An electroencephalogram feature extraction method combining ABR and ERP waveforms

Abstract

The embodiment of the application discloses a kind of electroencephalogram feature extraction methods combined with ABR and ERP waveform, comprising: obtaining left and right electroencephalogram waveform derived from the same test, wherein electroencephalogram waveform includes ABR waveform and ERP waveform;Respectively for left and right ABR waveform and left and right ERP waveform, extract the time feature, amplitude feature and density feature between unilateral waveform and bilateral waveform, and fatigue feature;The time feature, amplitude feature, density feature and fatigue feature of the left and right ABR waveform and the left and right ERP waveform are fused as the electroencephalogram feature for reflecting neural activity.This embodiment can extract multi-dimensional effective signal features from ABR and ERP electroencephalogram waveform, to comprehensively reflect the neural function state of brain and brain stem.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and in particular to a method for extracting electroencephalogram (EEG) features by combining ABR and ERP waveforms. Background Technology

[0002] In recent years, with the rapid development of electroencephalogram (EEG) technology and its derivative methods, using objective indicators of physiological signals to assist in the prediction of mental abnormalities has become a hot technology.

[0003] Among them, ABR (Auditory Brainstem Response) signals and ERP (Event-Related Potentials) signals have been widely used to assess the function of the central nervous system. ABR refers to recording the brain and brainstem's response to auditory stimuli, which can reflect the function of the brainstem's auditory conduction pathway; ERP, on the other hand, analyzes the electroencephalogram (EEG) signals evoked by specific stimuli (such as cognitive, emotional, or behavioral tasks) to obtain information reflecting the brain's cognitive function or emotional processing.

[0004] However, there is currently a lack of a method to extract multi-dimensional and effective signal features from ABR and ERP EEG waveforms to comprehensively reflect the neurological functional state of the brain and brainstem, thereby providing a more accurate and objective basis for predicting mental status. Summary of the Invention

[0005] This invention provides a method for extracting EEG features by combining ABR and ERP waveforms to solve the above-mentioned technical problems.

[0006] In a first aspect, embodiments of the present invention provide a method for extracting electroencephalogram (EEG) features by combining ABR and ERP waveforms, including:

[0007] Obtain EEG waveforms from the left and right sides of the same test, including ABR waveforms and ERP waveforms;

[0008] For the left and right ABR waveforms and the left and right ERP waveforms respectively, extract the time features, amplitude features, density features, and fatigue features between the single-sided waveform and the double-sided waveform;

[0009] The temporal, amplitude, density, and fatigue characteristics of the left and right ABR and ERP waveforms are fused together to form EEG features that reflect neural activity.

[0010] Secondly, embodiments of the present invention provide a mental health prediction model construction system that combines ABR and ERP features, including:

[0011] The waveform acquisition module is used to acquire left and right EEG waveforms with labeled data. The waveform types include ABR waveforms and ERP waveforms, and the labeled data includes normal and at least one type of mental abnormality.

[0012] The feature extraction module is used in the method described in the above embodiments to extract EEG features reflecting neural activity from the left and right EEG waveforms from the same test, respectively.

[0013] The model building module is used to train a machine learning classification model using EEG features with normal labels and labels for each abnormal type. The trained model is used to predict the degree to which the new left and right EEG waveforms tend to be labeled for each abnormal type based on the EEG features of the new left and right EEG waveforms.

[0014] Thirdly, embodiments of the present invention also provide a mental health prediction system combining ABR and ERP features, comprising:

[0015] The waveform acquisition module is used to acquire left and right EEG waveforms from the same test, including ABR waveforms and ERP waveforms.

[0016] The feature extraction module is used to extract time features, amplitude features, density features, and fatigue features between unilateral and bilateral waveforms for the left and right ABR waveforms and the left and right ERP waveforms respectively; and to fuse the time features, amplitude features, density features, and fatigue features of the left and right ABR waveforms and the left and right ERP waveforms as EEG features to reflect neural activity.

[0017] The prediction module is used to use a trained machine learning classification model to predict the degree to which the left and right EEG waveforms tend to each abnormality type label based on the EEG features, providing auxiliary information for medical decision-making.

[0018] In summary, this invention provides a method for extracting EEG features by combining ABR and ERP waveforms. By simultaneously acquiring and analyzing auditory evoked brainstem responses (ABR) and event-related potentials (ERPs), it effectively reduces the cumbersome steps of multiple devices and tests, improves detection efficiency and subject compliance, and ensures high-precision alignment of data at the same time reference. This integrated fusion of ABR and ERP breaks the traditional model of independent detection of the two, facilitating a more comprehensive understanding of the connection mechanism between brainstem and cortical cognitive functions. Furthermore, this embodiment not only focuses on the brainstem auditory conduction function characteristics reflected by ABR but also integrates key features of ERP in cognitive or emotional processing, achieving a holistic assessment from peripheral auditory pathways to higher cortical functions. Specifically, this embodiment captures data characteristics from multiple dimensions such as time, amplitude, density, and fatigue, and respectively mines effective features closely related to neural activity in ABR and ERP. By feature fusion, it comprehensively reflects the state of neural activity. Compared with traditional detection methods that rely too much on subjective scales, this in-depth feature that combines information from the brainstem and cortical levels is more objective and accurate, helps to fully reveal the multi-level characteristics of the mechanism of mental abnormalities, and provides richer auxiliary information for medical decision-making.

