A field brain injury rapid diagnosis system based on electroencephalogram data

Abstract

This invention discloses a rapid field brain injury diagnostic system based on electroencephalogram (EEG) data, belonging to the field of EEG data analysis and diagnostic technology. It includes an external stimulation module that utilizes sound waves and electrical impulses to stimulate limb muscles to induce specific event-related potentials and motor cortical activity rhythms for detection; an EEG data analysis module that performs multi-scale time window preprocessing and deep learning algorithm analysis; and an injury assessment module that determines the type and quantifies the degree of injury. This system can be implemented using a portable EEG acquisition helmet, a portable pulse stimulation device, and a chip-level handheld terminal to achieve a method from stimulation to detection to diagnosis. The overall device is lightweight, suitable for deployment in field scenarios, and can operate stably in complex environments.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) data analysis and diagnostic technology, specifically a rapid diagnostic system for brain injury in the field based on EEG data. Background Technology

[0002] In field environments, there are often risks of sudden impacts such as explosions, collisions, and falls. These impact risks may cause closed head injuries to combat personnel, meaning that they lack obvious external trauma symptoms and are therefore easily overlooked and difficult to implement accurate early emergency interventions.

[0003] These types of brain injuries often manifest as confusion, slowed reaction, or other cognitive impairments. However, in the absence of significant trauma, the diagnosis of mild brain injuries currently relies primarily on imaging examinations such as CT, MRI, and neurological functional assessments. While CT can detect hemorrhage and fractures, it lacks sensitivity to minute structural changes such as diffuse axonal damage. Other high-precision imaging devices are bulky, time-consuming, and not suitable for mobile deployment in the field. Neurological functional assessments, such as the Glasgow Coma Scale, can be performed on-site, but they are highly subjective, susceptible to environmental interference, and struggle to detect the extent and location of functional connectivity impairment within the brain. Summary of the Invention

[0004] 1) Technical problems to be solved

[0005] This invention provides a rapid diagnostic system for brain injury in the field based on electroencephalogram (EEG) data, which can quickly assess brain injury in target casualties in harsh field environments.

[0006] (ii) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution: a rapid field brain injury diagnosis system based on electroencephalogram (EEG) data, comprising:

[0008] The external stimulation module is used to apply sound wave stimulation and muscle electrical pulse stimulation to the target injured person in stages. Each stimulation has multiple intensity levels and is applied independently level by level. The sound wave stimulation is used to induce event-related EEG signal responses, and the muscle electrical pulse stimulation is used to induce EEG signal responses related to motor cortex activity.

[0009] The data acquisition module is used to acquire real-time data on the target wounded person under the stimulation of the sound waves. , and Electroencephalogram (EEG) signals in the band, and under the stimulation of the muscle electrical impulses and Electroencephalogram (EEG) signals in the specified frequency band;

[0010] The data preprocessing module removes artifacts from the EEG signals of each band and uses a multi-time-scale time window to extract the time-domain and frequency-domain features of the EEG signals of each band. Within the same time window, it calculates the frequency-domain coherence and phase lag index between different EEG signals and constructs a brain region connectivity matrix that reflects the functional coupling state.

[0011] The EEG data analysis module uses a deep learning model to jointly analyze the extracted time-domain features, frequency-domain features, and brain region connectivity matrix to identify abnormal EEG signals induced by external stimuli. Specifically, it captures the dynamic changes of EEG signals within different time windows through parallel convolution kernels in multi-scale convolutional layers, and extracts abnormal coupling patterns from the brain region connectivity matrix through spatial convolutional layers.

[0012] The injury assessment module determines the type of brain injury based on the abnormal EEG signals and their coupling patterns, locates the brain injury area based on the acquisition location of the abnormal EEG signals, compares them with healthy standard signals, quantifies the frequency band power deviation and the degree of functional connectivity loss, grades and assesses the degree of injury, and generates a diagnostic report including the injury type, injury area, and injury degree.

[0013] Furthermore, the external stimulation module is used to induce specific brain region activity responses in the target injured person by progressively increasing intensity of the sound wave stimulation and limb muscle electrical pulse stimulation, in order to assess perceptual function, motor function, and brain region activity; wherein, the sound wave stimulation is used to induce event-related potentials to assess perceptual and cognitive functions, and the limb muscle electrical pulse stimulation is used to stimulate motor cortex-related potentials to reflect the activity of the motor cortex.

