Electroencephalogram signal real-time acquisition, analysis and processing system based on deep learning algorithm

Abstract

The invention relates to the technical field of electroencephalogram signal acquisition and analysis, and discloses an electroencephalogram signal real-time acquisition, analysis and processing system based on a deep learning algorithm, which comprises a hardware acquisition module, a data transmission module, a preprocessing module, a learning analysis module and a visualization and feedback module, through efficient cooperation of an ADS1299 daisy chain architecture of the hardware acquisition module and an STM32 master control, flexible expansion and synchronous acquisition of 32-128 channels are realized, a distributed power supply and a multi-layer PCB integrated design are combined, signal acquisition accuracy and hardware stability are improved while high channel number and miniaturization balance are ensured, and the system has a wide application prospect. And the requirements of clinical and scientific research on multi-scene electroencephalogram acquisition are met.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electroencephalogram signal acquisition and analysis, and more particularly to a real-time electroencephalogram signal acquisition, analysis and processing system based on a deep learning algorithm. BACKGROUND

[0002] An electroencephalogram (EEG) is a comprehensive reflection of the electrical activity of brain neurons on the scalp or cortical surface. By recording and analyzing these signals, the functional state of the brain can be revealed.

[0003] Although the prior art can achieve basic electroencephalogram signal acquisition and preliminary analysis, providing basic data support for brain science research and clinical diagnosis, the high-density acquisition devices of the prior art are mostly designed in an integrated manner, are bulky, and have a fixed number of channels, which are difficult to adapt to diversified scene requirements. Portable devices compromise on the number of channels and signal quality, and have insufficient interface flexibility, difficulty in function expansion, limited artifact removal effect, and lack of efficient deep learning analysis integration, resulting in that the data processing efficiency and analysis accuracy are difficult to meet the high-standard requirements of scientific research and clinical application. SUMMARY

[0004] The technical problem solved by the present application is to provide a real-time electroencephalogram signal acquisition, analysis and processing system based on a deep learning algorithm, which can solve the problems raised in the background.

[0005] To solve the above technical problems, according to one aspect of the present application, more specifically, a real-time electroencephalogram signal acquisition, analysis and processing system based on a deep learning algorithm comprises:

[0006] The hardware acquisition module realizes high-precision acquisition of electroencephalogram signals through a modular architecture, supports flexible expansion of 32-128 channels, ensures effective capture of weak electroencephalogram signals, realizes a balance between high channel number and miniaturization, and improves the adaptability and quality of signal acquisition; the data transmission module transmits the electroencephalogram signals acquired by the hardware acquisition module to subsequent modules in real time through a communication interface, supports wired and wireless transmission modes, meets the data transmission requirements in different scenarios, and ensures the real-time and stability of data transmission; the preprocessing module filters and denoises the transmitted raw electroencephalogram signals, eliminates interference signals, improves the signal-to-noise ratio of the electroencephalogram signals, and provides high-quality data for subsequent analysis; the learning analysis module analyzes the preprocessed electroencephalogram signals based on a deep learning algorithm, realizes extraction and classification identification of electroencephalogram signal features, improves the accuracy and efficiency of electroencephalogram signal analysis, and meets the analysis requirements; the visualization and feedback module displays the results of the learning analysis module in an intuitive manner and feeds back when an abnormality occurs, facilitating users to quickly understand the analysis results and timely obtain key information.

[0007] Further, the hardware acquisition module comprises: a signal acquisition module, a main control and control module, and a power supply and hardware integration module.

[0008] The signal acquisition module: adopts the ADS1299 chip to realize 32-128 channel EEG signal synchronous acquisition through daisy chain architecture, supports single-ended signal input, processes the EEG signal through the pre-filtering circuit to ensure the accuracy of signal acquisition.

[0009] The main control and control module: takes the STM32F103C8T6 as the main control chip, communicates with the signal acquisition module through the SPI bus, receives the DRDY pin interrupt signal to trigger data reading, and realizes the timing control and data processing of the whole acquisition process.

