Stroke risk assessment system and method, storage medium

Abstract

The application discloses a stroke risk assessment system and method and a storage medium. After scalp electroencephalogram signals are picked up through an electroencephalogram cap module, the signals are transmitted to a 32-channel electroencephalogram signal amplifier module through lead wires, signal amplification, filtering and ADC conversion are completed by the 32-channel electroencephalogram signal amplifier module, and then the signals are transmitted to an upper computer data acquisition visualization and risk assessment software module through a wireless mode after digital processing and packaging. The upper computer data acquisition visualization and risk assessment software module receives data and performs real-time visual display, extracts stroke electroencephalogram features, inputs a stroke risk analysis model and finally outputs a risk grade assessment result. According to the technical scheme, the accuracy of stroke risk assessment is significantly improved by combining precise electroencephalogram signal feature extraction with a machine learning algorithm.

Description

Technical Field

[0001] This invention belongs to the field of electroencephalogram (EEG) signal processing technology, specifically relating to a stroke risk assessment system and method, and a storage medium. Background Technology

[0002] Stroke, a prevalent neurovascular disease, requires early identification and research into its pathological mechanisms for clinical diagnosis and rehabilitation intervention. Electroencephalography (EEG) signals, as a direct reflection of the electrophysiological activity of the cerebral cortex, have become a crucial technique in stroke-related research. Currently, existing technologies for studying the EEG characteristics of stroke patients employ classic signal analysis methods such as independent component analysis (ICA) and power spectral density (PSD). By creating EEG topographic maps and power spectral density maps, the visualization and quantitative analysis of EEG characteristics have been achieved, confirming that stroke patients exhibit characteristics such as enhanced slow waves, diminished fast waves, and asymmetrical distribution of cortical electrical activity. Furthermore, current technologies also incorporate multi-dimensional data (EEG signals, ECG signals, demographic information, and clinical history) to conduct correlation studies of stroke-related EEG activity.

[0003] In terms of EEG signal acquisition, existing technologies mostly employ electrode cap acquisition devices based on the international 10-20 system layout. The mainstream is a multi-channel silver / silver chloride (Ag / AgCl) electrode cap, with sampling rates covering 250Hz and above. Artifacts and noise in the EEG signals are removed through preprocessing methods such as filtering, notch filtering, and normalization to achieve signal purification. In terms of experimental design, basic EEG acquisition paradigms such as resting state and motor stimulation state have been established. By comparing the EEG characteristics of stroke patients and healthy individuals, the functional changes of the cerebral cortical neural network after stroke can be preliminarily explored.

[0004] In the screening and grouping of samples for stroke EEG research, existing technologies have incorporated tools such as neurological function scoring scales (e.g., NIHSS scale), cognitive assessment scales (e.g., Montreal Cognitive Assessment Scale), and hand-handling assessment scales (e.g., Edinburgh Handedness Scale) for subject inclusion and exclusion. Simultaneously, imaging techniques such as computed tomography (CT) scans are used to confirm stroke lesions, ensuring the validity of the research samples. Furthermore, existing technologies have attempted to combine ICA topographic mapping with PSD analysis to distinguish brain-derived signals from artifacts such as eye movements and electromyography (EMG), thereby extracting stroke-related specific EEG features.

[0005] Meanwhile, visualization tools for EEG analysis in existing technologies have been maturely applied. EEG topography can display the spatial distribution of electrical activity on the scalp surface through interpolation calculation and color coding. Power spectral density maps can quantitatively analyze the power distribution of EEG signals in each frequency band of δ (0.5–4 Hz), θ (4–8 Hz), α (8–13 Hz), β (13–30 Hz), and γ (>30 Hz), providing a quantitative means for identifying abnormal EEG frequency characteristics in stroke patients.

[0006] Although existing technologies have made some progress in the study of EEG characteristics in stroke, there are still many technical shortcomings and application pain points that urgently need to be addressed, in light of the needs of clinical application and in-depth research. These are as follows: 1. Insufficient scene adaptability of EEG acquisition equipment: Most of the current mainstream EEG acquisition equipment are large desktop devices with poor portability, which cannot meet the needs of primary medical institutions (counties and communities) and bedside on-site acquisition during the acute phase of stroke. Only a few portable devices have problems with insufficient number of channels and low signal acquisition accuracy, making it difficult to record the electrical activity of the entire cerebral cortex.

[0007] 2. Gaps and lack of standardization in early EEG research after stroke: Existing research mainly focuses on EEG characteristics during the recovery period of stroke, with very few studies on early EEG characteristics within 96 hours of the first ischemic stroke. Furthermore, there is a lack of standardized subject inclusion and exclusion criteria and EEG acquisition paradigms suitable for this acute phase, resulting in insufficient specific data on early stroke EEG characteristics and an inability to provide electrophysiological evidence for early clinical identification.

[0008] 3. Low standardization in EEG acquisition and analysis: In existing experimental paradigms, the acquisition of motor-stimulated EEG lacks a unified standard for the movement process, duration, and number of repetitions, resulting in poor data stability and reproducibility. At the same time, the combined IC+PSD analysis after EEG signal preprocessing has not formed a standardized feature extraction system for ischemic stroke, making it difficult to achieve accurate screening and quantification of stroke-related EEG features.

[0009] 4. Insufficient sample design and data support: Existing EEG studies on stroke have small sample sizes, lack scientific statistical calculations (such as G-power sample size estimation) for matching stroke patients with healthy controls, and have poor matching in terms of gender and age. In addition, there is a lack of specific studies targeting the high-risk population of stroke aged 40-70, resulting in low generalizability and reliability of the research results, and failing to establish an effective data foundation for EEG risk grading and risk profiling of stroke.

[0010] 5. Poor anti-interference and operational adaptability of grassroots data collection environments: The existing EEG acquisition process lacks standardized environmental requirements, and there are no clear standards for temperature, humidity, and electromagnetic interference control in grassroots settings. Furthermore, the procedures for wearing electrode caps and adjusting impedance are cumbersome and lack standardized operating procedures suitable for grassroots medical staff, which can easily lead to artifacts and data distortion in the acquired EEG signals.

[0011] 6. Insufficient connection between EEG characteristics and clinical applications: Current combined ICA and PSD analysis only focuses on identifying EEG features and does not establish a direct link between these features and neurological functional impairment and rehabilitation prognosis after stroke. Therefore, it cannot provide targeted electrophysiological evidence for subsequent stroke rehabilitation interventions and the clinical application of brain-computer interfaces. Summary of the Invention

[0012] To address the problems existing in the prior art, the present invention provides a stroke risk assessment system and method, and a storage medium.