[0014] Furthermore, the data acquisition module is used to acquire relevant EEG signals of the injured person in real time under the sound wave stimulation and limb muscle electrical pulse stimulation; specifically, the data acquisition module is connected to a multi-channel EEG electrode cap, which covers the frontal lobe, parietal lobe, temporal lobe, occipital lobe and central cortex of the target injured person.

[0015] Furthermore, the data acquisition module acquires data under acoustic stimulation. , and EEG signals in the band, among which, The EEG signals in the band reflect the inhibition of brain region activity. The EEG signals in this band are related to attention. Electroencephalogram (EEG) signals in certain bands reflect the coordination of brain regions;

[0016] Obtaining electrical impulse stimulation of limb muscles and EEG signals in the band, among which, The EEG signals in the band reflect the inhibitory state of the sensory cortex. Electroencephalogram (EEG) signals in certain bands reflect the readiness of motor function.

[0017] Furthermore, the data preprocessing module first performs artifact removal on all EEG signal data acquired by the data acquisition module; specifically, the data preprocessing module uses bandpass filtering to remove DC drift and high-frequency electromyographic noise, uses notch filtering to remove power frequency interference, and uses independent component analysis to identify and remove non-EEG related signals, including electrooculography and electromyography.

[0018] Furthermore, after removing artifacts from the acquired EEG data of each frequency band, the data preprocessing module pre-divides the data into multiple time windows of different and increasing lengths.

[0019] For each preprocessed EEG signal acquired after each external stimulus, the EEG signal is subjected to parallel sliding processing using all preset time windows of different lengths. Under the sliding processing of each time window, its corresponding time domain features and frequency domain features are extracted. The time domain features include amplitude peak value, mean and variance, and the frequency domain features include power spectral density.

[0020] Furthermore, within the same time window, frequency domain analysis is performed on any two EEG signals of the same band acquired under external stimuli of different intensities. The coherence coefficients of the two signals in the corresponding bands are calculated to reflect the coherence of the activities of the two corresponding brain regions. The phase lag index is calculated by extracting the sequential phase of the two signals. Based on the calculated frequency domain coherence and phase lag index, a corresponding symmetric matrix is ​​generated for each external stimulus, which is used as the brain region connectivity matrix.

[0021] Furthermore, regarding the deep learning model used in the EEG data analysis module, different sizes of convolutional kernels are set to extract EEG signal features at different time scales; wherein, small-sized convolutional kernels are used to extract short-term fluctuation features and capture high-frequency components; large-sized convolutional kernels are used to extract long-term trend features and capture low-frequency components.

[0022] Deep convolutional layers in deep learning algorithms use two-dimensional convolution to focus on the interaction features between different brain regions, and aggregate the connection strength between a group of brain regions by each convolutional kernel to extract their peak distribution; by increasing the number of convolutional layers, the cooperative strength between EEG signals corresponding to different brain regions is gradually refined.

[0023] Furthermore, the EEG data analysis module utilizes the deep learning algorithm to analyze the multi-dimensional EEG signal features extracted by the data analysis module, in order to extract abnormal EEG signals; specifically,

[0024] The EEG signal feature sequence, after being divided into time windows at different time scales set by the data analysis module, is input into the convolutional kernel layer, and the intensity change and peak distribution of the EEG signal in different time periods are output features.

[0025] The output features of the convolutional kernel layer are input into the deep convolutional layer, which outputs the coordination strength and abnormal EEG signals between various brain regions.

[0026] Furthermore, the injury assessment module integrates the abnormal EEG signals output by the EEG data analysis module and the characteristics of various brain regions to locate the injury area and assess the type and severity of injury in the target patient; correspondingly:

[0027] Based on the characteristic types of abnormal electroencephalogram (EEG) signals, determine the nature and type of brain injury;

[0028] By using the electrode locations corresponding to abnormal EEG signals, the corresponding functional distribution areas of the brain can be mapped to determine the damaged areas.

[0029] By comparing abnormal EEG signals with corresponding healthy standard signals, the frequency band power deviation can be analyzed and calculated to quantify the degree of damage.

