Objective recognition system of depression based on rhythmic analysis spectrum, medium and equipment

Abstract

This invention belongs to the field of medical artificial intelligence technology and provides an objective identification system, medium, and device for depression based on rhythmic analytical spectrum. The system includes an EEG acquisition device for collecting dual-channel EEG signals from a subject in a resting state and transmitting them to a data processing platform. The data processing platform, connected to the EEG acquisition device, acquires the raw dual-channel EEG signals and, based on the raw dual-channel EEG signals and a biomarker extraction model, obtains core biomarker features. Combining the core biomarker features with a pre-trained depression identification model, it outputs the depression identification result. A host computer, connected to the data processing platform, acquires and displays the depression identification result. This invention addresses the problems of unstable biomarkers, low accuracy, and poor individual adaptability in traditional EEG analysis methods for depression diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence technology, and in particular relates to an objective recognition system, medium and device for depression based on rhythmic analytical spectrum. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Electroencephalography (EEG) signals can directly reflect the brain's neural electrical activity and are an important means of finding objective biomarkers for depression. Numerous studies have shown that patients with depression have characteristic EEG abnormalities, among which asymmetry in prefrontal alpha waves (alpha power in the left prefrontal cortex is higher than that in the right) is the most widely recognized electrophysiological marker of depression.

[0004] Current methods for identifying depression rely on EEG analysis using power spectrum, but power spectrum analysis cannot distinguish between periodic oscillations and non-periodic 1 / f background noise. Age, gender, head movements, and muscle activity can significantly interfere with the accuracy of power values, leading to poor biomarker stability. Furthermore, inconsistent frequency band definitions—different studies using vastly different alpha band divisions (7-13Hz, 8-12Hz, 8-13Hz, etc.)—make it difficult to replicate research results and standardize clinical applications. Insufficient functional specificity—the traditional alpha band includes two completely different dynamic modes: sustained oscillations and transient bursts. General power analysis cannot distinguish specific components related to the pathology of depression. Fixed frequency band divisions ignore the natural differences in EEG frequencies between individuals, resulting in the missed detection of abnormal signals in some patients. Summary of the Invention

[0005] To address at least one of the technical problems existing in the background art, the present invention provides an objective identification system, medium, and device for depression based on rhythmic analytical spectrum.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides an objective identification system for depression based on rhythmic analytical spectrum, and an EEG acquisition device for acquiring dual-channel EEG signals of subjects in a resting state and transmitting them to a data processing platform; The data processing platform is connected to the EEG acquisition device to acquire raw dual-channel EEG signals and obtain core biomarker features based on the raw dual-channel EEG signals and the biomarker extraction model. The core biomarker features are combined with the pre-trained depression recognition model to output the depression recognition result. The biometric extraction model includes a dual-channel rhythmic spectrum calculation module, an individual frequency band automatic division module, a depression-specific feature extraction module, and a depression recognition module. The dual-channel rhythmicity spectrum calculation module is used to calculate the rhythmicity values ​​of the two channels at each frequency point within a specified range based on the preprocessed EEG signals. The individual frequency band automatic segmentation module is used to generate noise bands based on phase randomization substitution data. It automatically segments each channel into individual-specific high-rhythm bands and low-rhythm bands according to the rhythmicity value of each frequency point, and locates multiple target frequency bands. The depression-specific feature extraction module is used to extract core biomarker features based on the target frequency band positioning results; The host computer is connected to the data processing platform and is used to acquire and display the depression identification results.

[0007] Furthermore, the dual-channel rhythmic spectrum calculation module includes a time-frequency decomposition module and a rhythmic value calculation module; The time-frequency decomposition module is used to perform time-frequency decomposition on the preprocessed dual-channel EEG signal to obtain the complex wavelet coefficient matrix corresponding to each time sampling point and each analysis frequency. The rhythmicity value calculation module is used to combine the complex wavelet coefficient matrix of each time sampling point and each analysis frequency to calculate the mean value of the phase difference vector between the original signal and its 1.5-period replica, obtain the rhythmicity value of each frequency point, and generate the full-band rhythm spectrum.