[0013] To achieve the above objectives, the present invention provides the following solution: A stroke risk assessment system includes: an EEG cap module, a 32-channel EEG signal amplifier module, and a host computer data acquisition, visualization, and risk assessment software module; wherein, After the scalp EEG signals are picked up by the EEG cap module, they are transmitted to the 32-channel EEG signal amplifier module via lead wires. The 32-channel EEG signal amplifier module simulates the front end to complete signal amplification, filtering, and ADC conversion. After digital processing and encapsulation, the signals are then transmitted wirelessly to the host computer data acquisition, visualization, and risk assessment software module. The host computer data acquisition, visualization, and risk assessment software module receives the data and displays it in real time. At the same time, it extracts stroke EEG features, inputs them into the stroke risk analysis model, and finally outputs the risk level assessment result.

[0014] Preferably, the 32-channel EEG signal amplifier adopts a multi-layer high-density printed circuit board design, with digital circuits as the main architecture, integrating a high-performance core processing unit, multiple signal conditioning channels, wireless communication interface and power management system.

[0015] As a preferred option, the EEG cap module is made of elastic fabric material. The cap surface has 32 pre-set electrode mounting positions according to the international 10-20 system standard, covering the frontal, central, parietal, temporal, and occipital regions, with additional reference and ground electrode positions.

[0016] This invention also provides a method for assessing stroke risk, comprising: Collecting electroencephalogram (EEG) signals from stroke patients; Extracting EEG features from stroke EEG signals; Based on the extracted EEG features of stroke, a stroke risk analysis model was constructed. The 32-channel EEG signals collected in real time are input into the stroke risk analysis model, and the final risk level assessment result is output.

[0017] As a preferred approach, the extracted EEG features for stroke include: frequency band and energy features, entropy and complexity features, and statistical and time-frequency features.

[0018] Preferably, the frequency band and energy characteristics include: the power spectral density of each standard frequency band, the absolute power and relative power of each frequency band, and the power ratio of the frequency bands.

[0019] As preferred options, entropy and complexity features include: sample entropy, fuzzy entropy, permutation entropy, multi-scale entropy, and Lempel-Ziv complexity features.

[0020] Preferably, the statistical and time-frequency features include: time-domain statistics and time-frequency joint analysis features of the EEG signal; wherein, the time-domain statistics include mean, variance, skewness, kurtosis and zero-crossing rate; the time-frequency joint analysis features are based on short-time Fourier transform, wavelet transform or Hilbert-Huang transform to extract time-frequency energy distribution, instantaneous frequency and instantaneous amplitude.

[0021] The present invention also provides a stroke risk assessment system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program performs a stroke risk assessment method when executed by the processor.

[0022] The present invention also provides a storage medium storing a computer program that executes a stroke risk assessment method when running.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Comparison with existing EEG acquisition devices: Most current mainstream EEG acquisition devices are large desktop devices, lacking portability and unable to meet the needs of primary healthcare institutions and bedside acquisition during the acute phase. The 32-channel portable EEG device of this invention can easily adapt to different acquisition environments while ensuring high sampling rate and accuracy. Compared with existing devices, this invention optimizes the balance between portability and acquisition accuracy, and can meet the needs of emergency and early stroke detection in primary hospitals.

[0024] 2. Comparison with Existing EEG Feature Extraction Methods: Traditional EEG analysis methods mainly rely on simple frequency domain analysis and pay little attention to the spatial distribution and temporal characteristics of EEG. In contrast, this invention provides a more comprehensive EEG feature analysis by combining multi-dimensional features such as band power, entropy value, and time-frequency characteristics, making the identification of early stroke EEG abnormalities more sensitive. In clinical applications, this invention can effectively identify EEG abnormalities that existing methods have failed to detect, significantly improving the sensitivity of early diagnosis.

[0025] 3. Comparison with Existing Stroke Risk Assessment Methods: Traditional stroke risk assessment methods typically rely on clinical symptoms and imaging examinations, but these methods often have significant errors in the early stages of stroke. This invention, through real-time analysis of electroencephalograms (EEGs), can promptly identify abnormal changes during the acute phase of stroke, providing clinicians with a more accurate risk assessment tool. Compared to traditional assessment methods, this invention can identify stroke risk before the onset of clinical symptoms and provides strong support for early intervention. Attached Figure Description

[0026] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the structure of a stroke EEG risk assessment system; Figure 2 This is a front layout diagram of the PCB for a 32-channel EEG signal amplifier. Figure 3 This is a layout diagram of the back of the PCB for a 32-channel EEG signal amplifier. Figure 4 A schematic diagram showing the functional module division of the amplifier's core circuit; Figure 5 This is a picture of the EEG amplifier after the outer casing has been added. Figure 6 This is a topological diagram of the EEG cap structure and electrode distribution; Figure 7 This is a schematic diagram showing how the amplifier is connected to the EEG cap; Figure 8 Screenshot of the main interface of LabVIEW data acquisition software; Figure 9 Screenshot of the real-time waveform display interface for 32-channel EEG signals; Figure 10 Screenshot of the data storage and file management interface; Figure 11Flowchart of experimental paradigm for EEG acquisition in stroke; Figure 12 A flowchart of the overall method for EEG risk assessment of stroke; Figure 13 A framework diagram for extracting EEG features from three types of stroke; Figure 14 Flowchart for constructing a stroke risk analysis model.

[0028] Among them, 100 is the EEG cap module; 200 is the 32-channel EEG signal amplifier module; 300 is the host computer data acquisition visualization and risk assessment software module; U5 is the core processor; U1, U2, U3, U4, U6, U9, and U10 are auxiliary integrated circuits; P1 and P2 are pin header interfaces; X1 is the standard connector mounting position; Y1 is the crystal oscillator; C27 and C28 are the load capacitors; LED1 is the status indicator; SW1 is the push-button switch; and D1 is the diode. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Example 1 like Figure 1 As shown, this invention provides a stroke risk assessment system, comprising: an EEG cap module 100, a 32-channel EEG signal amplifier module 200, and a host computer data acquisition, visualization, and risk assessment software module 300; wherein, After the scalp EEG signals are picked up by the EEG cap module, they are transmitted via lead wires to the 32-channel EEG signal amplifier module. The 32-channel EEG signal amplifier module simulates the front end to complete signal amplification, filtering, and ADC conversion. After digital processing and encapsulation, the signals are transmitted wirelessly to the host computer data acquisition, visualization, and risk assessment software module. The host computer data acquisition, visualization, and risk assessment software module receives the data and displays it in real time. After acquisition, the data is saved in TXT format. The saved data is used to extract stroke EEG features, which are then input into the stroke risk analysis model, and finally, the risk level assessment result is output.