[0030] (iii) Beneficial effects:

[0031] Compared with the prior art, this invention has the following beneficial effects:

[0032] This invention utilizes sound waves and electrical impulses to stimulate limb muscles, inducing specific event-related potentials and motor cortical activity rhythms. Compared to simple static EEG detection, it can more specifically reveal the functional response states of different brain regions. By simultaneously capturing short-term and long-term EEG dynamic changes through multi-scale time windows, and calculating frequency domain coherence and phase lag indices within the same time scale, a connectivity matrix reflecting functional coupling between brain regions is constructed, providing quantitative evidence for the analysis of abnormal connectivity patterns and the localization of brain regions. By comparing abnormal EEG signals with healthy standard signals, and comprehensively considering frequency band power deviation and the degree of functional connectivity loss, a quantitative assessment of the type, location, and severity of injury is formed, generating a diagnostic report that can be directly used for battlefield medical decision-making.

[0033] The system can rely on a portable EEG acquisition helmet, a portable pulse stimulation device, and a chip-level handheld terminal. The entire process, from stimulation induction to result analysis, can be completed within minutes. The overall equipment is lightweight and suitable for deployment in field scenarios. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of a rapid field brain injury diagnosis system based on electroencephalogram (EEG) data provided in an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of the electrode patches when the target wounded person wears an EEG acquisition helmet, as described in an embodiment of the present invention.

[0036] Figure 3 This is a schematic diagram of the target brain regions in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the electroencephalogram (EEG) signals of various bands that need to be acquired in a rapid field brain injury diagnosis system based on EEG data provided in an embodiment of the present invention.

[0038] Figure 5 This is a schematic diagram of the EEG power density that needs to be measured in a rapid field brain injury diagnosis system based on EEG data provided in an embodiment of the present invention.

[0039] In the picture:

[0040] 100. External Stimulation Module; 200. Data Acquisition Module; 300. Data Preprocessing Module; 400. EEG Data Analysis Module; 500. Injury Assessment Module. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0043] Furthermore, the terms "first" and "second" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0044] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0045] During field training, soldiers may suffer potential brain injuries due to blast waves, falls, or other accidents.

[0046] Combination Figures 1 to 5 As shown, this invention provides a rapid field brain injury diagnostic system based on electroencephalogram (EEG) data. This system utilizes portable EEG devices, allowing medical personnel to quickly reach the scene and induce EEG responses in the injured person through external stimulation (sound waves and muscle electrical impulses). The system collects EEG signals in real time, analyzes and extracts abnormal features using a deep learning model, locates the damaged brain region, and finally outputs the injury type, severity, and brain region location, providing a scientific basis for on-site medical decisions, such as whether to evacuate the injured person or provide immediate on-site treatment.

[0047] For reference here Figure 1 First, there is the external stimulation module 100, which stimulates the target injured person. This module is used to induce specific brain region activity responses in the target injured person through progressively increasing intensity sound wave stimulation and limb muscle electrical impulse stimulation, thereby assessing sensory function, motor function, and brain region activity. It is important to note that the external stimulation here is applied using a time-sequential strategy, that is, after multiple sound wave stimuli are completed, limb muscle electrical impulse stimulation is initiated, and each set of stimuli is set with increasing intensity to avoid excessive interference.

[0048] Now Figure 3 For example, based on the brain regions marked in the figure: frontal lobe 1, temporal lobe 2, parietal lobe 3, central cortex 4, and occipital lobe 5, the external stimulation module 100 induces electrical activity responses in specific brain regions by designing sound waves and limb muscle electrical pulse stimulation of different frequencies, thereby identifying brain region damage.

[0049] Regarding the selection of sound wave frequency, sound waves stimulate the auditory cortex in temporal lobe 2 through the auditory pathway, and at the same time activate frontal lobe 1, which is related to memory and attention, through functional connections between brain regions. In some embodiments of the present invention, low frequencies (such as 40 Hz) can be selected to induce the P300 band of the parietal lobe and central cortex, and high frequencies (such as 200 Hz) can be selected to activate the visual cortex of the occipital lobe.

[0050] Regarding the order of application of sound wave stimulation, in some embodiments, the sound wave frequency can be gradually increased according to the increasing order of sound wave intensity, with each level lasting 1 second, to obtain the activity level and time response characteristics of brain regions at different frequencies.

[0051] Regarding electrical impulse stimulation of muscles, the electrical impulses are transmitted to the central nervous system via sensory nerves, and the central cortex response can reflect the integrity of the tactile pathway. In some embodiments, low-frequency (e.g., 10 Hz) stimulation can induce motor-related EEG rhythms, such as in the central cortex, while mid-frequency (e.g., 50 Hz) stimulation can be selected to activate the sensory cortex.