[0010] The power supply and hardware integration module: adopts a distributed power supply architecture to provide ±2.5V and ±3.3V multi-voltage domain output, realizes high-density integration of each module through a multi-layer PCB stacking architecture, controls the power supply node ripple and noise, and ensures stable operation of the hardware.

[0011] Further, the data transmission module comprises: a wireless transmission module, a wired and storage module;

[0012] The wireless transmission module: adopts the ESP8266 WiFi module and the Bluetooth module, is connected with the main control chip through the SPI interface, supports the TCP / IP protocol and the hotspot mode, realizes the wireless real-time transmission of the EEG data, and meets the data transmission demand in the mobile scene.

[0013] The wired and storage module: is integrated with the USB interface based on the CH340G chip and is equipped with a standard SD card storage interface, supports high-speed wired data transmission and 16 hexadecimal raw data offline storage, and guarantees the diversity and integrity of data transmission.

[0014] Further, the preprocessing module comprises: a filtering and noise reduction module, an artifact removal module, and a feature extraction module.

[0015] The filtering and noise reduction module: adopts the IIR and FIR digital filtering algorithm, contains the 50Hz power frequency notch filter and the 20-500Hz band-pass filter, and removes the environmental noise and the power frequency interference.

[0016] The artifact removal module: separates and eliminates the electrooculogram and electromyogram artifact signals through the independent component analysis algorithm, and improves the purity of the EEG signal.

[0017] The feature extraction module: extracts the time domain features and the frequency domain features of the EEG signal, and provides basic feature data for subsequent deep learning analysis.

[0018] Further, the learning analysis module comprises: a model construction module and a real-time inference module.

[0019] The model construction module: based on MATLAB, a CNN-LSTM hybrid model is constructed, the preprocessed electroencephalogram data is used for model training, and the model parameters are optimized to improve the classification accuracy.

[0020] The real-time inference module: receiving the preprocessed real-time electroencephalogram signal segment, performing inference calculation through the trained model, outputting the signal classification result, and ensuring that the inference delay is less than 50ms.

[0021] Further, the visualization and feedback module comprises: a real-time display module, a storage and alarm module;

[0022] The real-time display module: based on the MATLAB platform, 32-128 channel electroencephalogram waveforms are displayed in parallel, time baseline and amplitude scaling adjustment are supported, real-time FFT spectrum analysis function is integrated, and signal characteristics are intuitively presented;

[0023] The storage and alarm module: the original signal and analysis result are stored in EDF format or binary file, and when the analysis result is abnormal, an audible and visual alarm is triggered to ensure timely warning of abnormal signals.

[0024] Further, the system supports automatic identification and driving adaptation of the expansion interface, realizes the function module "plug and play" through embedded software, and reduces the expansion adaptation cost.

[0025] Further, the hardware acquisition module adopts an anti-interference design of single-point connection of analog ground and digital ground, cooperates with key signal line ground processing, and reduces the influence of electromagnetic interference on weak electroencephalogram signals.

[0026] The beneficial effects of the electroencephalogram signal real-time acquisition and analysis processing system based on deep learning algorithm are:

[0027] Through the efficient cooperation of the ADS1299 daisy chain architecture of the hardware acquisition module and the STM32 master control, 32-128 channel flexible expansion and synchronous acquisition are realized, combined with distributed power supply and multi-layer PCB integrated design, while ensuring high channel number and miniaturization balance, the accuracy and hardware stability of signal acquisition are improved, meeting the needs of clinical and scientific research for multi-scene electroencephalogram acquisition.

[0028] Through the algorithms such as digital filtering and independent component analysis of the preprocessing module to eliminate noise and artifacts, combined with the real-time inference of the CNN-LSTM model of the learning and analysis module, and the dynamic display and abnormal alarm of the visualization and feedback module, the efficiency and accuracy of electroencephalogram signal analysis are greatly improved, providing reliable support for clinical diagnosis and scientific research of electroencephalogram signals. BRIEF DESCRIPTION OF DRAWINGS

[0029] The application will be described in further detail below with reference to the drawings and specific implementation methods.