[0032] As one embodiment of the present invention, such as Figure 6 and Figure 7 As shown, the EEG cap is an elastic EEG cap used in conjunction with the amplifier module 200.

[0033] Furthermore, the EEG cap is made of elastic fabric, with 32 pre-set electrode mounting positions on its surface according to the international 10-20 system standard, covering the frontal, central, parietal, temporal, and occipital regions, along with additional reference and ground electrodes. The cap is available in various head circumference sizes and can be adapted to different subjects via adjustable straps.

[0034] The standard positions corresponding to the 32-channel electrodes include Fp1, Fp2, AF3, AF4, F7, F3, Fz, F4, F8, FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, CP5, CP1, CP2, CP6, P7, P3, Pz, P4, P8, PO3, PO4, O1, Oz, and O2.

[0035] Furthermore, the electrodes are preferably silver / silver chloride sintered disc electrodes. During use, a special conductive gel for EEG needs to be injected between the electrodes and the subject's scalp to reduce the contact impedance between each channel electrode and the scalp to below 5 kiloohms, thus ensuring the quality of EEG signal pickup.

[0036] Furthermore, the EEG cap connects to the amplifier's high-density signal input connector via a multi-core shielded lead cable. The cable has good shielding performance and flexibility, and is of moderate length, so it does not affect the subject's basic activities during the experiment.

[0037] As one embodiment of the present invention, such as Figure 2 , Figure 3 and Figure 4 As shown, the 32-channel EEG signal amplification module adopts a multi-layer high-density printed circuit board design, with digital circuits as the main architecture, and integrates a high-performance core processing unit, multiple signal conditioning channels, wireless communication interface and a complete power management system.

[0038] The PCB is made of FR-4 epoxy fiberglass board with a minimum of four layers. A reasonable stacking design balances signal integrity and electromagnetic compatibility. The board has four corner mounting holes and a center mounting hole to securely fix the PCB inside the chassis, ensuring the system's mechanical stability during clinical use and handling.

[0039] The overall board layout strictly adheres to the principle of functional zoning. The core processor area, analog signal conditioning area, power management area, interface connector area, and clock area are each independently zoned. High-speed digital traces employ impedance matching design, and traces between critical signal lines maintain reasonable spacing to suppress crosstalk.

[0040] Furthermore, the core chip on the board is U5, which uses BGA packaging and is located in the center of the PCB board. It is surrounded by a large number of decoupling capacitors (including C93, C94, etc.) to form a multi-level high-frequency decoupling network, ensuring the power integrity and signal stability of the chip when it is running at high speed.

[0041] The U5 chip is a high-performance processor that undertakes the core computing and control tasks of the system, specifically including: coordinating and controlling the synchronous sampling timing of the 32-channel analog front-end signals; aggregating, buffering, and performing preliminary processing on the digital EEG data converted by the 32-channel ADC; performing data frame encapsulation and communication protocol management; controlling the wireless communication module to achieve real-time data transmission with the host computer; and completing system status management, including power management, lead detachment detection, and device self-test.

[0042] The decoupling capacitor network around U5 employs a three-stage decoupling strategy: near, mid, and far. A 100nF high-frequency ceramic capacitor in a 0402 package is placed close to the BGA pad, a 1µF mid-frequency decoupling capacitor is placed approximately 3 to 5 mm from the chip, and a 10µF low-frequency energy storage capacitor is placed approximately 10 mm from the chip, covering the power supply noise suppression requirements across the entire frequency band from MHz to kHz.

[0043] Furthermore, surrounding the core processor U5, the board also houses multiple auxiliary integrated circuits, including U1, U2, U3, U4, U6, U9, and U10, all in QFP packages with numerous and neatly arranged pins. Each chip plays a functional role in the system, including the following aspects.

[0044] In terms of analog front-end signal conditioning, some ICs are responsible for low-noise amplification, anti-aliasing filtering, and high-precision analog-to-digital conversion of 32 channels of EEG signals. The chip integrates a programmable gain amplifier, and the gain can be configured via register programming in the range of 1 to 24 times. The default setting is 24 times gain to match the typical amplitude range of EEG signals (1 microvolt to 100 microvolts).

[0045] In terms of interface conversion and driving, some ICs are used to implement the conversion of digital interface protocols such as USB, SPI, and I2C, as well as the driving control of wireless communication modules.

[0046] In terms of memory control, some ICs are responsible for the read and write control of external memory, providing sufficient storage depth for data caching.

[0047] Multiple signal conditioning chips are connected to the core processor U5 via cascading or parallel bus to achieve synchronous data transmission and unified clock control for 32 channels, eliminating sampling phase differences between channels.

[0048] Furthermore, the PCB is equipped with a variety of interface connectors to meet the electrical interconnection needs between the system and external devices.

[0049] The high-density dual-row pin connector on the left, composed of numerous circular solder pads, serves as the system's main signal input interface for connecting the EEG cap's lead cables. This interface carries 32 EEG signal input channels, as well as reference electrode and ground electrode signals. The interface is compactly arranged with uniform pin spacing, ensuring the orderly introduction of multi-channel signals. The connector employs a gold-plated contact design, with a contact resistance of less than 20 milliohms, guaranteeing the reliability of weak EEG signal transmission.

[0050] P1 and P2 are small pin header interfaces used as debugging interfaces, program burning interfaces, or peripheral expansion ports. During the development phase, the P1 interface can be used to connect a JTAG or SWD debugger to perform firmware debugging and upgrades on the core processor. The P2 interface can be used for external trigger signal input, in conjunction with experimental paradigm synchronization markers.

[0051] X1 is an interface mounting position with mounting holes for installing industrial-grade standard connectors, such as USB Type-C interfaces or dedicated power interfaces. It is fixed by screws or clips and has strong mechanical strength against plugging and unplugging, adapting to the frequent use needs of clinical environments.

[0052] Furthermore, crystal oscillator Y1 provides a precise and stable master clock signal for the entire system. The load capacitors C27 and C28 at both ends of the crystal oscillator are precisely matched to ensure the accuracy and stability of the oscillation frequency, with a frequency deviation better than ±20ppm, providing a reliable time base reference for ADC synchronous sampling and digital communication.

[0053] LED1, a status indicator light, is mounted on the edge of the PCB board and displays the system's operating status through corresponding light guide holes on the casing. Different flashing modes indicate states such as normal power supply, data acquisition in progress, and abnormal alarm, allowing operators to quickly determine the equipment's working condition. A solid light indicates normal power supply, slow flashing indicates data acquisition in progress, and fast flashing indicates an abnormal alarm.