[0019] Based on the above methods, this embodiment also provides a mental prediction model construction system that combines ABR and ERP features and a mental prediction system that combines ABR and ERP features. It uses machine learning or statistical methods to perform fusion analysis on multi-dimensional EEG features, improves the accuracy of identifying mental abnormalities, and can also support further subtyping prediction, helping doctors to grasp the condition in the early stages and formulate personalized intervention measures. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an EEG feature extraction method combining ABR and ERP waveforms provided in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the structure of a mental prediction model construction system that combines ABR and ERP features, provided by an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of the structure of a mental prediction system combining ABR and ERP features provided in an embodiment of the present invention;

[0024] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0026] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for 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 the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0027] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] This invention provides a method for extracting EEG features by combining ABR and ERP waveforms. To illustrate this method, the conventional applications of ABR and ERP are first analyzed. Specifically, although both ABR and ERP are non-invasive EEG detection methods with high temporal resolution, in the prior art, they are mostly used independently or separately in hearing assessment and cognitive function research. However, mental disorders often involve multi-dimensional neurological dysfunctions, including different levels such as the sensory system, cognitive function, and emotion regulation. Using only a single neurophysiological indicator (such as ABR or ERP alone) is insufficient to cover the full picture of the pathological process and to meet the needs of early prediction. Therefore, this application integrates physiological features at the brainstem level and the cortex level, providing multi-angle neurological function assessment based on the mutually verifying and complementary information of the two, and capturing waveform details for the type of mental disorder to be predicted, constructing a variety of novel sub-features to improve the accuracy of neurological function assessment and provide more auxiliary information for medical decision-making.

[0029] Specifically, Figure 1 This is a flowchart illustrating a method for extracting electroencephalogram (EEG) features by combining ABR and ERP waveforms, as provided in an embodiment of the present invention. This method is executed by an electronic device or system module, such as... Figure 1 As shown, the method specifically includes:

[0030] S110. Obtain left and right EEG waveforms from the same test, including ABR waveforms and ERP waveforms.

[0031] In this context, "same test" refers to a complete set of tests performed on the same subject. A set of tests may include multiple stimuli, thereby generating both ABR and ERP waveforms. For each stimulus, dual-channel EEG waves from both sides of the brain are collected simultaneously, yielding left and right ABR and ERP waveforms, respectively.

[0032] This embodiment first acquires the left and right ABR and ERP waveforms from the same subject in a complete test, which serve as the data source for subsequent extraction of the subject's EEG features. Optionally, the left and right ABR and ERP waveforms can be sampled synchronously using the same hardware platform to ensure high-precision alignment on the time axis.

[0033] Specifically, ABR (Auditory Response) refers to the neural electrical activity response generated in the cochlea and brainstem auditory pathways after external auditory stimuli (such as clicks, short sonic booms, etc.) are applied to the human ear. ABR reflects the functional state of the brainstem auditory system conduction pathways and is often used in clinical hearing testing and early detection of neurological abnormalities. ERP (Empirical Performance Response) refers to the relatively time-locked potential changes generated in the human cerebral cortex under the influence of specific stimuli or events (such as visual, auditory, emotional, or cognitive tasks). ERP reflects cognitive processes, emotional processing, and higher neural activities, and is an important indicator for studying the mechanisms of mental abnormalities. Therefore, since the stimulation methods for inducing the two signals are different, ABR stimulation and ERP stimulation (such as clicks, visual images, cognitive tasks, etc.) can be triggered simultaneously or sequentially by a controller or software module, and the ABR and ERP waveforms can be acquired through electrodes in corresponding regions and bands.

[0034] For example, taking ABR as an example, a total of 1024 frame pairs (2048 frames in total) were collected, with each frame corresponding to 256 samples (10ms, 25.6kHz). Each frame was arranged in order according to the frame number to obtain the left and right waveforms respectively. The set of waveforms on both sides is called the left and right ABR waveforms.

[0035] S120 extracts the time characteristics, amplitude characteristics, density characteristics, and fatigue characteristics of the single-sided and double-sided waveforms for the left and right ABR waveforms and the left and right ERP waveforms, respectively.

[0036] After obtaining the two dual-channel waveforms, this embodiment processes the left and right ABR waveforms and the left and right ERP waveforms separately, extracting multi-dimensional features of the two waveforms, including time features, amplitude features, density features and fatigue features. Each feature category includes at least one sub-feature. These sub-features were finally determined after continuous analysis, optimization and verification, and can provide key and effective information for the detection of mental abnormalities.

[0037] In one specific implementation, the currently processed left and right EEG waveforms (left and right ABR waveforms or left and right ERP waveforms) are referred to as the current left and right EEG waveforms. After obtaining the raw waveforms of each side, the waveforms are first preprocessed to convert the waveform signals into clear, standardized data, facilitating subsequent feature extraction. Optionally, the preprocessing process includes:

[0038] Artifact removal: Removes corrupted or invalid signal frames, such as removing the first 100 frames of signal;

[0039] Time trimming: Convert time into sample index;

[0040] DC offset elimination: Eliminates baseline drift and electrode offset;

[0041] High-pass filtering: eliminates low-frequency noise and baseline drift while preserving neural activity;

[0042] Low-pass filtering: removes high-frequency noise while preserving neural responses;

[0043] Quality segmentation: Select the highest quality frame, sort all frames by peak-to-peak amplitude, and retain the middle part (with the best signal-to-noise ratio);

[0044] Quality assessment: Calculate the signal quality metrics. If the quality metrics meet the requirements, the signal can be used as a subsequent waveform. Optional, quality metrics... Here, averageValue refers to the average amplitude of the final waveform, and 100 is a scaling factor (which can be flexibly determined according to the range of data values).

[0045] For example, taking a raw one-sided ABR waveform as an example, the raw waveform consists of 2000 frame pairs, each frame is 10ms long and sampled at 25.6kHz. The raw waveform contains artifacts, noise, and baseline drift. After step-by-step preprocessing, the changes in each step are as follows:

[0046] 1. Artifact removal: 2000 frame pairs → 1800 frame pairs

[0047] 2. Pruning: 10ms frame → 5ms valid signal

[0048] 3. DC offset: Remove +1000μV baseline → center at 0μV

[0049] 4. FIR filter: Removes the baseline from 0-400Hz → Smooths the baseline

[0050] 5. Butterworth filter: Removes baselines >3000Hz → smooths neural responses

[0051] 6. Segmentation: 1800 frame pairs → 1024 frame pairs for optimal quality

[0052] 7. Quality: Calculate Q = 0.85 (good quality), which is a clean, standardized ABR waveform.