[0030] Figure 1 It is a schematic diagram of system principle;

[0031] Figure 2 It is a schematic diagram of function framework;

[0032] Figure 3 It is a schematic diagram of control circuit principle;

[0033] Figure 4 It is a schematic diagram of power management circuit part design. DETAILED DESCRIPTION

[0034] The application will be described in further detail below with reference to the drawings and specific implementation methods. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0035] According to one aspect of the present application, as shown in the accompanying drawings, a real-time EEG signal acquisition and analysis processing system based on deep learning algorithm is provided, comprising: Figures 1-4 A hardware acquisition module realizes high-precision acquisition of EEG signals through a modular architecture, supports flexible expansion of 32-128 channels, ensures effective capture of weak EEG signals, realizes a balance between high channel number and miniaturization, and improves the adaptability and quality of signal acquisition. The module includes:

[0036] A signal acquisition module: ADS1299 chips are cascaded through a daisy chain architecture to realize 32-128 channel EEG signal synchronous acquisition, support single-ended signal input, process EEG signals through a pre-filtering circuit to ensure the accuracy of signal acquisition;

[0037] Among them, the ADS1299 chip adopts a single-ended signal input mode, the IN1P to IN8P pins are connected to the EEG signal output by the pre-filtering circuit, and the SRB1 pin is used as the reference pin of the 8 channels;

[0038] And the VREFP and VREFN pins are connected in parallel through 10uF and 0.1uF capacitors, and the VREFP pin is directly connected to AVSS;

[0039] In addition, the CLKSEL pin is grounded, and by default, an external 2.048MHz active crystal oscillator is used to provide a clock signal. When multiple ADS1299 chips are cascaded through a daisy chain architecture, the DOUT pin of the previous chip is connected to the DAISY_IN pin of the next chip, only the first DRDY output pin and the BIA amplifier power supply are reserved, and the rest are floating, ensuring that all chip data can be read through one SPI interface (the function block diagram is shown in the accompanying

[0040] Figure 2 ​​

[0041] Master and control module: STM32F103C8T6 as the master chip, communicates with the signal acquisition module through the SPI bus, receives the DRDY pin interrupt signal to trigger data reading, realizes the timing control and data processing of the whole acquisition process;

[0042] Among them, when the STM32F103C8T6 master chip is connected with the ADS1299 through the SPI bus, the clock line is connected with a terminal resistor to suppress signal reflection, the data line is laid with equal length (deviation controlled within ±5mm), equipped with 8MHz and 2.048MHz dual crystal oscillator clock system, providing stable timing reference for the acquisition process, integrating standard SWD debugging interface, facilitating system development and maintenance (its control circuit principle diagram is shown in the attached Figure 3

[0043] Power supply and hardware integration module: adopts distributed power architecture, provides multi-voltage domain output of ±2.5V and ±3.3V, realizes high-density integration of each module through multi-layer PCB stacking architecture, controls power node ripple and noise, and guarantees stable operation of hardware;

[0044] Among them, the front end of the distributed power architecture adopts TP4056 lithium battery management chip to realize constant current / constant voltage charging of 4.2V lithium battery, 3.3V digital power supply is realized by two-stage voltage stabilization of TLV76733 synchronous step-down converter and LP5907MFX-3.3 low dropout regulator, ±2.5V analog power supply is realized by cooperation of TPS72325 (positive 2.5V), LM2776 charge pump inverter (generates-3.3V) and TLV70025 (stabilizes to-2.5V), multi-layer PCB adopts layered copper design, including digital ground, analog ground and power ground, realizing isolation of power network and ground plane (the power management circuit part design is shown in the attached Figure 4

[0045] Through such setting, the balance of high channel number and miniaturization is realized, ensuring the accuracy and stability of weak brain electrical signal acquisition, and guaranteeing the real-time and integrity of data transmission in different scenarios.