[0054] SW1 is a push-button switch used for system power on / off or function reset, supporting short press to turn the device on / off and long press to restore factory settings.

[0055] D1 is a diode placed in the power input circuit to provide reverse polarity protection and prevent circuit damage caused by operational errors.

[0056] The board is covered with a large number of 0402 and 0603 packaged surface mount resistors (Rxx series) and surface mount capacitors (Cxx series), distributed around each IC and on the signal path, serving the following functions: decoupling capacitors next to the power supply pins are used to filter out high-frequency power supply ripple noise; RC filter networks on the signal path are used to suppress out-of-band noise and high-frequency interference; voltage divider and bias resistor networks are used to establish an accurate DC operating point for the analog front end; pull-up and pull-down resistors are used to ensure deterministic logic levels in the idle state of the digital bus.

[0057] Furthermore, the 32-channel EEG signal amplifier module of this invention integrates a wireless communication module, supporting real-time data transmission with a host computer via Bluetooth or Wi-Fi. The wireless communication transmission has the following characteristics: it employs a reliable wireless transmission protocol, ensuring the real-time transmission bandwidth requirement of 32-channel EEG data at a sampling rate of 250 SPS, approximately 24 kilobytes per second; the wireless transmission distance is no less than 10 meters (in an open indoor space), meeting the spatial layout requirements of a clinical EEG acquisition room; the wireless method eliminates the physical constraints of data cables on the subject, making it particularly suitable for the comfort and safety of stroke patients when completing motor-related experimental tasks.

[0058] Furthermore, such as Figure 5 As shown, the PCB circuit board of the EEG amplifier is installed inside a custom-designed housing, which takes into account electromagnetic shielding, heat dissipation, portability, and clinical safety requirements.

[0059] The outer casing has a flat, box-like structure, with the top and bottom covers fastened together by screws. The sides of the casing have pre-drilled openings for the EEG cap signal input connector, a USB or power interface, LED indicator light guide holes, and button operation holes.

[0060] The PCB is secured inside the housing via mounting holes and isolation studs, ensuring board stability while providing electrical isolation from the metal casing. The entire unit is compact and lightweight, facilitating transport and deployment in clinical environments.

[0061] As one embodiment of the present invention, such as Figures 8 to 10 As shown, the host computer data acquisition and visualization software module is independently developed based on the National Instruments LabVIEW graphical programming platform and runs on a PC with a Windows operating system.

[0062] Furthermore, the software module establishes a connection with the EEG amplifier via wireless communication, including: The first processing unit is used for real-time visualization of 32-channel EEG signals. For example... Figure 9 As shown, the software interface can simultaneously display real-time rolling waveforms of 32 channels of EEG signals. Each channel is labeled according to standard electrode names, such as Fp1, Fp2, F3, F4 up to O1, O2, arranged vertically in the display area. The operator can intuitively observe the EEG activity characteristics of each brain region, including stroke-related electrophysiological features such as alpha rhythms, abnormal slow waves, and epileptiform discharges. The display interface supports adjustment of the time window width and amplitude sensitivity scaling, allowing the operator to flexibly configure settings according to different observation needs.

[0063] The second processing unit is used for starting and ending signal acquisition. The interface has clearly defined start and stop recording control buttons. The operator presses the start recording button to initiate data writing according to the experimental procedure, and presses the stop recording button to end data writing after the experiment, ensuring that only valid experimental data segments are recorded.

[0064] The third processing unit is used for autonomous selection of data storage path and TXT format storage. For example... Figure 10 As shown, the software provides a data save path selection function, allowing operators to specify the storage directory and filename of the data files via the file dialog box. The acquired 32-channel EEG data is saved in TXT text format, with the voltage values ​​of each channel recorded in columns at each sampling time using fixed delimiters, facilitating subsequent reading and analysis using tools such as MATLAB and Python. The TXT format has the advantages of strong cross-platform compatibility and no need for dedicated parsing tools, lowering the threshold for subsequent data processing.

[0065] The software module adopts LabVIEW's loop event-driven architecture, with the data receiving thread and the interface display thread running separately. Asynchronous data transmission is achieved through a queue mechanism, ensuring the real-time nature of data acquisition and the smoothness of interface operation.

[0066] This invention has the following characteristics: 1. Clinical diagnostic level: It fills the gap in the study of early EEG characteristics within 96 hours of ischemic stroke onset. The established standardized inclusion and exclusion criteria and collection paradigm can achieve accurate identification of early EEG characteristics of stroke, providing a new electrophysiological means for the early diagnosis of stroke and making up for the limitations of imaging examinations such as CT in the early stage of stroke.

[0067] 2. Equipment R&D: The self-developed 32-lead portable EEG device solves the contradiction between portability and acquisition accuracy of existing devices, is suitable for primary medical institutions and bedside acquisition scenarios, greatly improves the accessibility of EEG screening for stroke at the grassroots level, and promotes the popularization and application of EEG technology in stroke prevention and treatment in counties.

[0068] 3. Basic Research Level: By using G-power scientific calculations to determine the research sample size, a matched study was established between the stroke group and the healthy control group in the high-risk population of stroke aged 40-70. This provided a large-sample, high-quality basic data for stroke EEG risk grading and risk profiling, and promoted the development of stroke EEG research from qualitative analysis to quantitative and individualized analysis.

[0069] 4. Clinical application: The characteristics and patterns of ICA topography and PSD analysis in stroke patients were clarified, and the correlation between EEG characteristics and stroke cortical functional impairment was established. This provides targeted electrophysiological basis for the formulation of subsequent stroke rehabilitation intervention programs and the application of brain-computer interfaces in stroke rehabilitation, and promotes the translation of EEG technology from basic research to clinical practice.

[0070] 5. At the primary healthcare level: The established operational guidelines and environmental control standards for primary healthcare-adaptive EEG acquisition have lowered the technical threshold for EEG acquisition, enabling standardized acquisition of stroke EEG signals by primary healthcare institutions. This promotes the downward extension of my country's stroke prevention and control system to the grassroots level and enhances the county-level capacity for early stroke screening and intervention.

[0071] 6. Technical System Level: The established standardized system for the entire process of early EEG acquisition, preprocessing, and analysis in ischemic stroke provides a replicable and scalable technical paradigm for stroke EEG research, filling the standardization gap in this field and promoting the standardized development of stroke EEG characteristic research.

