference, and electrode contact noise, resulting in poor signal quality and affecting detection accuracy. Second, feature extraction is not comprehensive enough, extracting only single time-domain or frequency-domain features, which cannot fully reflect the EEG signal change patterns corresponding to neuropathic pain. Third, the detection model is not highly targeted and has not been optimized in conjunction with the specific EEG characteristics of neuropathic pain, resulting in low accuracy in distinguishing pain types. Fourth, the system lacks a dynamic calibration mechanism, and the detection accuracy is prone to decline after long-term use, and it cannot adapt to the differences in EEG signals among different individuals. Fifth, the detection process is cumbersome and the operation is complex, making it difficult to meet the needs of rapid clinical detection.
[0004] Therefore, developing a neuropathic pain detection and analysis system based on EEG signals that can accurately collect EEG signals, comprehensively extract pain-related features, accurately distinguish pain types, and has dynamic calibration functions and is easy to operate is key to overcoming the shortcomings of existing technologies.
[0005] Based on this, a neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals was designed. Summary of the Invention
[0006] In order to overcome the shortcomings of the prior art, the present invention provides a neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals, which effectively solves the problems mentioned in the background art.
[0007] To achieve the above objectives, the present invention provides the following technical solution: a neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals, comprising an EEG signal acquisition module, a signal preprocessing module, a feature extraction module, a pain detection and analysis module, a result output module, and a system calibration module, wherein each module is electrically connected in sequence to form a closed-loop detection and analysis system; The EEG signal acquisition module includes a 32-channel EEG electrode array, a signal buffer, and a preliminary noise reduction unit. The 32-channel EEG electrode array is attached to a preset area of the human scalp to acquire raw EEG signals. The signal buffer is used to buffer and amplify the acquired raw EEG signals to avoid signal attenuation. The preliminary noise reduction unit uses an RC low-pass filter circuit to filter out high-frequency interference signals in the raw EEG signals and transmits the preliminary noise-reduced EEG signals to the signal preprocessing module. The signal preprocessing module includes an adaptive filtering unit, a baseline correction unit, a standardization unit, and a signal segmentation unit. The adaptive filtering unit uses an improved adaptive noise cancellation algorithm to filter out power frequency interference, electromyographic interference, and electrode contact noise in the EEG signal. The baseline correction unit uses a moving average algorithm to correct baseline drift in the filtered EEG signal. The standardization unit uses a Z-score standardization algorithm to convert the corrected EEG signal into a standardized signal, eliminating the influence of individual differences. The signal segmentation unit segments the standardized EEG signal according to a preset time length and outputs the segmented standardized EEG signal. The feature extraction module includes a time-domain feature extraction unit, a frequency-domain feature extraction unit, and a spatial-domain feature extraction unit. The time-domain feature extraction unit is used to extract the peak value, trough value, peak-to-peak value, mean, variance, and waveform slope features of the segmented standardized EEG signal. The frequency-domain feature extraction unit uses a Fourier transform algorithm to convert the time-domain EEG signal into a frequency-domain signal and extracts the power spectral density features of the θ band, α band, β band, and γ band. The spatial-domain feature extraction unit is used to extract the correlation features and spatial distribution features of EEG signals from different channels. The pain detection and analysis module includes a model storage unit, a feature fusion unit, and a detection and analysis unit. The model storage unit stores a pre-set improved CNN-LSTM-AM deep learning model, which consists of a convolutional neural network (CNN), a bidirectional long short-term memory network (BiLSTM), and an attention mechanism (AM). CNN is used to extract spatial features of the feature parameters, BiLSTM is used to extract temporal series features of the feature parameters, and the attention mechanism is used to assign differentiated weights to different features. The feature fusion unit fuses the temporal, frequency, and spatial domain feature parameters to generate a fused feature vector. The detection and analysis unit inputs the fused feature vector into the improved CNN-LSTM-AM deep learning model and outputs the presence or absence of neuropathic pain and the pain type differentiation result through model computation. The pain types include peripheral neuropathic pain, central neuropathic pain, and mixed neuropathic pain. The result output module includes a display unit, a data storage unit, and an alarm unit. The display unit is used to intuitively present the pain detection and analysis results, EEG signal waveforms, and characteristic parameters in numerical, curve, and text form. The data storage unit is used to store the collected raw EEG signals, preprocessed signals, extracted characteristic parameters, and detection and analysis results, and supports data query and export. The alarm unit is used to issue an audible and visual alarm when a neuropathic pain signal is detected and the pain level reaches a preset threshold. The system calibration module includes a standard sample import unit, an error calculation unit, and a parameter optimization unit. The standard sample import unit is used to import standard EEG sample data with known pain types and pain states. The error calculation unit is used to calculate the deviation between the current detection and analysis results and the standard sample data. The parameter optimization unit is used to dynamically adjust the parameters of the improved CNN-LSTM-AM deep learning model based on the deviation value using a gradient descent algorithm to achieve calibration of the system detection accuracy.