[0052] Regarding the location of electrical impulse stimulation of limb muscles, in some embodiments, electrical impulse stimulation can be applied to the upper limb, lower limb, or both limbs of the target injured person simultaneously to detect the motor cortex's ability to process tactile stimuli.

[0053] In summary, the task of the external stimulation module 100 is to guide different functional areas of the target patient's brain to generate measurable response signals through its designed time-series stimulation strategy. Specifically, it first selects to apply sound wave stimulation to stimulate brain regions associated with auditory processing, cognitive responses, and event-related potentials (ERPs).

[0054] The frequency, intensity, and duration of the sound stimulation are gradually increased according to a preset intensity level. For example, a short pulse of low-frequency (e.g., 500Hz) low-volume (40dB) sound is used first, followed by a gradual increase to mid-frequency (1000Hz) medium-volume (60dB) and high-frequency (4000Hz) high-volume (80dB) sound, with each frequency applied sequentially according to a preset volume gradient. Such sound stimulation can induce event-related potentials (ERPs) in brain regions associated with auditory perception and attentional responses (e.g., temporal lobe, parietal lobe), facilitating the assessment of auditory perception and information processing speed.

[0055] After the various intensity levels of sound wave stimulation are applied, the external stimulation module 100 enters the second stage of stimulation, namely, applying electrical pulse stimulation to the injured person's limb muscles. This type of stimulation is mainly targeted at the extremities, such as specific muscle groups in the forearm or calf. The electrical pulse signal is transmitted to the target muscle through electrodes, triggering a slight contraction, thereby activating the neural activity of the brain's motor cortex and related sensorimotor pathways.

[0056] Electrical pulse stimulation is also performed according to the principle of increasing intensity. For example, a short pulse with low amplitude (5mA) is first applied to the forearm flexor muscles, which can only cause slight muscle twitching. Then, the pulse is gradually increased to medium amplitude (10mA) and high amplitude (15mA) to trigger significant contraction, so as to induce activity signals in the corresponding motor cortex. This time-sequential application strategy not only helps to detect the response patterns of motor functional pathways under different stimulation intensities, but also minimizes muscle fatigue or neural inhibition effects caused by overstimulation.

[0057] The aforementioned external stimulation module 100 can be implemented using a portable stimulation device. Its design integrates a small acoustic stimulator with a limb muscle electrical pulse stimulator, offering multi-frequency and intensity adjustment capabilities. Besides allowing users to select between acoustic or electrical pulse stimulation as needed within the device to meet diverse scenario requirements, a built-in battery can also be considered to support extended operation, making it suitable for environments without electricity in the field.

[0058] In summary, it can be understood that sound waves and limb muscle electrical pulse stimulation are applied in a preset order of increasing intensity. Sound waves mainly stimulate the frontal, temporal, and occipital lobes, while limb muscle electrical pulse stimulation focuses on detecting the central cortex and parietal lobe.

[0059] Regarding the data acquisition module 200, this module is used to acquire multi-band electroencephalogram (EEG) signals of the injured person in real time under acoustic stimulation and limb muscle electrical pulse stimulation, and to ensure that the quality of the acquired signals meets the analysis requirements.

[0060] For details, please refer to the following: Figure 2 The data acquisition module 200 is specifically connected to a multi-channel EEG device, that is, to an EEG electrode cap, in which each electrode covers key brain regions, namely the aforementioned frontal lobe 1, temporal lobe 2, parietal lobe 3, central cortex 4, and occipital lobe 5, to acquire multi-channel EEG data.

[0061] In this system, signals in each frequency band are localized to corresponding brain regions using EEG electrodes, forming a brain region-signal intensity mapping table. For example, sound wave stimulation induces a P300 response signal in the frontal lobe, which is then recorded. , and Electroencephalogram (EEG) signals in the [band name missing] band; electrical impulse stimulation of limb muscles induced motor rhythm signals in the central cortex, recorded in [the context missing]. and Amplitude changes of the band EEG signal.

[0062] To make it easier to understand, the following table 1 will be used to further illustrate this.

[0063] Table 1. Brain regions and functional analysis corresponding to each signal band.

[0064]

[0065] The aforementioned data acquisition module 200 can be implemented using a portable EEG acquisition helmet. Its design is lightweight, employing a flexible electrode array and an embedded wireless data transmission module. It is important to note that the flexible electrodes must ensure signal quality to adapt to various head shapes and complex environments. Furthermore, this EEG acquisition helmet requires no professional operation; data acquisition can begin with simple adjustments.