[0072] Example 2 This invention also provides a method for assessing stroke risk, comprising: Step 1: Acquiring EEG signals from stroke patients like Figure 11 As shown, based on the above hardware system and software platform, this invention provides a complete method for acquiring electroencephalogram (EEG) signals in stroke subjects, with the specific steps as follows.

[0073] Step S11, Equipment Connection and Preparation. Connect the EEG amplifier module 200 and the EEG cap module 100 via lead cables. Correctly place the EEG cap on the head of the stroke subject, aligning each electrode according to the international 10-20 system standard. Inject conductive gel between each electrode and the scalp, ensuring that the impedance of all 32 electrode channels is less than or equal to 5 kiloohms.

[0074] Step S12: Software Startup and Communication Establishment. Start the LabVIEW EEG data acquisition software on the PC. The amplifier establishes a real-time data transmission connection with the host computer software via wireless communication. The operator confirms that the 32-channel signals are displayed normally on the software interface.

[0075] Step S13: Execute the stroke experimental paradigm. This invention designs a specific experimental paradigm for EEG signal acquisition in stroke subjects. This paradigm includes multiple functional assessment task stages, where subjects complete corresponding cognitive or motor tasks based on on-screen instructions or voice prompts. The operator clicks the start recording button at the beginning of the paradigm, and the software simultaneously records 32 channels of EEG signals.

[0076] The experimental paradigm comprehensively encompasses the following assessment dimensions: The resting-state acquisition phase involves baseline EEG recordings of subjects in a quiet, awake state with their eyes closed or open, used to extract resting-state EEG rhythm characteristics. The cognitive task acquisition phase involves subjects performing attention, memory, or language-related cognitive tasks, used to assess the impact of stroke on cognitive functional networks. The motor imagery or motor attempt acquisition phase involves subjects imagining or actually attempting movement of the affected or unaffected limbs based on prompts, used to assess the functional status and lateralization degree of the motor cortex.

[0077] Appropriate rest intervals were set between each task phase, and the overall collection time was controlled within a reasonable range based on the subject's tolerance.

[0078] Step S14: Data Saving. After the experimental paradigm is completed, the operator clicks the Stop Recording button, selects the data saving path in the software, and saves the 32-channel EEG data in TXT format to the specified location for subsequent offline analysis.

[0079] Step 2: Stroke EEG Risk Assessment like Figures 12 to 14 As shown, after completing the EEG data acquisition, this invention further provides a stroke risk assessment method based on EEG features. This method comprises two core stages: the stroke EEG feature extraction stage and the stroke risk analysis model establishment stage.

[0080] Step S15: Extract EEG features of stroke patients like Figure 13 As shown, this invention constructs three types of EEG features for stroke from the perspectives of clinical significance and signal analysis, comprehensively characterizing the multi-dimensional pathological features of EEG signals in stroke patients.

[0081] The first type of characteristic is frequency band and energy characteristics. The frequency band and energy characteristics mentioned correspond to clinical indicators for assessing the severity of stroke.

[0082] Frequency domain analysis was performed on the acquired 32-channel EEG signals to extract features such as power spectral density, absolute and relative power, and power ratios of each standard frequency band. The standard frequency bands were defined as the delta band (0.5–4 Hz), theta band (4–8 Hz), the alpha band (8–13 Hz), the beta band (13–30 Hz), and the gamma band (30–45 Hz). Among the power ratios, the delta to alpha power ratio (DAR) was particularly important, as it has been extensively demonstrated in clinical studies to be significantly correlated with stroke severity and prognosis.

[0083] These characteristics can quantify the slowing of brain activity caused by stroke, namely the characteristic change of increased delta and theta activity and decreased alpha activity in the lesion area. They directly reflect the extent and severity of cerebral ischemia-reperfusion injury and are important electrophysiological indicators for clinically measuring the degree of neurological deficits in stroke.

[0084] Power spectral density (PSD) is an important property used to describe the distribution of a signal in the frequency domain, reflecting the distribution of signal power at different frequencies.

[0085] Power spectra for all EEG bands were calculated using Fast Fourier Transform (FFT) at a sampling frequency of 256 Hz and a default window length twice the lowest frequency of the band. First, the Welch method was used to calculate the power spectral density (PSD) of each electrode in the six frequency bands: δ, θ, α, β1, β2, and γ. To calculate the total power in a specific frequency band, the Simpson rule was used to integrate the PSD, and the option to calculate relative power (dividing the power of a specific frequency band by the total power of the entire spectrum) was optional to eliminate the influence of bandwidth. The signal was divided into multiple segments, each processed using a window function and then subjected to FFT. The PSD estimates of these segments were averaged to obtain the final PSD estimate. This method not only improves the accuracy of PSD estimation but also helps us to more accurately analyze and interpret the frequency domain characteristics of EEG activity. The specific calculation method is as follows: If the length is discrete signals Divided into There are sub-segments, and the number of points in each sub-segment is . The distance between the starting points of two consecutive sub-segments is 1. That is, the number of overlapping points of consecutive sub-segments is Then the first window after adding a window Each sub-segment can be represented as: in Represents a window function, and . No. The PSD estimate of the segment is obtained by using the periodogram method. for: in, This represents the Fast Fourier Transform. Finally, the PSD of each segment is estimated. Average signal PSD estimation for: Absolute Band Power (ABP) is defined as the power spectral density in a specific frequency band. Points on: Relative Band Power (RBP) is defined as the proportion of power in a specific frequency band to the total power across the entire frequency band, eliminating the influence of baseline power differences between individuals. The frequency band power ratio, calculated by the ratio of the absolute power of two frequency bands, characterizes the dynamic balance between brainwave rhythms. Its general form is: The second category of features is entropy and complexity features. The entropy and complexity features correspond to empirical quantitative indicators of clinical symptoms.

[0086] Nonlinear dynamic analysis of EEG signals was performed to extract features such as sample entropy, fuzzy entropy, permutation entropy, multi-scale entropy, and Lempel-Ziv complexity.

[0087] Stroke patients experience significant alterations in the complexity and irregularity of their electroencephalogram (EEG) signals due to ischemic and hypoxic brain damage. The entropy value of the brain region on the side of the lesion is typically reduced, reflecting either an increase in the ordered nature of neuronal activity in that area or a loss of normal, complex dynamics. These characteristics can capture subtle pathological changes in EEG signals that are difficult to detect through traditional linear frequency domain analysis, providing quantitative and objective evidence for clinical empirical diagnosis.