[0008] Preferably, the 32-channel EEG electrode array uses Ag / AgCl electrodes, and the electrode contact positions include the frontal lobe, parietal lobe, temporal lobe and occipital lobe regions. Conductive gel is placed between each electrode and the human scalp to ensure good contact between the electrode and the scalp. The sampling rate of the electrode array can be adjusted between 256Hz and 1024Hz.
[0009] Preferably, the improved adaptive noise cancellation algorithm constructs an interference signal model by collecting human electromyography reference signals and power frequency reference signals, and dynamically adjusts the filtering coefficients using an adaptive filtering algorithm to achieve accurate filtering of interference signals. The adjustment step size of the filtering coefficients is adaptively adjusted in real time according to the amplitude changes of the EEG signal.
[0010] Preferably, the preset time length of the signal segmentation unit is 1s-5s, and the segmented data adopts an overlapping segmentation method with an overlap rate of 50%-70% to ensure the continuity of feature extraction and avoid loss of feature information.
[0011] Preferably, the frequency range of the θ band is 4Hz-7Hz, the frequency range of the α band is 8Hz-12Hz, the frequency range of the β band is 13Hz-29Hz, and the frequency range of the γ band is 30Hz-50Hz. The power spectral density characteristics of each band are obtained by calculating the power integral of the signal within that band.
[0012] Preferably, the training process of the improved CNN-LSTM-AM deep learning model includes: importing a dataset of EEG samples labeled with pain states and pain types, dividing the dataset into a training set, a validation set, and a test set; training the model using the training set, adjusting the model hyperparameters using the validation set, validating the model performance using the test set, and storing the model in a model storage unit after training to support online updates and iterations of the model.
[0013] Preferably, the system further includes a power supply module, which uses a rechargeable lithium battery with an output voltage of 3.3V-5V, supports a battery life of no less than 8 hours, and is equipped with a power management unit to realize charging protection, over-discharge protection and low battery warning functions.
[0014] Preferably, the display unit of the result output module adopts a touch-screen LCD display, which supports multiple interface switching and can display real-time waveforms of EEG signals, feature parameter curves, detection and analysis results, and historical data query interface respectively; the data storage unit adopts an SD card or solid-state drive with a storage capacity of not less than 16GB and supports data export via USB interface.
[0015] Preferably, the calibration cycle of the system calibration module can be set manually or triggered automatically. The automatic triggering condition is that after a preset number of consecutive tests, the detection error exceeds a preset threshold, triggering the automatic calibration process.
[0016] Preferably, each module uses an SPI interface or a USB interface for data transmission, with a transmission rate of no less than 1Mbps, to ensure the real-time performance and stability of data transmission. At the same time, a data encryption unit is provided to encrypt the transmitted data and ensure data security.
[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention enables objective detection of neuropathic pain without relying on the patient's subjective description. Through purely objective analysis of electroencephalogram (EEG) signals, it effectively avoids misdiagnosis and missed diagnosis caused by subjective factors, providing a reliable objective basis for clinical diagnosis and treatment. 2. This invention has high precision in acquiring EEG signals. It comprehensively acquires EEG signals from different regions of the scalp through a 32-channel electrode array. Combined with an improved adaptive filtering algorithm, it effectively filters out various interference signals, ensuring signal quality and providing a reliable foundation for feature extraction and detection analysis. 3. The present invention has comprehensive feature extraction, and the combination of multi-dimensional features can fully reflect the changes in EEG signals corresponding to neuropathic pain. Combined with the improved deep learning model, it significantly improves the accuracy of pain detection and the precision of pain type differentiation. 4. This invention has a dynamic calibration function, which can dynamically adjust the model parameters according to standard sample data, adapt to the differences in EEG signals of different individuals, avoid the decline in detection accuracy after long-term use, and ensure the stability and reliability of detection results. 5. This invention is highly practical and can be widely used in the diagnosis of neuropathic pain, differentiation of pain types, and monitoring of the rehabilitation process in clinical practice. It can also be applied to basic research on neuropathic pain and has broad application prospects. Attached Figure Description
[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0019] In the attached diagram: Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the system workflow of the present invention; Figure 3 This is a logic block diagram of the pain detection and analysis module of the present invention; Detailed Implementation
[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0021] Depend on Figure 1 The present invention relates to a neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals, comprising an EEG signal acquisition module, a signal preprocessing module, a feature extraction module, a pain detection and analysis module, a result output module, and a system calibration module, wherein the modules are electrically connected in sequence to form a closed-loop detection and analysis system.