[0053] This preprocessing process ensures that only high-quality, standardized neural response data enters the analysis engine, significantly improving the reliability of feature extraction.

[0054] Then, based on the preprocessed current left and right EEG waveforms, the following waveform features are extracted respectively:

[0055] The first category is time features. For ease of distinction and description, the sub-features belonging to time features will be referred to as the first time feature, the second time feature, and so on, in the order of their appearance.

[0056] 1. For each moment of each unilateral waveform in the current left and right EEG waveforms, perform the following operations respectively: delineate a time window covering the current moment in the normal baseline waveform, determine a time point within the time window whose amplitude is closest to the amplitude of the current moment in the current unilateral waveform, and calculate the time offset between the time point and the current moment; calculate the root of the sum of squares of the time offsets corresponding to each moment to obtain the first time feature of the current unilateral waveform used to characterize the difference between neural activity and the normal baseline time.

[0057] The normal baseline waveform refers to the average waveform of a normal subject (a subject without any mental abnormalities, or a healthy subject) that is identical to the waveform currently being processed. For example, if the waveform currently being processed is the left ABR waveform under a specific stimulus, then the normal baseline waveform here refers to the average of the left ABR waveforms of multiple normal subjects under that specific stimulus, representing the normal baseline of the current left ABR waveform. If the waveform currently being processed changes, the designation of the normal baseline waveform will also change accordingly, remaining consistent with the waveform currently being processed.

[0058] The first temporal feature reflects the temporal difference between the current waveform and the normal control baseline. This feature matches the waveform to be processed with the temporal pattern of the normal baseline to analyze the temporal pattern of the waveform. Such temporal analysis can provide precise temporal characteristics of neural activity, reveal subtle neurological differences, and provide key biomarkers for subsequent prediction of mental disorder types.

[0059] Specifically, the time window covering the current moment is a very small window, such as a window covering 10 frames centered on the current moment. The size of this window is preset, and then for each moment in each one-sided signal, the closest matching amplitude in the normal baseline waveform is found. For example, if the amplitude at moment 1 in the current one-sided waveform is A1, within the 10-frame time window centered on moment 1 in the normal baseline waveform, the time point whose amplitude is closest to A1 is found and denoted as moment 2. The deviation between moment 2 and moment 1 is the time offset corresponding to moment 1. Performing the same operation for each moment in the current one-sided waveform yields the time offset for each moment. Taking the root of the square of the time offsets for all moments gives the first time feature of the current one-sided waveform. Performing the same operation for each one-sided waveform yields the first time feature for each one-sided waveform.

[0060] Optionally, if the first time feature is denoted as TimeCovariance, the root of the square is calculated as follows:

[0061]

[0062] Wherein, timepoint represents each moment in the current single-sided waveform, closetindex represents the time point that matches the amplitude of timepoint, abs represents the absolute value, and normalizationfactor represents the set scaling index to keep the value of the first time feature within a certain range; the square root operation refers to squaring all data, accumulating the sum of squares, and then taking the square root of the sum.

[0063] 2. For each moment of the left waveform in the current left and right EEG waveforms, perform the following operations: Define a time window covering the current moment in the right waveform of the current left and right EEG waveforms; within the time window, determine a time point whose amplitude is closest to the amplitude of the current moment in the left waveform; and calculate the time offset between the time point and the current moment; calculate the root of the square of the time offset corresponding to each moment to obtain the second time feature used to characterize the time difference between channels.

[0064] The second time feature is used for advanced time analysis to measure time differences between channels. This feature is helpful in detecting conduction delays or time asymmetry between hemispheres.

[0065] Specifically, for any moment in the left waveform, find the closest matching amplitude in the left waveform before and after that moment in the right waveform; calculate the time offset between the time point corresponding to that matching amplitude and that moment. Perform the same operation for each moment in the left waveform to obtain the time offset for each moment; take the root of the sum of squares of the time offsets for all moments to obtain the second temporal feature of the current left and right EEG waveforms.

[0066] Optionally, the second time feature can be denoted as TimeBetwCh, and the calculation of the square root is as follows:

[0067]

[0068] The meanings of each variable are the same as in formula (1), except that the timepoint and the matching time point closetindex at each moment are derived from the left and right waveforms under the same stimulus, respectively.

[0069] The second category is amplitude characteristics. Similarly, for ease of distinction and description, the sub-characteristics belonging to amplitude characteristics will be referred to as the first amplitude characteristic, the second amplitude characteristic, etc., in the order of their appearance, and so on.

[0070] 1. For each unilateral waveform in the current left and right EEG waveforms, perform the following operations respectively: calculate the amplitude difference between the current unilateral waveform and the normal baseline waveform at each time, and take the root of the square of the amplitude difference at each time to obtain the first amplitude feature of the current unilateral waveform used to characterize the difference between neural activity and the normal baseline amplitude.

[0071] The first amplitude feature represents the degree of deviation between the current neural activity and the overall amplitude distance of the normal pattern, and plays an important role in predicting mental abnormalities. Optionally, if the amplitude of the first feature is denoted as DistanceToConfidence, then:

[0072] distance=([(signal-healthysignal)*50]))

[0073]

[0074] Where distance represents the amplitude distance between the two waveforms being compared at the same moment, signal and healthysignal represent the amplitudes of the current single-sided waveform and the normal baseline waveform at the same moment, respectively; 50 is a scaling factor that controls the range of amplitude distance values ​​and can be adjusted according to the situation in practical applications.

[0075] 2. For the current left and right brain waveforms, detect the peaks and troughs of the left and right waveforms respectively, and calculate the amplitude difference between the peaks / troughs of the same order in the left and right waveforms; calculate the root of the square of each amplitude difference to obtain the second amplitude feature used to characterize the amplitude difference between channels.

[0076] The second amplitude feature is used to characterize the interchannel amplitude differences near the peak and / or trough values. Since the peak and trough values ​​represent typical amplitude values ​​in the waveform and to some extent represent the amplitude variation pattern of the entire waveform, cross-channel analysis of peak / trough values ​​can compare the different ways in which different hemispheres of the brain process the same stimulus, thus providing important neurophysiological insights that are crucial for identifying neural features.