[0066] In summary, it is understandable that the acquisition module provides high-quality multi-band signals for the subsequent preprocessing and analysis modules. The correspondence between brain regions and functions of different band signals is clear, laying the foundation for diagnosing the region and severity of brain injury in casualties and ensuring the reliability and adaptability of the system in the field environment.

[0067] Regarding the data preprocessing module 300, this module is used to perform artifact removal, feature extraction, and spatiotemporal analysis on the acquired EEG signals to ensure the quality and validity of the signals and to provide high-precision input for subsequent analysis modules.

[0068] Specifically, artifact removal is used to eliminate interference from non-EEG signals such as electromyography (EMG) interference, power line noise, and eye movement artifacts. In some embodiments, noise signals outside the EEG frequency band (1Hz-100 Hz) are removed by filtering, i.e., low-frequency noise filtering (<1 Hz) to avoid drift effects, and high-frequency noise filtering (>100 Hz) to remove EMG interference.

[0069] In addition, independent source signals in the original EEG signal are separated, and artifact components such as eye movement, blinking, and electromyography interference are removed.

[0070] Regarding the data preprocessing module 300's feature extraction of EEG signals to reflect the activity state of different brain regions, specifically, in some embodiments of the present invention, time windows of different time scales are first divided, such as setting multiple time scales like 150ms, 250ms, 500ms, and 800ms, to capture the dynamics of short-term and long-term EEG signals respectively.

[0071] It's important to note that the aforementioned multi-scale time windows are pre-defined by the data preprocessing module. Then, for each segment of EEG signal acquired after each external stimulus, these windows are used for sliding processing to extract features of that band at different time scales. In other words, the same EEG signal is placed within multiple time windows of varying lengths to extract features at different time scales. This multi-time-scale extraction method helps enhance the perception of complex brain movement patterns during the processing of neural signals.

[0072] More specifically, each external stimulus triggers a segment of EEG response, and this segment of EEG signal constitutes an analysis unit. Now, assuming the signal length acquired after a stimulus is 1000ms, we will analyze it in parallel using multiple pre-defined time windows. Specifically, we will perform sliding processing on this segment of EEG signal using each time window length, extracting temporal and frequency domain features within each time window, thereby extracting features at different time scales from this segment of EEG signal.

[0073] Regarding the temporal characteristics of EEG signals, the features that need to be extracted include mean amplitude and peak distribution. Specifically, mean amplitude can reflect signal strength and brain region activity; peak distribution of each EEG signal can detect high-frequency oscillations and abnormal activity.

[0074] Regarding feature extraction in the frequency domain, please refer to [reference needed]. Figure 4 That is, first obtain different frequency bands ( , , , and The spectrum signal is then analyzed, and its power distribution is then analyzed, that is, as shown in the example. Figure 5 The power spectral density (PSD) of each EEG signal is measured to reflect the characteristics of brain region activity.

[0075] Finally, on the same time scale, the correlation coefficient is calculated by determining the phase lock value, which measures whether two signals maintain a stable phase difference over time, and by calculating the frequency coherence coefficient between brain regions to reflect the synchronization and coordination between signals. The formula for calculating the correlation coefficient is as follows:

[0076] ;

[0077] in, and These are the power spectral densities of their respective EEG signals. Let be the cross-power spectral density of the two EEG signals.

[0078] For ease of understanding, please refer to Table 2 below to understand the explanation of the coherence and function between different brain regions.

[0079] Table 2. Correlation and Corresponding Functions among Different Brain Regions

[0080]

[0081] In summary, it can be understood that the data preprocessing module 300 ensures signal quality by removing artifacts, providing clean input data for subsequent analysis modules, extracts key brain region activity information and abnormal features through feature extraction, and finally provides the coordinated activity status of the injured brain injury area based on brain region coherence analysis, thus supporting the injury assessment module 500.

[0082] Regarding the EEG data analysis module 400, this module performs in-depth analysis on the extracted multi-feature EEG signal data output by the preprocessing module, extracts abnormal EEG signals, identifies key temporal dynamics and brain region functional characteristics, and provides an analytical basis for the subsequent damage assessment module 500.

[0083] First, the framework of the deep learning algorithm in the deep learning model is explained. This deep learning algorithm mainly consists of multi-scale convolutional layers and deep convolutional layers. Among them, the multi-scale convolutional layers are used to capture the dynamic features of EEG signals at different time scales and identify the change patterns at key time points. That is, parallel convolution extraction is performed on EEG sequences under multiple time windows to capture the rhythmic changes, rising or falling edges, and oscillating structures of the spectrum of the signal at different durations.