[0008] Furthermore, in the dual-channel rhythmicity spectrum calculation module, the formula for calculating the rhythmicity value at each frequency point within a specified range is as follows: , in, For frequency f The corresponding rhythmicity value ranges from [0,1]. A higher value indicates a more stable signal phase and a stronger oscillation duration. f The target analysis frequency is defined within the target analysis frequency band; t represents the time sampling point of the EEG signal, and T represents the total number of sampling points for the entire signal. This represents the number of sampling points contained in 1.5 oscillation cycles corresponding to frequency f; These are the complex wavelet coefficients at time point t and frequency f after complex Morlet wavelet time-frequency decomposition; These are the complex wavelet coefficients at time point t+L(f) and frequency f after complex Morlet wavelet time-frequency decomposition. The complex conjugate of the complex wavelet coefficients at the corresponding time-frequency point of the EEG signal with a lag of 1.5 cycles.

[0009] Furthermore, the individual frequency band automatic segmentation module includes a noise band generation module, an individual-specific preliminary segmentation module, and a target frequency band positioning module; The noise band generation module is used to generate phase randomization substitution data, calculate the rhythmic spectrum of each substitution data, and determine the noise band within a set confidence interval; The individual-specific preliminary segmentation module is used to classify regions that are above the baseline and exceed the upper limit of the noise band as high-rhythm bands, and regions that are below the baseline and exceed the lower limit of the noise band as low-rhythm bands, using the median of the rhythmicity values ​​of all frequency points as the baseline. The target frequency band positioning module is used to automatically label each frequency band with the alpha high rhythmic band within a set range as the anchor point, and to locate the alpha high rhythmic band, the theta low rhythmic band, and the beta1 low rhythmic band of the two channels respectively.

[0010] Furthermore, in the dual-channel rhythmic spectrum calculation module, the rhythmic value of each frequency point in the range of 3~45Hz for the two channel signals is calculated.

[0011] Furthermore, the depression-specific feature extraction module includes an alpha rhythmic asymmetric index calculation module, a theta band oscillation burst detection module, and a beta1 average rhythmic feature calculation module. The alpha rhythmic asymmetry index calculation module is used to calculate the average value of rhythmicity values ​​at all frequency points within the alpha band of channels F3 and F4, and to calculate the asymmetry index based on the average value to obtain the alpha rhythmic asymmetry index. The oscillation burst detection module in the theta band is used to calculate the average duration and burst rate of all bursts in the theta band of the F3 channel; The beta1 average rhythmicity characteristic calculation module is used to calculate the average value of the rhythmicity values ​​of all frequency points within the beta1 band of channels F3 and F4. The average value of both values ​​is taken to obtain the beta1 average rhythmicity characteristic.

[0012] Furthermore, the oscillation burst detection module within the theta band is used to calculate the average duration and burst rate of all bursts in the theta band of the F3 channel, including: Extract all time-series power data within the theta band of channel F3, calculate the 90th percentile P90 of the power, and use it as the burst power threshold; The timing signal is traversed frame by frame, and signal segments with instantaneous power > P90 and duration ≥ 2 cycles are selected. All valid burst events are counted, and the total number of bursts M and the duration of each burst are recorded. ; Combined with the total signal acquisition time theta bandwidth Calculate the standardized burst rate .

[0013] Furthermore, the depression identification model adopts a gradient boosting tree ensemble learning model, which classifies the condition into three levels: mild, moderate and severe based on the probability values ​​output by the model and the HAMD score.

[0014] A second aspect of the present invention provides a computer-readable storage medium.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps: Dual-channel EEG signals were collected from subjects at rest and transmitted to a data processing platform. The system acquires raw dual-channel EEG signals and obtains core biomarker features based on the raw dual-channel EEG signals and a biomarker extraction model. It then combines the core biomarker features with a pre-trained depression recognition model to output depression recognition results. Based on the preprocessed EEG signals, the rhythmic values ​​of the two channel signals at each frequency point within the specified range were calculated respectively; Based on noise bands generated from phase randomization substitution data, each channel is automatically divided into individual-specific high-rhythmic bands and low-rhythmic bands according to the rhythmicity value of each frequency point, and multiple target frequency bands are located. Core biomarker features are extracted based on the target frequency band localization results.

[0016] A fourth aspect of the present invention provides a computer device.