[0022] 1) EEG signal acquisition module It is the core unit of the system for acquiring raw EEG signals. It is used to collect raw EEG signals from different areas of the human scalp and perform preliminary noise reduction processing to ensure the integrity and initial purity of the signal. The module includes a 32-channel EEG electrode array, a signal buffer, and a preliminary noise reduction unit.
[0023] The 32-channel EEG electrode array uses Ag / AgCl electrodes, which are characterized by good biocompatibility, excellent conductivity, and low noise, enabling stable acquisition of EEG signals from the human scalp. The electrodes are positioned to cover the frontal, parietal, temporal, and occipital lobes, regions where EEG signals are closely related to the occurrence and development of neuropathic pain, allowing for comprehensive capture of pain-related EEG changes. Conductive gel is placed between each electrode and the scalp to reduce contact impedance, ensuring good contact and preventing signal distortion due to poor contact. The sampling rate of the electrode array can be adjusted between 256Hz and 1024Hz, allowing for flexible selection based on testing needs, balancing detection accuracy and data volume.
[0024] The signal buffer is constructed using an operational amplifier to buffer and amplify the acquired raw EEG signal. The amplification factor is adjustable to avoid signal attenuation during transmission and ensure that the signal can be stably transmitted to subsequent modules.
[0025] The preliminary noise reduction unit uses an RC low-pass filter circuit with a filter cutoff frequency of 50Hz. It is mainly used to filter out high-frequency interference signals (such as environmental electromagnetic interference) in the original EEG signal, while retaining the effective frequency components of the EEG signal, thus completing the preliminary noise reduction process. The EEG signal after preliminary noise reduction is then transmitted to the signal preprocessing module.
[0026] 2) Signal preprocessing module This module is used to further process the preliminary noise-reduced signal transmitted by the EEG signal acquisition module, eliminate various interference signals, correct signal baseline drift, and achieve signal standardization, providing a high-quality signal foundation for subsequent feature extraction. The module includes an adaptive filtering unit, a baseline correction unit, a standardization unit, and a signal segmentation unit.
[0027] The adaptive filtering unit employs an improved adaptive noise cancellation algorithm. This algorithm constructs an interference signal model by acquiring human electromyography (EMG) reference signals (acquired through dedicated EMG electrodes) and power frequency reference signals (50Hz). It then dynamically adjusts the filtering coefficients using the adaptive filtering algorithm to achieve precise filtering of interference signals. Compared with traditional adaptive filtering algorithms, the improved algorithm's filtering coefficient adjustment step size can be adaptively adjusted in real time according to the amplitude changes of the EEG signal. This effectively filters out power frequency interference, EMG interference, and electrode contact noise, while avoiding filtering out valid EEG signals and ensuring signal integrity.
[0028] The baseline correction unit uses a moving average algorithm with a sliding window length of 100ms-500ms. By performing a moving average calculation on the filtered EEG signal, the signal baseline is obtained, and then the EEG signal is corrected for baseline drift. This eliminates baseline drift caused by changes in human physiological state and electrode contact state, making the EEG signal benchmark more consistent.
[0029] The standardized unit uses the Z-score normalization algorithm to convert the baseline-corrected EEG signal into a standardized signal, eliminating the influence of differences in EEG signal amplitude and electrode position between different individuals, making the EEG signals of different individuals comparable, and providing a unified signal standard for subsequent feature extraction and detection analysis.
[0030] The signal segmentation unit segments the standardized EEG signal into segments according to a preset time length, which is 1s-5s and can be flexibly set according to the detection requirements. The segmented data adopts an overlapping segmentation method with an overlap rate of 50%-70% to ensure the continuity of feature extraction and avoid the loss of feature information due to segmentation. The segmented standardized EEG signal is then transmitted to the feature extraction module.