[0088] Sample entropy (SE) is a method for measuring the complexity and uncertainty of a time series signal. It assesses the complexity of a signal by calculating the probability of patterns appearing in the signal. The higher the sample entropy, the more complex the signal, and vice versa. The specific calculation method is as follows: For a given time series The construction length is pattern vector For each mode vector Calculate the distance (usually the maximum distance) to other pattern vectors, and statistically determine the distance to be less than 1. The number of pattern vectors. Sample entropy is defined as: in, and These represent the number of matching pattern vectors.

[0089] Fuzzy entropy: Replaces the hard threshold decision in the sample entropy with a continuous fuzzy membership function, thereby eliminating boundary discontinuities and improving estimation stability.

[0090] For length of Construct a mean-removed sequence. 3D template vector: The similarity between any two templates is measured using an exponential fuzzy membership function: in For Chebyshev distance, For gradient parameters, This is the width parameter. Defines the scale. Average similarity: Fuzzy entropy is defined as the logarithm of the ratio of average similarity between adjacent scales: Permutation entropy: It quantifies the complexity of a time series by analyzing its local ordinal pattern, and has the advantages of being computationally simple and robust to noise.

[0091] For sequence Construct the embedding dimension as The time delay is Phase space vectors: right The components are arranged in ascending order to obtain a unique permutation pattern. There are a total of Count the possible permutations. Count the permutations. relative frequency of occurrence The permutation entropy is defined as: Multiscale entropy: In view of the complexity of physiological signals at multiple time scales, sample entropy is calculated at different scales through a coarse-graining process to obtain the curve of complexity changing with scale.

[0092] Coarsening process: For scale factor The original sequence is divided into segments of length . Construct the first non-overlapping window, taking the average value within each window. Coarse-grained sequences at scale: when hour, This is the original sequence.

[0093] Multi-scale entropy: The sample entropy is calculated separately for each scale of the coarse-grained sequence to obtain the multi-scale entropy curve. Lempel-Ziv complexity algorithm flow: 1. Binarization: Converting time series data into binary data. Binarization is performed using its mean (or median) as a threshold to obtain the symbol sequence. ; 2. Pattern matching: A greedy parsing algorithm is used to match the sequence... Decompose from left to right into as few substrings (dictionary) as possible, where each new substring has not appeared in the parsed prefix; 3. Complexity Counting: Let the number of substrings obtained from parsing be... ; 4. Normalization: For Normalization is performed to eliminate the influence of sequence length: in LZC represents the theoretical maximum dictionary number for a random binary sequence. A LZC value close to 1 indicates that the sequence is nearly random, while a LZC value close to 0 indicates that the sequence is highly regular.

[0094] The third category of features is statistical and time-frequency features. The statistics and time-frequency characteristics correspond to dynamic monitoring indicators for stroke patients.

[0095] This type of feature comprises two parts: time-domain statistics of the EEG signal and time-frequency joint analysis features. Time-domain statistics include mean, variance, skewness, kurtosis, and zero-crossing rate. Time-frequency joint analysis features are extracted based on short-time Fourier transform, wavelet transform, or Hilbert-Huang transform, and include time-frequency energy distribution, instantaneous frequency, and instantaneous amplitude.

[0096] These features can characterize the dynamic changes of brain signals in stroke patients over time, reflect the real-time fluctuations in brain function after a stroke, and provide dynamic tracking indicators for continuous monitoring of stroke progression or recovery.

[0097] The Hurst exponent (H) is a metric for measuring the long-range dependence and persistence of a time series. It assesses the persistence or anti-persistence of a signal by analyzing its fluctuation behavior.

[0098] For a given time series Construct cumulative deviation sequence in, This represents the mean of the time series. For cumulative deviation series... Calculate its range and standard deviation Normalized range By fitting the relationship between the normalized range and the time length in logarithmic coordinates, the Hearst exponent is finally calculated: Higuchi Fractal Dimension (HFD) is a method used to describe the complexity and self-similarity of time series data. It evaluates the complexity of a signal by calculating its dimension at different scales.

[0099] For a given time series According to different scale factors Constructing new time series ,in As the initial point, The scaling factor. For each new time series Calculate its length For a given scale factor Calculate the average length of all new sequences. The fractal dimension is estimated by fitting the relationship between the mean length and the scale factor in logarithmic coordinates. Mean The mean reflects the DC component or overall potential shift of the signal: After processing with a good reference electrode, the mean value of the EEG signal is usually close to zero.

[0100] Variance Variance measures the dispersion of signal amplitude and is directly related to the energy of EEG oscillations. Skewness Skewness describes the asymmetry of the signal amplitude distribution relative to the mean: This indicates a right-skewed (positively skewed) distribution. Indicates left deviation (negative deviation).

[0101] Kurtosis Kurtosis describes the thickness of the tails in a signal's amplitude distribution, reflecting the frequency of extreme amplitude events (such as spikes and artifacts). The theoretical kurtosis value for a normal distribution is 3. Excess kurtosis is typically used. A value of 0 corresponds to a normal distribution. This indicates a thick-tailed (spike) distribution.

[0102] Zero-Crossing Rate (ZCR) The zero-crossing rate is the statistical measure of the number of times a signal crosses the zero baseline per unit time. It approximately reflects the dominant frequency component of the signal and is positively correlated with the oscillation frequency. in This is an indicative function. The zero-crossing rate can also be calculated based on the mean-removed sequence to reduce interference from DC offset.

[0103] Based on Short-Time Fourier Transform (STFT) The short-time Fourier transform performs discrete Fourier analysis on locally stationary signal segments using a sliding window, and there is a trade-off between time and frequency resolution determined by the uncertainty principle. ).

[0104] STFT definition: in For The analysis window function centered on the center (commonly the Hann window or Gaussian window).

[0105] Time-frequency energy distribution (spectrogram): Spectrograms visually reflect the two-dimensional distribution of brain electrical energy in the time-frequency plane and can be used to track dynamic changes in frequency band power related to events (such as the ERD / ERS phenomenon).

[0106] Based on wavelet transform (WT) Continuous wavelet transform (CWT) uses a mother wavelet that satisfies the admissibility condition. Using the basis functions, the analysis basis is constructed through scale scaling and time shifting, which overcomes the limitation of fixed time and frequency resolution of STFT, and has high frequency resolution in the low frequency band and high time resolution in the high frequency band.

[0107] Definition of continuous wavelet transform: in The scale parameter (inversely proportional to frequency) , (the center frequency of the mother wavelet) For translation parameters, Indicates conjugate.

[0108] Wavelet energy spectrum (Scalogram): The energy at each scale is: Based on Hilbert-Huang Transform (HHT) The Hilbert-Huang transform, proposed by Huang et al. (1998), is specifically designed for analyzing nonlinear and non-stationary signals. It does not require prior assumptions about the basis function forms and is fully adaptive. Its core consists of two steps.