[0084] Deep convolutional layers are used to extract spatial distribution features between different brain regions, and further analyze the coupling strength and abnormal peak distribution between EEG signals in different brain regions.

[0085] More specifically, regarding multi-scale convolutional layers, different sizes of convolutional layers, such as 3, 5, and 7 time steps, are used to extract features at different time scales. To elaborate further, small convolutional kernels are used to extract short-term fluctuation features to capture high-frequency components, while large convolutional kernels are used to extract long-term trend features to capture low-frequency components.

[0086] The EEG signal feature sequences, divided into different time scales and time windows by the data analysis module, are input into the multi-scale convolutional layer mentioned above. Through translation convolution operation, global coverage of temporal dynamics is achieved, and finally, the intensity changes and peak distribution of the EEG signal in different time periods are output as output features.

[0087] Regarding deep convolutional layers, two-dimensional convolution is used to focus on the interaction features between different brain regions, and each convolutional kernel aggregates a set of connection strengths between brain regions to extract their peak distribution. In some embodiments of the present invention, the number of convolutional layers is increased to progressively refine the cooperative strength between EEG signals corresponding to different brain regions.

[0088] The output features from the previous multi-scale convolutional layer, which can also be understood as the output feature map of the scale convolutional layer, are input into the depth convolutional layer, which outputs the coordination strength and abnormal EEG signals between various brain regions, which can also be understood as the output depth feature map.

[0089] The aforementioned EEG data analysis can be achieved using a handheld terminal with an embedded high-efficiency chip, which supports real-time preprocessing of EEG data and the operation of deep learning models. The handheld terminal employs an embedded chip design, which not only optimizes the operating efficiency of deep learning algorithms but also enables rapid processing of EEG data even in low-power conditions. It is important to note that the device casing must be waterproof and shockproof to withstand harsh outdoor environments.

[0090] In summary, it can be understood that the EEG data analysis module 400, as the core analysis module of the system, extracts high-dimensional features and labels abnormal signals. The temporal dynamic capture of the multi-scale convolutional layers of the deep learning algorithm supports the real-time requirements of rapid diagnosis, and the spatial distribution analysis of the deep convolutional layers provides an important basis for lesion localization.

[0091] Finally, there is the damage assessment module 500. This module integrates the abnormal signals and brain region characteristics output by the EEG data analysis module 400 to locate the damaged area, assess the type and severity of the damage, and generate a detailed diagnostic report.

[0092] The main tasks of this module can be simply understood as identifying the type of brain injury, locating the injured area, and determining the degree of injury. Specifically, brain injury type identification is based on the characteristic types of abnormal signals to determine the nature of the brain injury, such as whether the injury is functional or structural.

[0093] More specifically, the abnormal signals originate from the abnormal EEG signals extracted by the EEG data analysis module 400, including abnormal peak distributions, regions of reduced coherence, and abnormal frequency band activity. Regarding the mapping method within the damaged areas, in some embodiments, an electrode-brain region correspondence table is used to map the electrode locations marked with abnormal signals to functional brain regions.

[0094] Now, combining the deep learning algorithm in the EEG data analysis module 400, the spatial convolutional layer is used to analyze and extract the abnormal coupling patterns of the brain region functional connectivity matrix. The brain region functional connectivity matrix is ​​a two-dimensional structural representation that describes the functional coupling relationship between different brain regions and is often used to reflect the intensity and synchronicity of information interaction between brain regions.

[0095] Specifically, after calculating the frequency domain coherence and phase lag index between each pair of brain regions, the above indicators are calculated for multiple brain regions within each time window, forming a symmetric matrix. Now, assume the number of brain regions is... Then construct a The connection matrix, in which the first... Items represent brain regions and Functional connectivity strength within a certain frequency band and time window.

[0096] The goal of the EEG data analysis module 400 is to automatically extract abnormal connectivity features related to brain injury from a large number of connectivity matrices. Its recognition logic relies on the spatial modeling capabilities of the convolutional neural network (CNN) structure. The multi-scale convolutional layers within it can identify damage signals such as "sudden loss of connectivity in a brain region within a specific time window" or "periodic decrease in coupling synchronicity" by extracting the dynamic changes in the connectivity matrix or EEG features over time. In a normal state, the connectivity matrix exhibits a certain degree of symmetry and density. Therefore, the spatial convolutional layers can extract abnormal coupling patterns by identifying features such as significantly weakened row / column connections, regional isolation, and sparse global connectivity in a damaged area.