[0031] 3) Feature extraction module This module is used to extract feature parameters closely related to neuropathic pain from the segmented standardized EEG signals output by the signal preprocessing module, providing core data support for pain detection and analysis. The module includes a time-domain feature extraction unit, a frequency-domain feature extraction unit, and a spatial-domain feature extraction unit, enabling comprehensive extraction of multi-dimensional features.
[0032] The temporal feature extraction unit is used to extract temporal feature parameters of the segmented standardized EEG signal, specifically including peak value, trough value, peak-to-peak value, mean, variance, and waveform slope. Peak value and trough value reflect the amplitude variation range of the EEG signal, peak-to-peak value reflects the fluctuation amplitude of the signal, mean reflects the overall amplitude level of the signal, variance reflects the dispersion of the signal, and waveform slope reflects the rate of change of the signal. These temporal features can intuitively reflect the temporal variation pattern of the EEG signal when neuropathic pain occurs.
[0033] The frequency domain feature extraction unit employs a Fourier transform algorithm to convert time-domain EEG signals into frequency-domain signals, extracting the power spectral density features of characteristic bands associated with neuropathic pain. These characteristic bands include theta band (4Hz-7Hz), alpha band (8Hz-12Hz), beta band (13Hz-29Hz), and gamma band (30Hz-50Hz). The power variations in these bands are closely related to the occurrence, type, and intensity of neuropathic pain. The power spectral density features of each band are obtained by calculating the power integral of the signal within that band, quantitatively reflecting the energy distribution of each band. The frequency ranges of each band are non-overlapping and clearly defined: 4Hz-7Hz is the theta band, 8Hz-12Hz is the alpha band, 13Hz-29Hz is the alpha band, and 30Hz-50Hz is the gamma band, effectively avoiding the problem of ambiguous frequency assignment.
[0034] The spatial domain feature extraction unit is used to extract the correlation and spatial distribution features of EEG signals from different channels. The correlation feature is obtained by calculating the correlation coefficient between any two channels of EEG signals, reflecting the coordinated changes in EEG activity in different scalp regions. The spatial distribution feature is obtained by spatially interpolating the signal amplitude of each channel to obtain the spatial distribution map of the EEG signals, reflecting the spatial distribution pattern of pain-related EEG signals. The combination of multi-dimensional features can comprehensively capture the changes in EEG signals corresponding to neuropathic pain and improve the accuracy of detection and analysis.
[0035] 4) Pain Detection and Analysis Module It is the core analysis unit of the system, used to analyze the multi-dimensional feature parameters output by the feature extraction module, so as to determine the presence or absence of neuropathic pain and distinguish the types of pain; this module includes a model storage unit, a feature fusion unit and a detection and analysis unit.
[0036] The model storage unit stores a pre-defined improved CNN-LSTM-AM deep learning model, which consists of a convolutional neural network (CNN), a bidirectional long short-term memory network (BiLSTM), and an attention mechanism (AM). CNN is used to extract spatial features of feature parameters, effectively capturing the spatial relationships between different features. BiLSTM is used to extract temporal features of feature parameters, capturing the dynamic changes in EEG signals. The attention mechanism assigns differentiated weights to different features, focusing on features strongly correlated with neuropathic pain, suppressing interference from irrelevant features, and improving the model's detection accuracy and specificity.
[0037] The training process of the improved CNN-LSTM-AM deep learning model is as follows: First, import the EEG sample dataset labeled with pain state (pain / no pain) and pain type (peripheral neuropathic pain, central neuropathic pain, mixed neuropathic pain), and divide the dataset into training set, validation set, and test set in a 7:2:1 ratio; Second, train the model using the training set, adjusting the model parameters through the backpropagation algorithm, and adjust the model hyperparameters (such as learning rate, number of iterations, and number of network layers) using the validation set to ensure the model's generalization ability; Finally, verify the model performance using the test set. After training, store the model in the model storage unit, supporting online updates and iterations of the model, and continuously optimize the model performance based on new sample data.
[0038] The feature fusion unit is used to fuse time-domain, frequency-domain, and spatial-domain feature parameters. By using feature concatenation, it integrates feature parameters from various dimensions into a unified fused feature vector, eliminating the limitations of single features, achieving complementarity of multi-dimensional features, and improving the accuracy of detection and analysis.