[0109] Empirical Mode Decomposition (EMD) EMD will convert the original signal Adaptively decomposed into the sum of a finite number of intrinsic mode functions (IMFs): in For the first One IMF, This represents the residual trend term. Each IMF must satisfy two conditions: ① the difference between the number of extreme points and the number of zero-crossing points does not exceed 1; ② the mean envelope defined by the envelopes of local maxima and minima is zero. IMFs are extracted iteratively using a sifting algorithm.

[0110] Hilbert spectral analysis For each IMF Applying the Hilbert transform, we obtain its analytic signal: Where PV represents the Cauchy principal value integral. From this, the instantaneous amplitude and instantaneous frequency are extracted: Step S22, Stroke Risk Analysis Model like Figure 14 As shown, this invention, based on the acquisition, preprocessing, and extraction of three types of EEG features, fuses these three types of features to construct a multidimensional stroke EEG feature vector, which is then used as input to establish a stroke risk analysis model. The construction and training process of the stroke risk analysis model includes the following steps: Step S21: Construction of training dataset and label set Electroencephalogram (EEG) data were collected from diagnosed stroke patients and healthy controls to construct a model training dataset. The EEG data were preferably acquired using standardized acquisition equipment, electrode placement, and sampling parameters to ensure consistency and comparability of data among different subjects.

[0111] After obtaining the raw EEG data, the same preprocessing operations as those described above are performed on it, including but not limited to artifact removal, filtering, segmentation, baseline correction and standardization, in order to reduce noise interference and improve the stability of subsequent feature extraction.

[0112] Subsequently, the aforementioned three types of EEG features are extracted from the preprocessed EEG data, and the three types of features are spliced, weighted, or dimensionality-reduced to construct a multidimensional stroke EEG feature vector for each subject, which serves as input data for the stroke risk analysis model.

[0113] At the same time, each training sample is assigned a corresponding clinical label, which is derived from the clinical gold standard assessment results and preferably includes at least one of the following: whether a stroke has occurred; stroke risk level; National Institutes of Health Stroke Scale score, i.e., NIHSS score; modified Rankin Scale score, i.e., mRS score; lesion lateralization information; and severity grading.

[0114] The clinical labels are used to constrain model parameter updates during supervised learning, enabling the model to learn the mapping relationship between EEG features and stroke risk, lesion distribution, and disease severity.

[0115] Step S222: Divide the training set, validation set, and test set. The constructed training dataset and label set are divided into training, validation, and test sets according to a preset ratio. Preferably, the training set is used for model parameter learning, the validation set is used for hyperparameter tuning and model selection during training, and the test set is used to independently evaluate the model performance after training is completed.

[0116] To prevent sample distribution shift, stratified sampling is preferred for data partitioning, ensuring a relatively consistent class distribution between the stroke patient group and the healthy control group within each subset. For scenarios with a small sample size, K-fold cross-validation can also be used to evaluate the model's generalization performance.

[0117] Step S223: Construction of Stroke Risk Analysis Model Based on the multidimensional EEG feature vectors of stroke and their corresponding clinical labels, a stroke risk analysis model is constructed. This risk analysis model can be implemented using a machine learning model or a deep learning model.

[0118] Optionally, the machine learning model includes, but is not limited to: support vector machine; random forest; gradient boosting decision tree.

[0119] Among them, support vector machines are suitable for small-sample classification scenarios in high-dimensional feature spaces; random forests achieve risk classification by integrating multiple decision trees through voting, and have strong noise resistance and robustness; gradient boosting decision trees gradually improve prediction accuracy by iteratively optimizing the loss function, and are suitable for modeling structured EEG feature vectors.

[0120] Furthermore, the deep learning model includes, but is not limited to: deep neural networks; convolutional neural networks; long short-term memory networks; and hybrid architectures of convolutional neural networks and long short-term memory networks.

[0121] Among them, deep neural networks can perform multi-layer nonlinear mapping on the fused high-dimensional EEG features to mine deep discriminative features; convolutional neural networks are suitable for extracting local pattern features in EEG signals; long short-term memory networks are suitable for modeling the dynamic dependence of EEG signals over time; and the hybrid architecture of convolutional neural networks and long short-term memory networks can simultaneously characterize the spatial local features and temporal dynamic features of EEG signals, thereby improving the accuracy of stroke risk analysis.

[0122] Step S224: Model Training The multidimensional stroke EEG feature vectors and their corresponding clinical labels from the training set are input into the stroke risk analysis model for supervised training.

[0123] For classification tasks, the model outputs classification results such as whether a stroke has occurred, risk level, or lesion laterality, and the model can be optimized using the cross-entropy loss function. For regression tasks, the model outputs stroke risk probability values, NIHSS score prediction values, or mRS score prediction values, and the model can be optimized using the mean squared error loss function or the mean absolute error loss function.

[0124] When the model outputs multiple prediction targets simultaneously, a multi-task joint loss function can be constructed. By weighted summation of the losses from different tasks, the synergistic prediction of stroke risk level, lesion laterality, and severity can be achieved.

[0125] During training, gradient descent or its improved algorithms are preferably used to iteratively update the model parameters. These improved algorithms include, but are not limited to, stochastic gradient descent, Adam optimization, or RMSProp optimization. To improve model stability and prevent overfitting, at least one of the following strategies can be introduced during training: Dropout random deactivation; L1 or L2 regularization; early stopping strategy; learning rate decay strategy; and data augmentation strategy.

[0126] The data augmentation strategies may include resampling of EEG segments, sliding window segmentation, slight noise perturbation, or channel combination expansion to improve the model's adaptability to individual differences and changes in the data collection environment.

[0127] Step S225: Model Validation and Parameter Optimization During model training, a validation set is input into the current model to evaluate its predictive performance on unseen data. Based on the validation results, the model structure and training parameters are optimized.

[0128] The parameters to be optimized include, but are not limited to: feature dimension or feature fusion method; kernel function type and penalty parameter of support vector machine; number of trees and tree depth of random forest; learning rate, number of trees and leaf node parameters of gradient boosting decision tree; number of network layers, convolution kernel size, number of hidden layer units and batch size of deep learning model; learning rate, momentum parameter and regularization coefficient of optimization algorithm.

[0129] When the loss function value on the validation set tends to stabilize, or when the model performance index no longer improves after several consecutive iterations, the model training can be terminated, and the model parameters with the best validation performance can be retained to obtain the trained stroke risk analysis model.

[0130] Step S226: Model Testing and Performance Evaluation Input the test set that was not involved in the training and validation process into the trained stroke risk analysis model to independently evaluate the model's actual generalization ability.