[0046] Data transmission module, the brain electrical signal acquired by the hardware acquisition module is transmitted to the subsequent module in real time through the communication interface, supporting wired and wireless transmission methods, meeting the data transmission requirements in different scenarios, ensuring the real-time and stability of data transmission. This module includes:

[0047] Wireless transmission module: adopts ESP8266 WiFi module and Bluetooth module, connected with the master chip through SPI interface, supports TCP / IP protocol and hotspot mode, realizes wireless real-time transmission of brain electrical data, meets the data transmission requirements in mobile scenarios;​​

[0048] Among them, the ESP8266 WiFi module supports both standalone operation in hotspot mode and access to existing network environments. It is connected to the STM32 main controller through the SPI high-speed interface, and its transmission bandwidth can meet the real-time transmission needs of 32-channel EEG data.

[0049] The Bluetooth module is designed with low power consumption and is suitable for short-range connection of mobile devices, enabling convenient data interaction.

[0050] Wired and storage module: based on CH340G chip integrated USB interface and equipped with standard SD card storage interface, supporting high-speed wired data transmission and 16 hexadecimal raw data offline storage, ensuring the diversity and integrity of data transmission;

[0051] Among them, the USB interface based on CH340G chip can realize stable connection with the host computer, ensuring the reliability of data transmission;

[0052] The standard SD card storage interface supports offline data collection and can be used as data cache when the wireless transmission bandwidth is insufficient. The stored 16 hexadecimal raw data format is compatible with the host computer software, facilitating subsequent analysis and processing.

[0053] Through the cooperation of the wireless transmission module and the wired and storage module, the flexibility and stability of data transmission in different scenarios are realized, ensuring the complete flow of EEG signals from collection to analysis.

[0054] Preprocessing module: filters and denoises the transmitted raw EEG signals, eliminates interference signals, and improves the signal-to-noise ratio of EEG signals, providing high-quality data for subsequent analysis. This module includes:

[0055] Filtering and noise reduction module: uses IIR and FIR digital filtering algorithms, including 50Hz power frequency notch filtering and 20-500Hz bandpass filtering, to remove environmental noise and power frequency interference;

[0056] Among them, the IIR digital filtering algorithm is designed through the MATLAB signal processing toolbox, supporting user-defined filtering parameters to adapt to different noise scenarios;

[0057] The FIR digital filtering algorithm uses linear phase design to avoid signal phase distortion and ensure that the timing characteristics of the filtered EEG signals are not affected;

[0058] In addition, the 50Hz power frequency notch filter specifically suppresses power interference, with an attenuation effect of more than -40dB.

[0059] Artifact removal module: uses independent component analysis algorithm to separate and eliminate electrooculogram and electromyogram artifact signals, improving the purity of EEG signals;

[0060] Among them, the independent component analysis algorithm is realized based on the ICA toolbox of MATLAB, and the electrooculogram (EOG) and electromyogram (EMG) and other artifact signals are decomposed into independent components by blind source separation of multi-channel electroencephalogram signals;

[0061] And the operator can manually mark the artifact components and remove them through the visual interface, or automatically identify and remove the artifacts through the preset feature threshold;

[0062] In addition, the algorithm supports batch processing mode, which can remove artifacts in batches for historical data, improving processing efficiency.

[0063] Feature extraction module: extract the time domain features and frequency domain features of the electroencephalogram signal, and provide basic feature data for subsequent deep learning analysis;

[0064] Among them, the time domain features include the mean, variance, peak value, kurtosis and other statistical quantities of the signal, reflecting the amplitude change characteristics of the electroencephalogram signal;

[0065] And the frequency domain features are obtained by FFT transformation, including the energy value and power spectral density of the alpha band (8-13Hz), beta band (13-30Hz), theta band (4-8Hz) and gamma band (30-100Hz), corresponding to different functional states of the brain;

[0066] In addition, the feature extraction module also supports the calculation of time-frequency domain joint features (such as wavelet transform coefficients), providing more rich input dimensions for the deep learning model.

[0067] It realizes the deep purification and feature mining of the original electroencephalogram signal, provides high-quality input data for the learning analysis module, and lays the foundation for accurate analysis.