[0039] The detection and analysis unit is used to input the fused feature vector into the improved CNN-LSTM-AM deep learning model. Through model operation, it outputs the results of the presence or absence of neuropathic pain and the results of pain type differentiation. During the model operation, the attention mechanism is used to assign weights to different features in the fused feature vector, focusing on features that are strongly correlated with neuropathic pain, thereby achieving accurate pain detection and type differentiation. The detection and analysis results are transmitted to the result output module.
[0040] 5) Result Output Module This module is used to receive the detection and analysis results output by the pain detection and analysis module and present them to the user in an intuitive and easy-to-understand form, while also storing the data and providing alarm prompts. The module includes a display unit, a data storage unit, and an alarm unit.
[0041] The display unit uses a touch LCD screen with a screen size of no less than 7 inches. It supports multiple interface switching and can display real-time EEG waveforms, characteristic parameter curves, detection analysis results (presence or absence of pain, type of pain), and historical data query interface. Among them, the real-time EEG waveforms can display the signal changes of each channel, the characteristic parameter curves can intuitively present the changing trends of each characteristic parameter, and the detection analysis results are presented in the form of text and icons, making it easy for users to quickly obtain detection information.
[0042] The data storage unit uses an SD card or solid-state drive with a storage capacity of no less than 16GB. It is used to store the raw EEG signals, preprocessed signals, extracted feature parameters, and detection and analysis results. It supports data querying by time, detection object, and other conditions, and also supports data export via USB interface for easy subsequent data analysis and archiving.
[0043] The alarm unit uses an audible and visual alarm system with a preset pain level threshold. When a neuropathic pain signal is detected and the pain level reaches the preset threshold, the alarm unit will issue an audible and visual alarm to remind the user to pay attention to the pain status of the subject and facilitate timely diagnosis and treatment. The alarm threshold can be flexibly adjusted according to clinical needs.
[0044] 6) System calibration module This module is designed to dynamically optimize the system's detection accuracy, preventing a decline in accuracy over long-term use, while also adapting to differences in EEG signals among different individuals to ensure the reliability of the detection results. It includes a standard sample import unit, an error calculation unit, and a parameter optimization unit.
[0045] The standard sample import unit is used to import standard EEG sample data of known pain types and pain states. These standard sample data have been clinically validated and have high accuracy, serving as a reference benchmark for system calibration.
[0046] The error calculation unit is used to calculate the deviation between the current detection and analysis results and the standard sample data. The deviation is calculated using the mean square error algorithm, which quantitatively reflects the detection accuracy of the current system.
[0047] The parameter optimization unit is used to dynamically adjust the parameters of the improved CNN-LSTM-AM deep learning model based on the deviation value using the gradient descent algorithm, thereby optimizing the model's detection accuracy. The system calibration cycle can be set manually or triggered automatically. The automatic triggering condition is that after a preset number of consecutive detections (e.g., 10 times), if the detection error exceeds a preset threshold (e.g., 5%), the automatic calibration process is triggered to ensure that the system always maintains high detection accuracy.
[0048] 7) Power supply module The system also includes a power supply module to provide stable power support for each module. The power supply module uses a rechargeable lithium battery with an output voltage of 3.3V-5V, supporting a battery life of no less than 8 hours to meet the needs of clinical mobile testing. It is also equipped with a power management unit to realize charging protection, over-discharge protection and low battery warning functions, extend the service life of lithium batteries and ensure the stable operation of the system.
[0049] 8) Data transmission and encryption Data transmission between modules is achieved via SPI or USB interfaces with a transmission rate of no less than 1Mbps, ensuring real-time performance and stability. A data encryption unit is also included, employing the AES encryption algorithm to encrypt transmitted data, preventing tampering or theft and protecting the security and privacy of the test data.