[0097] The deep learning model in the EEG data analysis module 400 learns by comparing the connection matrix features of healthy samples in the training set, and establishes the recognition features shown in Table 3 below.

[0098] Table 3 Abnormal Coupling Patterns and Corresponding Identification Features

[0099]

[0100] Suppose a trainee is mildly affected by a blast wave during training, with no obvious external injuries, but exhibits temporary confusion and difficulty concentrating. The information extracted under the aforementioned sound wave stimulation... The waveband decreased significantly in the frontal lobe region, and the event-related potential waveform was delayed under muscle electrical impulse stimulation. The response of the waveband is weak in the central cortex, and the amplitude of motion-related potentials decreases. The deep learning model output shows a time-dynamic variation pattern. Frequency band response hysteresis; brain region functional connectivity matrix showing the connection between the prefrontal and parietal lobes The band functional connectivity strength is decreased, and the low-frequency coherence is weakened. The system judges it to be a mild concussion, without focal lesions, but with short-term neurological dysfunction.

[0101] The aforementioned health standard signals are essentially the characteristic statistical distribution of EEG data from healthy individuals, or reference data from patients with similar injuries. Scoring models can be built using statistical data, and scoring rules within the model can be constructed based on the degree of deviation of the features. For example:

[0102] Mild: Power deviation less than 20%, synergy slightly below the threshold;

[0103] Moderate: Power deviation is between 20% and 50%, and synergy is significantly reduced;

[0104] Severe: Power deviation > 50%, shock-type brain injury.

[0105] In summary, it is understandable that the injury assessment module 500 provides a scientific basis for on-site medical decisions (such as whether to evacuate the wounded or provide immediate on-site treatment) by outputting the type, severity, and location of the brain region involved in the injury. Furthermore, the output diagnostic data can be uploaded to a military medical platform for further evaluation or long-term tracking of rehabilitation progress. Moreover, the system's various modules are functionally implemented through the rational design of portable acquisition, stimulation, and processing devices, ensuring both functional integrity and portability, making it highly suitable for application in military training scenarios. This design not only lowers the barrier to entry for equipment use but also meets diverse on-site needs, effectively improving the efficiency and accuracy of medical treatment.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The scope of patent protection of the present invention shall be determined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention shall also be included within the scope of protection of the present invention.

Claims

1. A rapid field brain injury diagnostic system based on electroencephalogram (EEG) data, characterized in that, include: The external stimulation module is used to apply sound wave stimulation and muscle electrical pulse stimulation to the target injured person in stages. Each stimulation has multiple intensity levels and is applied independently level by level. The sound wave stimulation is used to induce event-related EEG signal responses, and the muscle electrical pulse stimulation is used to induce EEG signal responses related to motor cortex activity. The data acquisition module is used to acquire real-time data on the target wounded person under the stimulation of the sound waves. , and Electroencephalogram (EEG) signals in the band, and under the stimulation of the muscle electrical impulses and Electroencephalogram (EEG) signals in the specified frequency band; The data preprocessing module removes artifacts from the EEG signals of each band and uses a multi-time-scale time window to extract the time-domain and frequency-domain features of the EEG signals of each band. Within the same time window, it calculates the frequency-domain coherence and phase lag index between different EEG signals and constructs a brain region connectivity matrix that reflects the functional coupling state. The EEG data analysis module uses a deep learning model to jointly analyze the extracted time-domain features, frequency-domain features, and brain region connectivity matrix to identify abnormal EEG signals induced by external stimuli. Specifically, it captures the dynamic changes of EEG signals within different time windows through parallel convolution kernels in multi-scale convolutional layers, and extracts abnormal coupling patterns from the brain region connectivity matrix through spatial convolutional layers. The injury assessment module determines the type of brain injury based on the abnormal EEG signals and their coupling patterns, locates the brain injury area based on the acquisition location of the abnormal EEG signals, compares them with healthy standard signals, quantifies the frequency band power deviation and the degree of functional connectivity loss, grades and assesses the degree of injury, and generates a diagnostic report including the injury type, injury area, and injury degree.