[0017] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the following steps: The system acquires raw dual-channel EEG signals and obtains core biomarker features based on the raw dual-channel EEG signals and a biomarker extraction model. It then combines the core biomarker features with a pre-trained depression recognition model to output depression recognition results. Based on the preprocessed EEG signals, the rhythmic values ​​of the two channel signals at each frequency point within the specified range were calculated respectively; Based on noise bands generated from phase randomization substitution data, each channel is automatically divided into individual-specific high-rhythmic bands and low-rhythmic bands according to the rhythmicity value of each frequency point, and multiple target frequency bands are located. Core biomarker features are extracted based on the target frequency band localization results.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention distinguishes between periodic oscillations and non-periodic 1 / f background noise, provides a clear frequency band definition, and takes into account the natural differences in EEG frequencies between individuals. This avoids the problem of confused frequency band definitions, inability to distinguish specific components related to the pathology of depression, and the resulting missed detection of abnormal signals in some patients. This invention improves the stability, accuracy, and individual suitability of biomarkers in the diagnosis of depression.

[0019] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a block diagram of an objective depression identification system based on a rhythmic analytical spectrum architecture provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the LAVI rhythmicity spectrum calculation results provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the ABBA automatic boundary detection algorithm provided in an embodiment of the present invention; Figure 4 These are the core biomarkers and diagnostic results for depression provided in the embodiments of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] Example 1 like Figure 1 As shown, this embodiment provides an objective identification system for depression based on rhythmic analytical spectrum, including: EEG acquisition equipment is used to acquire dual-channel EEG signals from subjects in a resting state and transmit them to a data processing platform; In this embodiment, the subject's resting state includes closed eyes, awake, and relaxed state. The dual-channel EEG signal includes the prefrontal cortex F3 (left frontal cortex) and F4 (right frontal cortex) dual-channel EEG signal. The acquisition parameters are sampling rate ≥1000Hz, bandpass filter 0.5-100Hz, and the reference electrode uses the average reference of both mastoid processes. In this embodiment, the preferred EEG acquisition device is a professional EEG acquisition system conforming to the international 10-20 electrode positioning standard, including the original Nature Communications g. GAMMAcap (gTec), Biosemi ActiveTwo, Neuroscan NuAmps, Brain Products actiCHamp, and a 128-channel EGI high-density EEG analyzer. For primary care screening scenarios, a lightweight portable EEG device with 16 / 32 channels can be used. The device electrodes primarily cover the F3 and F4 frontal dual channels, and bilateral mastoid processes are used as reference electrodes. The acquisition parameters must meet the following requirements: sampling rate ≥1000Hz, 0.5–100Hz bandpass filtering, and electrode impedance less than 20kΩ. The dual-channel EEG signals of the subject in a closed-eye resting state acquired by the device are transmitted in real time to a device equipped with LAVI and ABBA. A dedicated data processing platform for the core rhythm analysis algorithm completes the entire process of signal preprocessing, rhythm spectrum calculation, individualized frequency band division, depression feature extraction, and disease identification. The data processing platform connects to the EEG acquisition equipment and includes a preprocessing module, a dual-channel rhythm spectrum calculation module, an individualized frequency band automatic division module, and a depression-specific feature extraction module. The preprocessing module is used to acquire raw EEG signals, preprocess the acquired raw EEG signals, and obtain preprocessed EEG signals. In this embodiment, the preprocessing includes detrending and mean-removing the original signal to eliminate DC offset and linear trend; using independent component analysis (ICA) to decompose the signal and remove eye movement, blinking and electromyography artifacts; using spectral interpolation to remove 50Hz power frequency interference to avoid the disruption of the rhythm of adjacent frequencies by traditional notch filters; and resampling the signal to 1000Hz to unify the data format.

[0026] The dual-channel rhythmicity spectrum calculation module is used to calculate the rhythmicity values ​​of the two channels at each frequency point within a specified range based on the preprocessed EEG signals. In this embodiment, the Lagged Angle Vector Index (LAVI) algorithm is used to calculate the EEG rhythmicity value. The core of this algorithm is to quantify the stability of the phase-temporal structure by using the mean of the phase difference vector between the signal and its 1.5-period lag replica, thus distinguishing between two dynamic modes of brain activity: continuous oscillation and transient bursts. Unlike traditional power spectrum analysis, which only focuses on signal energy, LAVI is unaffected by 1 / f aperiodic noise and exhibits greater robustness.