[0050] like Figure 2As shown, the workflow of this system, specifically the following steps: Step 1: System initialization, power supply module starts up, powering each module, each module completes self-test to ensure normal system operation; Step 2: Attach the 32-channel EEG electrode array to the preset area of the subject's scalp, apply conductive gel to ensure good contact between the electrodes and the scalp, and set the sampling rate of the electrode array; Step 3: The EEG signal acquisition module is started to acquire the raw EEG signal of the subject. The signal is amplified by the signal buffer, and high-frequency interference is filtered out by the preliminary noise reduction unit. The preliminary noise-reduced signal is then transmitted to the signal preprocessing module. Step 4: The signal preprocessing module processes the signal after preliminary noise reduction. It filters out interference such as power frequency and electromyography through the adaptive filtering unit, corrects baseline drift through the baseline correction unit, standardizes the signal through the standardization unit, segments the standardized signal through the signal segmentation unit, and transmits the segmented signal to the feature extraction module. Step 5: The feature extraction module extracts time-domain, frequency-domain, and spatial-domain feature parameters from the segmented standardized signal and transmits the extracted feature parameters to the pain detection and analysis module. Step 6: The feature fusion unit of the pain detection and analysis module fuses multi-dimensional feature parameters into a fused feature vector. The detection and analysis unit inputs the fused feature vector into the improved CNN-LSTM-AM deep learning model, and outputs the pain detection and analysis results (pain presence or absence, pain type) through model calculation, which are then transmitted to the result output module. Step 7: The result output module displays the detection and analysis results, EEG signal waveforms and characteristic parameters through the display unit, and stores the relevant data through the data storage unit. If the pain level reaches the preset threshold, the alarm unit will issue an audible and visual alarm. Step 8: The system calibration module imports standard sample data, calculates the deviation between the current test result and the standard sample data, and if the deviation exceeds the preset threshold, the model parameters are adjusted through the parameter optimization unit to complete the system calibration. Step 9: Determine whether to end the detection. If the detection ends, shut down the system; if the detection continues, return to step 3 and repeat the above process.
[0051] like Figure 3The diagram shows the logical block diagram of the pain detection and analysis module. The specific logic is as follows: The time-domain, frequency-domain, and spatial-domain feature parameters output by the feature extraction module are input into the feature fusion unit. The feature fusion unit concatenates and fuses the three types of features to generate a fused feature vector. The fused feature vector is input into the improved CNN-LSTM-AM model. The CNN extracts spatial features, the BiLSTM extracts time-series features, and the attention mechanism assigns differentiated weights to different features. After the model operation, it outputs the results of whether pain is present and the results of pain type differentiation, which are transmitted to the result output module. Simultaneously, the model parameters can be dynamically adjusted through the system calibration module to optimize model performance.
[0052] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0053] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals, characterized in that, It includes an EEG signal acquisition module, a signal preprocessing module, a feature extraction module, a pain detection and analysis module, a result output module, and a system calibration module. These modules are electrically connected in sequence to form a closed-loop detection and analysis system. The EEG signal acquisition module includes a 32-channel EEG electrode array, a signal buffer, and a preliminary noise reduction unit. The 32-channel EEG electrode array is attached to a preset area of the human scalp to acquire raw EEG signals. The signal buffer is used to buffer and amplify the acquired raw EEG signals to avoid signal attenuation. The preliminary noise reduction unit uses an RC low-pass filter circuit to filter out high-frequency interference signals in the raw EEG signals and transmits the preliminary noise-reduced EEG signals to the signal preprocessing module. The signal preprocessing module includes an adaptive filtering unit, a baseline correction unit, a standardization unit, and a signal segmentation unit. The adaptive filtering unit uses an improved adaptive noise cancellation algorithm to filter out power frequency interference, electromyographic interference, and electrode contact noise in the EEG signal. The baseline correction unit uses a moving average algorithm to correct baseline drift in the filtered EEG signal. The standardization unit uses a Z-score standardization algorithm to convert the corrected EEG signal into a standardized signal, eliminating the influence of individual differences. The signal segmentation unit segments the standardized EEG signal according to a preset time length and outputs the segmented standardized EEG signal. The feature extraction module includes a time-domain feature extraction unit, a frequency-domain feature extraction unit, and a spatial-domain feature extraction unit. The time-domain feature extraction unit is used to extract the peak value, trough value, peak-to-peak value, mean, variance, and waveform slope features of the segmented standardized EEG signal. The frequency-domain feature extraction unit uses a Fourier transform algorithm to convert the time-domain EEG signal into a frequency-domain signal and extracts the power spectral density features of the θ band, α band, β band, and γ band. The spatial-domain feature extraction unit is used to extract the correlation features and spatial distribution features of