[0077] Optionally, the second amplitude feature can be denoted as CovarianceBetwChSup. First, an edge detection algorithm is used to detect the peaks and / or valleys of the left and right waveforms respectively. Then, for each side of the waveform, the detected peaks and / or valleys are arranged sequentially. Peaks or valleys with the same order on both sides are compared to obtain the channel difference for each pair of peaks or valleys. For example, the first peak on the left is compared with the first peak on the right, then the first valley on the left is compared with the first valley on the right, then the second peak on the left is compared with the second peak on the right, and so on. The root of the sum of squares of the differences for each comparison is then used to obtain the second amplitude feature.

[0078] The third category is density features. Similarly, for ease of distinction and description, the sub-features belonging to density features will be referred to as the first density feature, the second density feature, etc., in the order of their appearance, and so on.

[0079] 1. For each unilateral waveform in the current left and right EEG waveforms, perform the following operations respectively: accumulate the absolute values ​​of the amplitude at multiple moments in the current unilateral waveform, and average the sum over the total duration covered by the multiple moments to obtain the first density feature of the current unilateral waveform used to characterize the signal density.

[0080] The density feature incorporates a time-averaged value over amplitude. Low density indicates reduced neural synchrony, which may be a sign of mental abnormality; high density indicates strong neural response and normal development. The first density feature is used to measure the absolute signal energy density across a time window.

[0081] Optionally, the first density feature can be denoted as Density. For a given unilateral waveform, the absolute values ​​of the amplitudes at multiple moments in the waveform can be scaled proportionally and then summed. This sum is then divided by the length of the time window corresponding to those multiple moments (i.e., the size of the time window covered by the summed absolute amplitude values). This results in the first density feature of the given unilateral waveform. The time window can be the entire waveform, a segment of interest within the waveform, or the average of multiple segments taken sequentially.

[0082] 2. For the current left and right EEG waveforms, calculate the difference in the absolute amplitude of the left and right waveforms at each time point, accumulate the difference in the absolute amplitude of multiple time points, and average the sum over the total duration covered by the multiple time points to obtain a second density feature used to characterize the difference in absolute signal density between channels.

[0083] The second density feature is used to measure the difference in amplitude density between the left and right channels, which is of great significance for detecting differences in processing between the left and right channels and identifying anomalies in unilateral pathways.

[0084] Optionally, the second density feature can be denoted as DensityBetwCh. The difference in absolute amplitude at that moment is obtained by subtracting the absolute amplitude values ​​of the left and right waveforms at the same instant. Then, the differences in absolute amplitude values ​​at multiple instants are scaled by a certain ratio and summed. This sum is then divided by the length of the time window corresponding to the multiple instants (i.e., the size of the time window covered by the summed absolute amplitude differences). This yields the second density feature of the current left and right waveforms. Similarly, the time window can be the entire waveform, a segment of interest within the waveform, or the average of multiple segments taken sequentially.

[0085] 3. For the current left and right EEG waveforms, calculate the difference in signed amplitude of the left and right waveforms at each time point, accumulate the differences in signed amplitude at multiple time points, and average the sum over the total duration covered by the multiple time points to obtain the third density feature used to characterize the difference in signed signal density between channels.

[0086] The third density feature is based on the second density feature and performs channel expansion analysis to measure the directional amplitude difference of energy density between channels in order to capture the directional differences of neural activity in different types of mental disorders.

[0087] Optionally, the third density feature can be denoted as SpreadBetwCh. Its calculation method differs from that of the second density feature: when calculating the amplitude difference, the second density feature uses the absolute values ​​of the two amplitudes to subtract, that is, neither amplitude has a positive or negative sign; but when calculating the amplitude difference, the third density feature uses the two amplitudes with positive or negative signs to subtract, thus reflecting the difference in direction.

[0088] The fourth category is fatigue characteristics:

[0089] For each unilateral waveform in the current left and right EEG waveforms, the following operations are performed respectively: the current unilateral waveform is divided into the first half and the second half, the sequence correlation coefficient of the first half and the second half is calculated, and fatigue characteristics are obtained to characterize the fatigue pattern of neural activity over time.

[0090] Fatigue features, used to reflect signal exhaustion patterns, are a method for fatigue analysis. Fatigue analysis can capture how the attentional and sensory systems change over time, providing a key dimension for understanding the neurophysiological differences in conditions such as ADHD and ASD. In this embodiment, fatigue features primarily calculate and compare the differences in neural responses to test stimuli between the first and second halves of the time interval.

[0091] Optionally, the fatigue characteristic can be denoted as Exhaustion, and the specific calculation method is Correlation(early_frames, late_frames), where early_frames and late_frames represent the first half and the second half of the time series, respectively, and Correlation represents the correlation coefficient between the time series.

[0092] To more clearly illustrate the above features, the type, name, and meaning of each feature are shown in the table below:

[0093]

[0094]

[0095] It should be noted that the above-mentioned features, and the specific sub-features under each feature category, are several key representative features that were finally determined after continuous analysis, optimization, and verification in this embodiment. In practical applications, in addition to the features mentioned above, other features can be introduced based on the characteristics of the neural activity of subjects in different mental states; this embodiment does not impose specific limitations.

[0096] After performing the above processing on the left and right ABR waveforms and the left and right ERP waveforms respectively, the time subdivision characteristics, amplitude subdivision characteristics, density subdivision characteristics and fatigue subdivision characteristics of the left and right ABR waveforms are obtained, as well as the time subdivision characteristics, amplitude subdivision characteristics, density subdivision characteristics and fatigue subdivision characteristics of the left and right ERP waveforms.

[0097] S130. The time characteristics, amplitude characteristics, density characteristics and fatigue characteristics of the left and right ABR waveforms and left and right ERP waveforms are fused together to form EEG characteristics for reflecting neural activity.