[0027] The dual-channel rhythmic spectrum calculation module specifically includes a time-frequency decomposition module and a rhythmic value calculation module; Specifically, the time-frequency decomposition module is used to perform time-frequency decomposition on the preprocessed F3 and F4 channel signals respectively; In this embodiment, a complex Morlet wavelet with a width of 5 periods is used for time-frequency decomposition. The operation is achieved by performing a Fourier transform on the time-domain EEG signal, multiplying it with the frequency domain form of the wavelet, and then performing a frequency domain convolution with an inverse Fourier transform. The operation is performed frequency-by-frequency for the target analysis frequency band to obtain a complex wavelet coefficient matrix corresponding to each time sampling point and each analysis frequency. The complex coefficients simultaneously carry the amplitude and phase information of the EEG signal at the corresponding time and frequency band. In this embodiment, the target analysis frequency band is selected as 3-45Hz as the target analysis frequency band for EEG signals. This band can completely cover all target feature frequency bands highly correlated with the diagnosis of depression, including theta (4-7Hz), alpha (6-14Hz), and beta1 (14.7-17.8Hz), enabling the complete extraction of all rhythmic biomarkers required for the diagnosis of depression. Frequency bands such as delta below 3Hz are significantly affected by baseline drift and insufficient acquisition time, have large noise fluctuations, and lack specific electrophysiological abnormalities for depression. Including them in the analysis would reduce the reliability of the results. Frequency bands above 45Hz are close to 50 / 60Hz power frequency interference, and power frequency signals themselves are highly rhythmic, which would seriously contaminate the LAVI phase consistency calculation results and destroy the frequency band division accuracy. Setting these upper and lower limits can avoid dual noise interference from high and low frequencies. The calculation stability and robustness are optimal in the 3-45Hz range. Below 3Hz, an extremely long acquisition time of several minutes is required, which is not suitable for rapid clinical screening scenarios. Above 45Hz... The signal distortion is obvious, and existing recognized studies have also confirmed that depression-related EEG abnormalities are concentrated in the 4-40Hz range. Therefore, the 3-45Hz range can balance the accuracy of analysis and clinical applicability without losing effective pathological information.

[0028] The rhythmicity value calculation module is used to calculate the mean of the phase difference vector between the original signal and its 1.5-cycle-later replica for each frequency point f, thereby obtaining the rhythmicity value of each frequency point and generating a full-band rhythmic spectrum. Specifically, the formula for calculating the rhythmicity value at each frequency point is as follows: , in, For frequency f The corresponding rhythmicity value ranges from [0,1]. A higher value indicates a more stable signal phase and a stronger oscillation duration. f The target analysis frequency is defined within the target analysis frequency band; t represents the time sampling point of the EEG signal, and T represents the total number of sampling points for the entire signal. This represents the number of sampling points contained in 1.5 oscillation cycles corresponding to frequency f; These are the complex wavelet coefficients at time point t and frequency f after complex Morlet wavelet time-frequency decomposition. To represent the complex conjugate of the complex wavelet coefficients at the corresponding time-frequency point of an EEG signal with a lag of t+L(f) of 1.5 cycles, the superscript... The complex conjugate operator cancels out interference from signal amplitude and energy, retaining only the phase difference between the two time points, thus naturally avoiding the influence of 1 / f aperiodic background noise in the calculation results; The complex wavelet coefficients at time point t+L(f) and frequency f after complex Morlet wavelet time-frequency decomposition are given. The parentheses t+L(f) represent the time sampling point after lag, t is the current reference time sampling point, and f is the target analysis frequency. In this embodiment, it is limited to the 3~45Hz range to fully cover the frequency bands of theta, alpha, and beta1 depression-related features. The denominator is a normalization term used to eliminate the interference of signal amplitude and energy on the result. This formula quantifies the stability of neural oscillations by comparing the phase correlation between the original signal and a signal that lags by 1.5 periods. It naturally avoids the influence of 1 / f non-periodic noise and only reflects the phase and timing characteristics of the signal.