[0068] Learning analysis module, based on deep learning algorithm to analyze the preprocessed electroencephalogram signal, realize the extraction and classification recognition of electroencephalogram signal characteristics, improve the accuracy and efficiency of electroencephalogram signal analysis, meet the analysis demand. This module includes:

[0069] Model construction module: based on MATLAB to build CNN-LSTM hybrid model, use preprocessed electroencephalogram data to train the model, optimize the model parameters to improve the classification accuracy;

[0070] Among them, the CNN part is responsible for extracting the spatial features of the electroencephalogram signal (such as the correlation of signals in different channels), and the LSTM part focuses on capturing the time domain dynamic characteristics (such as the signal time sequence change during the seizure);

[0071] And the model training adopts 5-fold cross-validation, adjusts the learning rate, iteration number and other parameters to optimize the classification accuracy, and the goal is to improve the accuracy of the validation set;

[0072] In addition, the training data covers normal EEG, interictal / ictal, sleep stages, and other multi-scene scenarios to ensure model generalization.

[0073] Real-time inference module: receives pre-processed real-time EEG signal segments, performs inference calculation through the trained model, and outputs signal classification results, ensuring an inference delay of <50ms;

[0074] Among them, real-time inference uses a sliding window mechanism (e.g., window length 5-10s, step 1-2s) to segment continuous EEG signals;

[0075] The model is converted to C language code using MATLAB Coder tool and integrated into an embedded system for accelerated inference;

[0076] In addition, the system outputs the consistent results of the last three windows after confirmation, reducing the probability of misjudgment.

[0077] This realizes a closed loop from offline model training to online real-time analysis, greatly improving the accuracy and timeliness of EEG signal classification, and providing an efficient tool for clinical diagnosis and scientific analysis.

[0078] The visualization and feedback module displays the results of the learning analysis module in an intuitive way and provides feedback when abnormalities occur, making it easy for users to quickly understand the analysis results and obtain key information in a timely manner. This module includes:

[0079] Real-time display module: based on the MATLAB platform, it realizes the parallel display of 32-128 channel EEG waveforms, supports time baseline and amplitude scaling adjustment, and integrates real-time FFT spectrum analysis function to intuitively present signal characteristics;

[0080] Among them, the channel waveforms are coded with different colors, and the abnormal signal segments (such as epileptic discharges) are automatically marked in red for easy and fast positioning;

[0081] The time baseline can be adjusted within 1s-60s, and the amplitude can be scaled from ±50μV to ±500μV to meet the observation needs of different signal intensities;

[0082] In addition, the FFT spectrum analysis results are displayed synchronously with the waveforms, and the energy proportion of each frequency band can be dynamically updated to help users correlate time and frequency domain characteristics.

[0083] Storage and alarm module: stores the original signal and analysis results in EDF format or binary file, and triggers sound and light alarms when the analysis results are abnormal to ensure timely warning of abnormal signals;

[0084] The EDF format file contains complete patient information, acquisition parameters and signal annotation, is compatible with mainstream electroencephalogram analysis software (such as EEGLAB), and is convenient for data sharing and secondary analysis.

[0085] The sound-light alarm is realized through a buzzer (frequency 2 kHz) and a red LED lamp, the alarm threshold can be customized according to clinical requirements, and hierarchical alarm (such as mild abnormality and severe abnormality) is supported.

[0086] In addition, the system automatically records the alarm time point and the corresponding signal segment to form an alarm log, which is convenient for subsequent review and diagnosis reference.

[0087] Both of them cooperate with each other, realize the intuitive presentation of the electroencephalogram signal, guarantee the timely response of the abnormal situation, and improve the clinical practicability of the system.

[0088] Of course, the above description is not a limitation of the present application, and the present application is not limited to the above examples. Changes, modifications, additions or replacements made by ordinary skilled in the art within the essential scope of the present application also belong to the protection scope of the present application.