[0098] This step fuses the various subdivided features constructed above to obtain new EEG features that comprehensively reflect the function of the nervous system. Optionally, the EEG features in this embodiment will be used to predict the probability that a subject belongs to (or is prone to) certain mental abnormalities. Therefore, based on typical mental abnormality types, subdivided features that can effectively reflect the characteristics of each abnormality type can be extracted from the time characteristics, amplitude characteristics, density characteristics, and fatigue characteristics of the left and right ABR waveforms and left and right ERP waveforms, and combined with these features as the object of feature fusion. Among them, mental abnormalities refer to mental disorders involving emotion, cognition, behavior, and social functioning. Typical types of mental abnormalities include depression, schizophrenia, and bipolar disorder.

[0099] For example, for ADHD (Attention Deficit Hyperactivity Disorder) and ASD (Autism Spectrum Disorder), the extracted subdivision feature combinations include the first amplitude feature, second amplitude feature, first time feature, and fatigue feature of the left and right ABR waveforms and left and right ERP waveforms. The subdivision feature combinations corresponding to different mental disorder types may not be exactly the same, and the subdivision feature combinations for some disorder types may include all of the above subdivision features.

[0100] Of course, if in S120, in addition to the above-mentioned time features, amplitude features, density features and fatigue features, other features mentioned above are also introduced as the extraction range of subdivision features, then the subdivision feature combination in this step can also include the other features or the subdivision features of the other features.

[0101] Furthermore, depending on the fusion perspective, this embodiment provides the following two optional implementation methods for fusing subdivided features:

[0102] In the first alternative implementation, the individual sub-features are combined as a whole and used as EEG features to reflect each type of abnormality. For example, if a certain type of abnormality corresponds to a combination of sub-features, then the set of all sub-features in that combination is considered a fused feature and used as the EEG feature to reflect that type of abnormality. In this fusion method, the individual sub-features are independent of each other and there is no data interaction; the EEG feature for each type of abnormality will be a set of multiple sub-features.

[0103] The second optional implementation involves sequentially arranging all sub-features in each sub-feature combination into sub-feature vectors, and then calculating the eigenvalues ​​of the covariance matrix of each sub-feature vector. The eigenvalues ​​corresponding to each sub-feature combination are then used as EEG features reflecting each abnormality type. This method fuses the sub-features in the sub-feature combination into new features by solving for the eigenvalues, focusing more on the correlation and synergistic effect between sub-features. Under this fusion method, the EEG feature for each abnormality type will be a completely new feature or feature set.

[0104] Specifically, for any anomaly type, assuming that the combination of subdivided features related to that anomaly type includes 8 subdivided features x1, x2, ..., x8, then the features are first arranged sequentially to obtain the feature vector X = [x1, x2, ..., x8]. T Then calculate the vector mean. (Here, n = 8), thus obtaining the central value vector X. center = [x1-μ,x2-μ,…,x8-μ] T Then, calculate the covariance matrix: This is an n×n matrix. Finally, the eigenvalues ​​λ of the covariance determinant are calculated using the following characteristic equation: det(C-λI)=0, where I represents the identity matrix.

[0105] These eigenvalues, also known as the eigenvalues ​​of the covariance matrix, reflect the main data characteristics of the covariance matrix. One or more of the calculated eigenvalues ​​are randomly selected as the final EEG feature. It should be noted that regardless of which eigenvalue(s) are selected as the EEG feature, the method of selecting eigenvalues ​​for all tests must be consistent. For example, the largest eigenvalue can be selected as the EEG feature for each abnormality type, or the first two eigenvalues ​​can be selected as the EEG feature for each abnormality type.

[0106] Based on the above EEG characteristics Figure 2 This is a schematic diagram of a system for constructing a mental health prediction model combining ABR and ERP features, provided by an embodiment of the present invention. This system is used to train a machine learning classification model, converting extracted EEG features into predicted probabilities of mental abnormalities. Figure 2 The system includes a waveform acquisition module, a feature extraction module, and a model building module.

[0107] The waveform acquisition module is used to acquire left and right EEG waveforms with labeled data. The waveform types include ABR waveforms and ERP waveforms, and the labeled data includes normal and at least one type of mental abnormality.

[0108] Optionally, the waveform acquisition module may include a stimulus generation unit and a signal acquisition unit. The stimulus generation unit generates specific types of stimuli (such as click sounds, visual images, cognitive tasks, etc.) and sets parameters such as intensity, duration, frequency, and randomness to induce ABR and ERP waveforms. Subjects generate neural activity under these stimuli. The signal acquisition unit, which is in direct contact with the subject, acquires ABR and ERP signals and includes electrode sensors, wires, amplifiers, and filtering circuits. This embodiment utilizes this module to acquire waveforms from a large number of subjects to construct a training set for a machine learning classification model.

[0109] In one specific implementation, a large number of subjects can be divided into a normal control group and different abnormality type groups. Each group of subjects and their waveforms have labels consistent with the group name (normal label and labels for each abnormality type). Furthermore, each group can be further stratified, for example, based on age (1-16 years, 17-75 years), gender, and stimulus type (T1, T4, T9). Similarly, the label types can also be further subdivided.

[0110] The feature extraction module is used to extract EEG features reflecting neural activity from the left and right EEG waveforms from the same test using the EEG feature extraction method provided in any of the above embodiments.

[0111] Similarly, the left and right EEG waveforms from the same test refer to the left and right ABR and ERP waveforms obtained after performing a complete set of tests on the same subject. The specific waveform forms are the same as those described in the feature extraction method. The methods S110 to S120 above are used to process each type of left and right EEG waveform from each subject to obtain several types of features for each type of left and right EEG waveform. Then, the method S130 is used to fuse the subdivided feature combinations corresponding to each abnormality type for each subject in the normal control group to obtain the EEG features corresponding to each abnormality type for each subject; simultaneously, for each subject in each abnormality type group, the subdivided feature combinations corresponding to the abnormality type label of that group are fused to obtain the EEG features corresponding to each subject and the abnormality type label of that group. In other words, subjects in the normal group have EEG features for each abnormality type, while subjects in the abnormality type group have EEG features for their own abnormality type labels.