[0131] The performance evaluation metrics include, but are not limited to: accuracy, sensitivity, specificity, precision, F1 score, and area under the receiver operating characteristic curve (AUC).

[0132] The above indicators can be used to comprehensively evaluate the model's ability to distinguish between stroke patients and healthy controls, its ability to identify risk levels, and its ability to predict the severity of the disease.

[0133] Step S227: Model Deployment and Risk Output After the model training is completed and passes testing and validation, the stroke risk analysis model is deployed in the rapid stroke assessment system. For new subjects, their EEG data is first collected, and preprocessing, three types of EEG feature extraction, and feature fusion are completed according to the same data processing workflow as the training phase to obtain the corresponding multidimensional stroke EEG feature vector. Subsequently, the feature vector is input into the trained stroke risk analysis model to obtain the stroke risk assessment result.

[0134] The risk assessment results include, but are not limited to: A probability value for the risk of stroke, wherein the probability value is a continuous output between 0 and 1; Risk levels are categorized, with a preferred classification including low risk, medium risk, and high risk. Additional information for diagnosis of the lesion on the side of the lesion; Auxiliary assessment results of the severity of the illness.

[0135] Specifically, a preset threshold can be set based on the risk probability values ​​output by the model, further mapping continuous outputs to graded risk levels for quick interpretation by clinicians. The results of lesion lateralization and severity assessment can be output simultaneously with the risk level, providing supplementary information for subsequent diagnostic and treatment decisions.

[0136] The workflow of the method of the present invention is as follows: Equipment deployment phase: The EEG amplifier is connected to the EEG cap, the EEG cap is worn on the subject's head, and conductive gel is injected.

[0137] Software preparation phase: Start the LabVIEW software and establish a wireless communication connection with the amplifier.

[0138] Signal confirmation phase: Visual confirmation of 32-channel signals, checking the signal quality of each channel.

[0139] Data acquisition phase: The stroke experimental paradigm is implemented, and subjects complete designated tasks and their electroencephalogram (EEG) signals are recorded.

[0140] Data saving stage: Select the save path and store the 32-channel EEG data in TXT format.

[0141] Risk assessment phase: Extract EEG features of stroke (three types of features), input them into the risk analysis model for reasoning, and output the risk level.

[0142] This invention has the following characteristics: 1. Enhancing the Accuracy of EEG Signal Analysis: This invention significantly improves the accuracy of stroke risk assessment by combining multi-dimensional EEG characteristics (such as band power, entropy and complexity features, and time-frequency characteristics). In particular, changes such as increased power in delta and theta waves and decreased power in alpha waves can accurately identify abnormal patterns in the EEG of stroke patients. Through these characteristics, this invention significantly improves the reliability of stroke risk assessment, providing a more accurate basis for early clinical diagnosis.

[0143] 2. Optimizing the Standardization and Reproducibility of EEG Signal Acquisition: Existing technologies exhibit low standardization in EEG acquisition, particularly in motor-stimulated EEG acquisition, which is susceptible to interference and suffers from poor data stability. To address this issue, this invention introduces strict time control and action specifications into the acquisition paradigm, ensuring data stability and reproducibility. By rigorously controlling the acquisition process, this invention guarantees the consistency and reliability of experimental data.

[0144] 3. Improved Portability and Field Adaptability: Compared to existing large desktop EEG devices, the 32-channel portable EEG device of this invention is more suitable for primary healthcare institutions and on-site data acquisition during the acute phase. This device is compact, lightweight, and possesses high-precision multi-channel signal acquisition capabilities. Through optimized design, this device meets the portability requirements of primary healthcare institutions while ensuring high sampling rate and accuracy.

[0145] Example 3 The present invention also provides a storage medium storing a computer program that executes a stroke risk assessment method when running.

[0146] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A stroke risk assessment system, characterized in that, include: The system comprises an EEG cap module, a 32-channel EEG signal amplifier module, and a host computer data acquisition, visualization, and risk assessment software module. The scalp EEG signals are picked up by the EEG cap module and transmitted via leads to the 32-channel EEG signal amplifier module. The 32-channel EEG signal amplifier module, acting as a front-end, amplifies, filters, and performs ADC conversion. After digital processing and encapsulation, the signals are wirelessly transmitted to the host computer data acquisition, visualization, and risk assessment software module. This software module receives the data, displays it in real-time, extracts stroke EEG features, inputs them into a stroke risk analysis model, and finally outputs a risk level assessment result.

2. The stroke risk assessment system as described in claim 1, characterized in that, The 32-channel EEG signal amplifier adopts a multi-layer high-density printed circuit board design, with digital circuits as the main architecture, integrating a high-performance core processing unit, multiple signal conditioning channels, wireless communication interface and power management system.

3. The stroke risk assessment system as described in claim 2, characterized in that, The EEG cap module is made of elastic fabric. The cap surface has 32 pre-set electrode mounting positions according to the international 10-20 system standard, covering the frontal, central, parietal, temporal and occipital regions, with additional reference and ground electrode positions.

4. A method for assessing stroke risk, characterized in that, include: Collecting electroencephalogram (EEG) signals from stroke patients; Extracting EEG features from stroke EEG signals; Based on the extracted EEG features of stroke, a stroke risk analysis model was constructed. The 32-channel EEG signals collected in real time are input into the stroke risk analysis model, and the final risk level assessment result is output.

5. The stroke risk assessment method as described in claim 4, characterized in that, The extracted EEG features of stroke include: frequency band and energy features, entropy and complexity features, and statistical and time-frequency features.

6. The stroke risk assessment method as described in claim 5, characterized in that, Frequency band and energy characteristics include: power spectral density of each standard frequency band, absolute power and relative power of each frequency band, and frequency band power ratio.

7. The stroke risk assessment method as described in claim 5, characterized in that, Entropy and complexity features include: sample entropy, fuzzy entropy, permutation entropy, multi-scale entropy, and Lempel-Ziv complexity features.

8. The stroke risk assessment method as described in claim 5, characterized in that, Statistical and time-frequency characteristics include: time-domain statistics and time-frequency joint analysis characteristics of EEG signals; among which, time-domain statistics include mean, variance, skewness, kurtosis and zero-crossing rate; time-frequency joint analysis characteristics are based on short-time Fourier transform, wavelet transform or Hilbert-Huang transform to extract time-frequency energy distribution, instantaneous frequency and instantaneous amplitude.

9. A storage medium, characterized in that, The storage medium stores a computer program that, when executed, performs the stroke risk assessment method as described in any one of claims 4-8.

Images