2. The rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 1, characterized in that, The external stimulation module is used to induce specific brain region activity responses in the target injured person through progressively increasing intensity sound wave stimulation and limb muscle electrical pulse stimulation, in order to assess perceptual function, motor function, and brain region activity; wherein, sound wave stimulation is used to induce event-related potentials to assess perceptual and cognitive functions, and limb muscle electrical pulse stimulation is used to stimulate motor cortex-related potentials to reflect the activity of the motor cortex.

3. The rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 1, characterized in that, The data acquisition module is used to acquire relevant EEG signals of the injured person in real time under the sound wave stimulation and limb muscle electrical pulse stimulation. Specifically, the data acquisition module is connected to a multi-channel EEG electrode cap, which covers the frontal lobe, parietal lobe, temporal lobe, occipital lobe and central cortex of the target injured person.

4. The rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 3, characterized in that, The data acquisition module acquires data under acoustic stimulation. , and EEG signals in the band, among which, The EEG signals in the band reflect the inhibition of brain region activity. The EEG signals in this band are related to attention. Electroencephalogram (EEG) signals in certain bands reflect the coordination of brain regions; Obtaining electrical impulse stimulation of limb muscles and EEG signals in the band, among which, The EEG signals in the band reflect the inhibitory state of the sensory cortex. Electroencephalogram (EEG) signals in certain bands reflect the readiness of motor function.

5. A rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 4, characterized in that, The data preprocessing module first performs artifact removal on all EEG signal data acquired by the data acquisition module. Specifically, the data preprocessing module uses bandpass filtering to remove DC drift and high-frequency electromyographic noise, notch filtering to remove power frequency interference, and independent component analysis to identify and remove non-EEG related signals, including electrooculography and electromyography.

6. The rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 5, characterized in that, After removing artifacts from the acquired EEG data of each frequency band, the data preprocessing module pre-divides the data into multiple time windows of different and increasing lengths. For each preprocessed EEG signal acquired after each external stimulus, the EEG signal is subjected to parallel sliding processing using all preset time windows of different lengths. Under the sliding processing of each time window, its corresponding time domain features and frequency domain features are extracted. The time domain features include amplitude peak value, mean and variance, and the frequency domain features include power spectral density.

7. A rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 5, characterized in that, Within the time window of the same duration, frequency domain analysis is performed on any two EEG signals of the same band acquired under external stimuli of different intensities. The coherence coefficient of the two signals in the corresponding band is calculated to reflect the coherence of the activities of the two corresponding brain regions. The phase lag index is calculated by extracting the sequential phase of the two signals. Based on the calculated frequency domain coherence and phase lag index, a corresponding symmetric matrix is ​​generated for each external stimulus, which is used as the brain region connectivity matrix.

8. The rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 1, characterized in that, Regarding the deep learning model used in the EEG data analysis module, different sizes of convolutional kernels are set to extract EEG signal features at different time scales; among them, small-sized convolutional kernels are used to extract short-term fluctuation features and capture high-frequency components; large-sized convolutional kernels are used to extract long-term trend features and capture low-frequency components. Deep convolutional layers in deep learning algorithms use two-dimensional convolution to focus on the interaction features between different brain regions, and aggregate the connection strength between a group of brain regions by each convolutional kernel to extract their peak distribution; by increasing the number of convolutional layers, the cooperative strength between EEG signals corresponding to different brain regions is gradually refined.

9. A rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 8, characterized in that, The EEG data analysis module utilizes the deep learning algorithm to analyze the multi-dimensional EEG signal features extracted by the EEG data analysis module in order to extract abnormal EEG signals; specifically, The EEG signal feature sequence, after being divided into time windows at different time scales by the EEG data analysis module, is input into a multi-scale convolutional layer, and the intensity change and peak distribution of the EEG signal in different time periods are output as output features. The output features of the multi-scale convolutional layer are input into the deep convolutional layer, which outputs the coordination strength and abnormal EEG signals between various brain regions.

10. A rapid field brain injury diagnostic system based on electroencephalogram (EEG) data according to claim 8, characterized in that, The injury assessment module integrates abnormal EEG signals and characteristics of various brain regions output by the EEG data analysis module to locate the injury area and assess the type and severity of injury in the target patient; correspondingly: Based on the characteristic types of abnormal electroencephalogram (EEG) signals, determine the nature and type of brain injury; By using the electrode locations corresponding to abnormal EEG signals, the corresponding functional distribution areas of the brain can be mapped to determine the damaged areas. By comparing abnormal EEG signals with corresponding healthy standard signals, the frequency band power deviation can be analyzed and calculated to quantify the degree of damage.

Images