[0112] The model building module is used to train a machine learning classification model using EEG features with normal labels and labels for each abnormal type. The trained model is used to predict the degree to which the new left and right EEG waveforms tend to be labeled for each abnormal type based on the EEG features of the new left and right EEG waveforms, thus providing auxiliary information for medical decision-making.

[0113] The labels for EEG features are identical to those of the subjects from whom they originated, including labels for normal features and various abnormality types. Machine learning classification models can employ Bayesian models, random forest models, neural network models, etc. The following section will use a Bayesian model as an example to illustrate the model construction process.

[0114] In one specific implementation, a binary classification Bayesian model is constructed for each abnormality type. The input of the model is the subject's EEG characteristics, and the output is the probability of the subject being normal and the probability of belonging to the corresponding abnormality type. Specifically, the Bayesian model assumes that the features are independent and that the distributions of the normal group and the abnormality type group are independent. Based on this assumption, taking the EEG characteristics fused by the first optional implementation method as an example, the following operations can be performed for each abnormality type label:

[0115] Step 1: Determine the normal probability P based on the number of tests with both a normal label and a label for the current anomaly type (assuming it's the kth anomaly type). 正常 and the probability P of the current anomaly type 异常kThis step calculates the prior probabilities in the Bayesian model. Optionally, based on the number of people in the normal group and the number of people in the abnormal group of the current abnormal type, the probability of the normal population and the probability of the current abnormal population can be determined, which are P here. 正常 and P 异常k .

[0116] Step 2: Determine the M subdivision features included in the subdivision type combination corresponding to the current anomaly type. For the m-th subdivision feature (m = 1, 2, ..., M), determine the Gaussian distribution probability G of this subdivision feature in the normal label group. 正常,m And the Gaussian distribution probability G of this subdivision feature in the current anomaly type label group. 异常k,m .

[0117] This step calculates the conditional probability in the Bayesian model. Optionally, taking the normal group as an example, for any sub-feature, based on the mean and root mean square error of that sub-feature in the normal group, a Gaussian probability density function of that sub-feature in the normal group can be fitted. This is G. 正常,m By performing the above operations on all sub-features, we can obtain the G value for each sub-feature. 正常,m The other groups are similar.

[0118] Step 3: Calculate the probabilities (i.e., P) 正常 P 异常k and each G 正常,m These factors together constitute the Bayesian classification model. Based on the prior probabilities and conditional probabilities mentioned above, the posterior probabilities of the two categories (normal or current abnormal type) can be calculated using Bayes' theorem, which is the model output.

[0119] Similar to the EEG features obtained through the second optional implementation method described above, each of the above subdivided features can be replaced with the feature values ​​selected as EEG features. In particular, when only one feature value is selected as an EEG feature, M=1, and M represents the number of feature values ​​selected as EEG features under the current abnormality type.

[0120] Based on the above EEG characteristics and mental prediction models, a system is constructed. Figure 3 This is a schematic diagram of a mental health prediction system combining ABR and ERP features provided in an embodiment of the present invention. This system utilizes the machine learning classification model trained in the above embodiment to predict the probability that a new subject belongs to (or is prone to) various mental health abnormality labels based on the subject's EEG waveform. Figure 3The system includes a waveform acquisition module, a feature extraction module, and a prediction module. The waveform acquisition module and the feature extraction module can reuse the waveform acquisition module and the feature extraction module in the above-mentioned mental prediction model construction system. The specific data processing methods in the modules are also the same. The difference is that the waveform data processed by the two modules in this system are unlabeled new data.

[0121] Specifically, the waveform acquisition module is used to acquire new left and right EEG waveforms from the same test, including new ABR waveforms and new ERP waveforms. These waveforms come from unlabeled new subjects, and this embodiment will use these waveforms to predict scientific mental state labels for these subjects.

[0122] The feature extraction module is used to extract the temporal, amplitude, and density features, as well as fatigue features, of the new unilateral and bilateral ABR and ERP waveforms, respectively. These new temporal, amplitude, density, and fatigue features are then fused to form new EEG features reflecting neural activity. The specific feature extraction and feature fusion processes are the same as described in the above method embodiments.

[0123] The prediction module is used to use a trained machine learning classification model to predict the degree to which the new left and right EEG waveforms tend to each abnormality type label based on the new EEG features, providing auxiliary information for medical decision-making.

[0124] Taking the aforementioned Bayesian model as an example, after inputting the new left and right EEG waveform feature data into the Bayesian model corresponding to a certain abnormality type, the model will be based on P 正常 P 异常k and each G 正常,m Calculate the probability that the subject is likely to be labeled as normal or currently abnormal based on each sub-feature and / or feature value in the new EEG features. Taking each sub-feature (i.e., the first fusion method mentioned above) as an example, the calculation method for any sub-feature is as follows:

[0125] P(Diagnosis|feature)=P(feature|Diagnosis)*P(Diagnosis) / P(feature)

[0126] Where P(Diagnosis|feature) represents the probability that a subject tends to the current abnormality type based on the value of this sub-feature, and P(feature|Diagnosis) represents the probability that the value of this sub-feature is in the current abnormality type label group (consisting of G). 异常k,m P(Diagnosis) represents the probability of the current anomaly type (i.e., P0).异常k P(feature) represents the probability of the occurrence of the value of the subdivision feature:

[0127] P(feature)=P(feature|HC)*P(HC)+P(feature|Diagnosis)*P(Diagnosis)

[0128] Where P(HC) represents the normal probability under the current anomaly type (i.e., Pnormal). 正常 P(feature|HC) represents the probability that the value of this subdivision feature is in the normal group (consisting of G). 正常,m ).

[0129] By performing the above calculations for each sub-feature, we can obtain the probability that the subject is likely to be normal and the probability that the subject is likely to be the current abnormal type, based on each sub-feature.