Claims

1. A real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms, characterized in that: include: The hardware acquisition module achieves high-precision acquisition of EEG signals through a modular architecture, supports flexible expansion from 32 to 128 channels, ensures effective capture of weak EEG signals, achieves a balance between high channel count and miniaturization, and improves the adaptability and quality of signal acquisition. The data transmission module transmits the EEG signals acquired by the hardware acquisition module to subsequent modules in real time through a communication interface, supports wired and wireless transmission methods, meets the data transmission needs of different scenarios, and ensures the real-time performance and stability of data transmission. The preprocessing module filters and denoises the transmitted raw EEG signals, eliminates interference signals, improves the signal-to-noise ratio of the EEG signals, and provides high-quality data for subsequent analysis. The learning and analysis module analyzes the preprocessed EEG signals based on deep learning algorithms, enabling the extraction and classification of EEG signal features, thereby improving the accuracy and efficiency of EEG signal analysis and meeting analysis requirements. The visualization and feedback module displays the results of the learning and analysis module in an intuitive way and provides feedback when anomalies occur, making it convenient for users to quickly understand the analysis results and obtain key information in a timely manner.

2. The real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The hardware acquisition module includes: a signal acquisition module, a main control and control module, and a power supply and hardware integration module; Signal acquisition module: The ADS1299 chip is cascaded in a daisy-chain architecture to achieve synchronous acquisition of 32-128 channels of EEG signals. It supports single-ended signal input and processes EEG signals through a pre-filter circuit to ensure the accuracy of signal acquisition. Main control and control module: The STM32F103C8T6 is used as the main control chip. It communicates with the signal acquisition module through the SPI bus, receives the interrupt signal of the DRDY pin to trigger data reading, and realizes the timing control and data processing of the entire acquisition process. Power supply and hardware integration module: Adopting a distributed power architecture, it provides multi-voltage domain output of ±2.5V and ±3.3V. Through a multi-layer PCB stacking architecture, it achieves high-density integration of each module, controls power node ripple and noise, and ensures stable hardware operation.

3. The real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The data transmission module includes: a wireless transmission module and a wired and storage module; Wireless transmission module: It adopts ESP8266 WiFi module and Bluetooth module, and connects to the main control chip through SPI interface. It supports TCP / IP protocol and hotspot mode to realize wireless real-time transmission of EEG data and meet the data transmission needs in mobile scenarios. Wired and storage module: Based on the CH340G chip, it integrates a USB interface and is equipped with a standard SD card storage interface, supporting high-speed wired data transmission and offline storage of hexadecimal raw data, ensuring the diversity and integrity of data transmission.

4. The real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The preprocessing module includes: a filtering and noise reduction module, an artifact removal module, and a feature extraction module; Filtering and noise reduction module: It adopts IIR and FIR digital filtering algorithms, including 50Hz power frequency notch filtering and 20-500Hz bandpass filtering to remove environmental noise and power frequency interference; Artifact Removal Module: Separates and eliminates electrooculography (EOG) and electromyography (EMG) artifact signals through independent component analysis algorithms, thereby improving the purity of EEG signals; Feature extraction module: Extracts time-domain and frequency-domain features of EEG signals to provide basic feature data for subsequent deep learning analysis.

5. The real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The learning and analysis module includes: a model building module and a real-time inference module; Model building module: Based on MATLAB, a CNN-LSTM hybrid model is built, and preprocessed EEG data is used for model training to optimize model parameters and improve classification accuracy; Real-time inference module: Receives pre-processed real-time EEG signal segments, performs inference calculations using a trained model, and outputs signal classification results, ensuring inference latency <50ms.

6. The real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The visualization and feedback module includes: a real-time display module and a storage and alarm module; Real-time display module: Based on the MATLAB platform, it realizes parallel display of 32-128 channels of EEG waveforms, supports adjustment of time baseline and amplitude scaling, integrates real-time FFT spectrum analysis function, and intuitively presents signal characteristics; Storage and alarm module: Stores raw signals and analysis results in EDF format or binary files. When an abnormality occurs in the analysis results, it triggers an audible and visual alarm to ensure timely warning of abnormal signals.

7. The real-time acquisition, analysis, and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The system supports automatic identification and driver adaptation of extended interfaces, and realizes "plug and play" of functional modules through embedded software, reducing the cost of extension adaptation.

8. The real-time acquisition, analysis and processing system for electroencephalogram (EEG) signals based on deep learning algorithms according to claim 1, characterized in that: The hardware acquisition module adopts an anti-interference design with single-point connection between analog ground and digital ground, and is combined with grounding processing of key signal lines to reduce the impact of electromagnetic interference on weak EEG signals.

Images