EEG signals from different channels. The pain detection and analysis module includes a model storage unit, a feature fusion unit, and a detection and analysis unit. The model storage unit is used to store a preset improved CNN-LSTM-AM deep learning model, which consists of a convolutional neural network (CNN), a bidirectional long short-term memory network (BiLSTM), and an attention mechanism (AM). CNN is used to extract the spatial features of the feature parameters, BiLSTM is used to extract the time-series features of the feature parameters, and the attention mechanism is used to assign differentiated weights to different features. The feature fusion unit is used to fuse time-domain, frequency-domain, and spatial-domain feature parameters to generate a fused feature vector; the detection and analysis unit is used to input the fused feature vector into the improved CNN-LSTM-AM deep learning model, and output the presence or absence of neuropathic pain and the pain type differentiation result through model operation. The pain types include peripheral neuropathic pain, central neuropathic pain, and mixed neuropathic pain. The result output module includes a display unit, a data storage unit, and an alarm unit. The display unit is used to intuitively present the pain detection and analysis results, EEG signal waveforms, and characteristic parameters in numerical, curve, and text form. The data storage unit is used to store the collected raw EEG signals, preprocessed signals, extracted characteristic parameters, and detection and analysis results, and supports data query and export. The alarm unit is used to issue an audible and visual alarm when a neuropathic pain signal is detected and the pain level reaches a preset threshold. The system calibration module includes a standard sample import unit, an error calculation unit, and a parameter optimization unit. The standard sample import unit is used to import standard EEG sample data with known pain types and pain states; the error calculation unit is used to calculate the deviation value between the current detection and analysis results and the standard sample data; the parameter optimization unit is used to dynamically adjust the parameters of the improved CNN-LSTM-AM deep learning model according to the deviation value using the gradient descent algorithm, so as to calibrate the system detection accuracy.
2. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The 32-channel EEG electrode array uses Ag / AgCl electrodes, and the electrode contact positions include the frontal lobe, parietal lobe, temporal lobe and occipital lobe regions. Conductive gel is placed between each electrode and the human scalp to ensure good contact between the electrode and the scalp. The sampling rate of the electrode array can be adjusted between 256Hz and 1024Hz.
3. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The improved adaptive noise cancellation algorithm constructs an interference signal model by collecting human electromyography reference signals and power frequency reference signals, and dynamically adjusts the filtering coefficients using an adaptive filtering algorithm to achieve accurate filtering of interference signals. The adjustment step size of the filtering coefficients is adaptively adjusted in real time according to the amplitude changes of the EEG signal.
4. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The preset time length of the signal segmentation unit is 1s-5s. The segmented data adopts an overlapping segmentation method with an overlap rate of 50%-70% to ensure the continuity of feature extraction and avoid loss of feature information.
5. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The frequency range of the θ band is 4Hz-7Hz, the frequency range of the α band is 8Hz-12Hz, the frequency range of the β band is 13Hz-29Hz, and the frequency range of the γ band is 30Hz-50Hz. The power spectral density characteristics of each band are obtained by calculating the power integral of the signal within that band.
6. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The training process of the improved CNN-LSTM-AM deep learning model includes: importing a dataset of EEG samples labeled with pain states and pain types, dividing the dataset into a training set, a validation set, and a test set; training the model using the training set, adjusting the model hyperparameters using the validation set, validating the model performance using the test set, and storing the model in the model storage unit after training to support online updates and iterations of the model.
7. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The system also includes a power supply module, which uses a rechargeable lithium battery with an output voltage of 3.3V-5V, supports a battery life of no less than 8 hours, and is equipped with a power management unit to realize charging protection, over-discharge protection and low battery warning functions.
8. The neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The display unit of the result output module adopts a touch LCD screen, which supports multiple interface switching and can display real-time waveforms of EEG signals, characteristic parameter curves, detection and analysis results, and historical data query interface respectively; the data storage unit adopts an SD card or solid-state drive with a storage capacity of not less than 16GB and supports data export via USB interface.
9. A neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The calibration cycle of the system calibration module can be set manually or triggered automatically. The automatic triggering condition is that after a preset number of consecutive tests, the detection error exceeds a preset threshold, triggering the automatic calibration process.
10. A neuropathic pain detection and analysis system based on electroencephalogram (EEG) signals according to claim 1, characterized in that: The modules communicate with each other via SPI or USB interfaces, with a transmission rate of no less than 1Mbps, ensuring the real-time performance and stability of data transmission. They are also equipped with a data encryption unit to encrypt the transmitted data and ensure data security.