[0130] Finally, the probabilities predicted for each subdivision feature that the subject tends to the current abnormal type are multiplied and normalized and scaled to obtain the final probability that the new left and right EEG waveforms tend to the normal label and the final probability that they tend to the current abnormal type label.

[0131] Laplace smoothing can be used to avoid zero probability and improve robustness. Normalization scaling aims to numerically scale the product of probabilities so that the final probability of the normal label and the final probability of the current anomalous type label are combined to 1, and both final probabilities are output. Furthermore, the greater the probability that a subject tends to the current anomalous type predicted based on a single sub-feature, the greater the co-occurrence of that single sub-feature.

[0132] If the EEG features are not derived from subdivided features, but rather from the eigenvalues ​​of the covariance matrix of the subdivided feature vectors, then simply replacing each subdivided feature in the above method with its respective eigenvalue is sufficient. Each eigenvalue represents the combined information of multiple subdivided features, reflecting the interrelationships between them.

[0133] By using the Bayesian model corresponding to each anomaly type to make predictions, the probability that the new subject tends to each anomaly type can be obtained.

[0134] It is worth noting that, since Bayesian classifiers assume that the sub-features are independent of each other, the first optional implementation of fusing EEG features in the above embodiments has better accuracy when the sub-features are independent, while the second optional implementation has better accuracy when the sub-features are correlated. In practical applications, the appropriate method can be selected based on the correlation within the combination of sub-features under different abnormality types.

[0135] In summary, this invention provides a method for extracting EEG features by combining ABR and ERP waveforms. By simultaneously acquiring and analyzing auditory evoked brainstem response (ABR) and event-related potentials (ERP), it effectively reduces the cumbersome steps of multiple devices and tests, improves detection efficiency and subject compliance, and ensures high-precision alignment of data at the same time reference. This integrated fusion of ABR and ERP breaks the traditional model of independent detection of the two, facilitating a more comprehensive understanding of the connection mechanism between brainstem and cortical cognitive functions. Furthermore, this embodiment not only focuses on the brainstem auditory conduction function characteristics reflected by ABR but also integrates key features of ERP in cognitive or emotional processing, achieving a holistic assessment from peripheral auditory pathways to higher cortical functions. Specifically, this embodiment captures data characteristics from dimensions such as time, amplitude, density, and fatigue, respectively mining effective features closely related to neural activity in ABR and ERP. It not only focuses on the contrast characteristics between the left and right channels and the healthy baseline but also pays attention to the activity differences between the left and right sides, as well as the overall pattern differences of the waveform curves. Finally, feature fusion comprehensively reflects the state of neural activity. Compared to traditional detection methods that rely too heavily on subjective scales, this in-depth feature, which combines information from both the brainstem and cortical levels, is more objective and accurate. It helps to comprehensively reveal the multi-layered characteristics of the mechanisms of mental abnormalities and provides richer auxiliary information for medical decision-making.

[0136] Based on the above methods, this embodiment also provides a mental prediction model construction system that combines ABR and ERP features and a mental prediction system that combines ABR and ERP features. It uses machine learning or statistical methods to perform fusion analysis on multi-dimensional EEG features, improves the accuracy of identifying mental abnormalities, and can also support further subtyping prediction, helping doctors to grasp the condition in the early stages and formulate personalized intervention measures.

[0137] It should be noted that all user data involved in this application is information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0138] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 4 As shown, the device includes a processor 60, a memory 61, an input device 62, and an output device 63; the number of processors 60 in the device can be one or more. Figure 4 Taking a processor 60 as an example; the processor 60, memory 61, input device 62, and output device 63 in the device can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.

[0139] The memory 61, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the EEG feature extraction method combining ABR and ERP waveforms in this embodiment of the invention. The processor 60 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 61, thereby realizing the aforementioned EEG feature extraction method combining ABR and ERP waveforms.

[0140] The memory 61 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 61 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory 61 may further include memory remotely located relative to the processor 60, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0141] Input device 62 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 63 may include display devices such as a display screen.

[0142] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the EEG feature extraction method combining ABR and ERP waveforms of any embodiment.

[0143] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0144] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0145] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0146] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting EEG features by combining ABR and ERP waveforms, characterized in that, include: Obtain EEG waveforms from the left and right sides of the same test, including ABR waveforms and ERP waveforms; For the left and right ABR waveforms and the left and right ERP waveforms respectively, extract the time features, amplitude features, density features, and fatigue features between the single-sided waveform and the double-sided waveform; The time characteristics, amplitude characteristics, density characteristics, and fatigue characteristics of the left and right ABR waveforms and left and right ERP waveforms are fused together to form EEG characteristics that reflect neural activity. The extraction of density features between single-sided and double-sided waveforms includes: For the current left and right EEG waveforms, the difference in the absolute amplitude of the left and right waveforms at each time point is calculated. The difference in the absolute amplitude of multiple time points is accumulated, and the sum is averaged over the total duration covered by the multiple time points to obtain a second density feature used to characterize the difference in absolute signal density between channels. For the current left and right EEG waveforms, the difference in signed amplitude of the left and right waveforms at each time point is calculated. The differences in signed amplitude at multiple time points are accumulated, and the sum is averaged over the total duration covered by the multiple time points to obtain a third density feature used to characterize the difference in signed signal density between channels.

2. The method according to claim 1, characterized in that, The extraction of time features between the single-sided and double-sided waveforms includes: For each moment of each unilateral waveform in the current left and right EEG waveforms, the following operations are performed: a time window covering the current moment is defined in the normal baseline waveform; within the time window, the time point whose amplitude is closest to the amplitude of the current moment in the current unilateral waveform is determined, and the time offset between the time point and the current moment is calculated; the root of the sum of squares of the time offsets corresponding to each moment is calculated to obtain the first time feature of the current unilateral waveform used to characterize the difference between neural activity and the normal baseline time. For each moment in the left waveform of the current left and right EEG waveforms, the following operations are performed: a time window covering the current moment is defined in the right waveform of the current left and right EEG waveforms; within the time window, the time point whose amplitude is closest to the amplitude of the current moment in the left waveform is determined, and the time offset between the time point and the current moment is calculated; the root of the sum of squares of the time offsets corresponding to each moment is calculated to obtain the second time feature used to characterize the time difference between channels.

3. The method according to claim 1, characterized in that, The extraction of amplitude features between single-sided and double-sided waveforms includes: For each unilateral waveform in the current left and right EEG waveforms, perform the following operations respectively: calculate the amplitude difference between the current unilateral waveform and the normal baseline waveform at each time, and calculate the square root of the amplitude difference at each time to obtain the first amplitude feature of the current unilateral waveform used to characterize the difference between neural activity and the normal baseline amplitude. For the current left and right brain waveforms, the peaks and troughs of the left and right waveforms are detected respectively, and the amplitude difference between the peaks / troughs of the same order in the left and right waveforms is calculated; the root of the square of each amplitude difference is obtained to obtain the second amplitude feature used to characterize the amplitude difference between channels.

4. The method according to claim 1, characterized in that, The extraction of density features between single-sided and double-sided waveforms also includes: For each unilateral waveform in the current left and right EEG waveforms, the following operations are performed respectively: the absolute values ​​of the amplitude at multiple moments in the current unilateral waveform are accumulated, and the sum is averaged over the total duration covered by the multiple moments to obtain the first density feature of the current unilateral waveform used to characterize the signal density.

5. The method according to claim 1, characterized in that, The extraction of fatigue features between single-sided and double-sided waveforms includes: For each unilateral waveform in the current left and right EEG waveforms, the following operations are performed respectively: the current unilateral waveform is divided into the first half and the second half, the sequence correlation coefficient of the first half and the second half is calculated, and fatigue characteristics are obtained to characterize the fatigue pattern of neural activity over time.

6. The method according to claim 1, characterized in that, The process of fusing the temporal, amplitude, density, and fatigue characteristics of the left and right ABR and ERP waveforms to form EEG features reflecting neural activity includes: Based on typical types of mental abnormalities, extract subdivided feature combinations related to each abnormality type from the time characteristics, amplitude characteristics, density characteristics, and fatigue characteristics of the left and right ABR waveforms and left and right ERP waveforms. The subdivided features are combined to serve as EEG features reflecting each type of abnormality; or all subdivided features in each subdivided feature combination are arranged sequentially as subdivided feature vectors, and the eigenvalues ​​of the covariance matrix of each subdivided feature vector are solved. The eigenvalues ​​corresponding to each subdivided feature combination are then used as EEG features reflecting each type of abnormality.

7. The method according to claim 1, characterized in that, Before extracting the time, amplitude, and density features, as well as fatigue features, between the single-sided and double-sided waveforms for the left and right ABR and ERP waveforms respectively, the following steps are also included: For each single-sided waveform, artifact removal, DC offset elimination, high-pass filtering, low-pass filtering, quality segmentation, and quality evaluation are performed respectively. The single-sided waveform that meets the quality requirements in the evaluation results is taken as the final single-sided waveform.

8. A mental health prediction model construction system combining ABR and ERP features, characterized in that, include: The waveform acquisition module is used to acquire left and right EEG waveforms with labeled data. The waveform types include ABR waveforms and ERP waveforms, and the labeled data includes normal and at least one type of mental abnormality. The feature extraction module is used to extract brainwave features reflecting neural activity from the left and right EEG waveforms derived from the same test, respectively, using the method described in any one of claims 1-7. The model building module is used to train a machine learning classification model using EEG features with normal labels and labels for each abnormal type. The trained model is used to predict the degree to which the new left and right EEG waveforms tend to be labeled for each abnormal type based on the EEG features of the new left and right EEG waveforms.

9. The system according to claim 8, characterized in that, The model building module trains the machine learning classification model in the following manner: For each type of exception label, perform the following operations: Based on the number of tests with normal labels and current anomaly type labels, determine the normal probability and the probability of the current anomaly type, respectively; Based on each sub-feature in the EEG features with normal labels, determine the Gaussian distribution probability of each sub-feature in the normal label group; Based on each sub-feature in the EEG features with the current abnormality type label, determine the Gaussian distribution probability of each sub-feature in the current abnormality type label group; The Bayesian classification model is composed of various probabilities. The Bayesian classification model is used to: respond to the EEG features of new left and right EEG waveforms, calculate the probability that each sub-feature in the EEG features tends to the normal label and the current abnormal type label based on the probabilities in the model; and use the probability that each sub-feature tends to each label to calculate the probability that the new left and right EEG waveforms tend to the current abnormal type label.

10. A mental health prediction system combining ABR and ERP features, characterized in that, include: The waveform acquisition module is used to acquire left and right EEG waveforms from the same test, including ABR waveforms and ERP waveforms. The feature extraction module is used to extract time features, amplitude features, density features, and fatigue features between unilateral and bilateral waveforms for the left and right ABR waveforms and the left and right ERP waveforms respectively; and to fuse the time features, amplitude features, density features, and fatigue features of the left and right ABR waveforms and the left and right ERP waveforms as EEG features to reflect neural activity. The prediction module is used to use a trained machine learning classification model to predict the degree to which the left and right EEG waveforms tend to each abnormality type label based on the EEG features, so as to provide auxiliary information for medical decision-making. The extraction of density features between single-sided and double-sided waveforms includes: For the current left and right EEG waveforms, the difference in the absolute amplitude of the left and right waveforms at each time point is calculated. The difference in the absolute amplitude of multiple time points is accumulated, and the sum is averaged over the total duration covered by the multiple time points to obtain a second density feature used to characterize the difference in absolute signal density between channels. For the current left and right EEG waveforms, the difference in signed amplitude of the left and right waveforms at each time point is calculated. The differences in signed amplitude at multiple time points are accumulated, and the sum is averaged over the total duration covered by the multiple time points to obtain a third density feature used to characterize the difference in signed signal density between channels.

Images