[0029] The individual frequency band automatic segmentation module is used in conjunction with the automatic band boundary detection algorithm (ABBA) to define the noise band based on the median of the rhythmic values ​​of all frequency points and the phase randomization substitution data. Based on the high and low values ​​of the rhythmic values ​​λ(f) of each frequency point, it divides the individual-specific high rhythmic band and low rhythmic band, locates the target frequency band, and extracts the features within each target frequency band. The individual frequency band automatic segmentation module employs the Automatic Band Boundary Detection (ABBA) algorithm. It uses the 3-45Hz full-band rhythmicity value λ(f) calculated by the LAVI module in the previous step. The core objective is to address the problems of traditional fixed frequency band segmentation ignoring individual differences and failing to distinguish between real neural rhythms and random noise. Using 200 sets of surrogate data generated by the IAAFT algorithm, completely replicating the original EEG 1 / f energy spectrum but with phase scrambled, a noise band with a 95% confidence interval is constructed. Simultaneously, the median of all frequency rhythmicity values ​​in a single channel is taken as the global rhythmicity baseline. A dual criterion is used: "rhythmicity values ​​above the baseline and exceeding the noise upper limit are classified as high-rhythmic bands; values ​​below the baseline and falling below the noise lower limit are classified as low-rhythmic bands." This automatically generates personalized functional frequency bands suitable for each individual. For this depression diagnosis scenario, the algorithm is specifically customized, performing independent calculations only on the prefrontal F3 / F4 dual channels, prioritizing individual-specific frequency bands and limiting them to 3-45Hz. The computational range is specifically targeted to the three core frequency bands related to the pathology of depression: theta, alpha, and beta1. In practice, the generation of alternative data, calculation of noise bands and baselines are completed first, and then the high and low rhythm bands are divided frequency by frequency. Finally, the 6-14Hz alpha high rhythm band, which has the highest stability in the population, is used as the anchor point to accurately locate the precise ranges of theta (4-7Hz) low rhythm band, the alpha high rhythm band (6-14Hz), and the beta1 (14.7-17.8Hz) low rhythm band. The individual frequency band automatic segmentation module includes a noise band generation module, an individual-specific preliminary segmentation module, and a target frequency band positioning module. Specifically, the noise band generation module is used to generate phase randomization substitution data, calculate the rhythmic spectrum of each substitution data, and determine the noise band within a set confidence interval; As a further implementation method, this step uses Iterative Amplitude Adjusted Fourier Transform (IAAFT) to generate phase randomized replacement data, strictly replicating the 1 / f non-periodic power spectrum characteristics of the original EEG signal, only scrambling the signal phase. After generating 200 sets of independent replacement signals in batches, the full-band rhythmic spectrum is calculated group by group using the LAVI algorithm and parameters that are completely consistent with the original signal. Then, the rhythmic value distribution of 200 sets of data is statistically analyzed for each frequency point in the range of 3~45Hz. The upper and lower thresholds of rhythmic fluctuations at each frequency are calculated according to the statistical 95% confidence interval. The set of thresholds for all frequencies constitutes the complete noise band, ensuring that the noise judgment criteria are completely consistent with the calculation rules of the original signal.

[0030] This technical solution can accurately distinguish between real neural rhythms and rhythmic pseudo-fluctuations caused by equipment interference, electromyography, head movements, and other noise signals. It can also generate scientific thresholds that conform to the inherent characteristics of EEG, while achieving individualized threshold adaptation to solve the problem of poor generalization of traditional uniform experience thresholds. Furthermore, it can improve the statistical reliability of high and low rhythm band division through large-sample statistical design, and ultimately filter noise interference from the source. This ensures that the subsequently extracted biomarkers of depression, such as the alpha rhythm asymmetry index and theta burst characteristics, can truly reflect pathological changes, thereby reducing the probability of misdiagnosis and missed detection, and improving the accuracy of disease screening, disease grading, and efficacy monitoring.

[0031] The individual-specific preliminary segmentation module is used to classify regions that are above the baseline and exceed the upper limit of the noise band as high-rhythm bands, and regions that are below the baseline and exceed the lower limit of the noise band as low-rhythm bands, using the median of the rhythmicity values ​​of all frequency points as the baseline. The target frequency band positioning module is used to automatically label each frequency band with the alpha high rhythmic band within a set range as the anchor point, and to locate the alpha high rhythmic band, the theta low rhythmic band and the beta1 low rhythmic band of the two channels respectively. Specifically, the rhythmic curve is scanned point by point along the frequency axis, and candidate frequency band segments are divided by the intersection of adjacent baselines and high and low rhythmic attributes are initially marked. Then, the significance is screened by whether the peak / valley value exceeds the noise band threshold to eliminate pseudo-frequency bands. On this basis, the effective high rhythmic band with the highest rhythmic peak in the 6~14Hz range is used as the alpha anchor point. The remaining frequency bands are marked in sequence along the high and low sides of the frequency axis according to the alternating peak and valley pattern. Finally, for the diagnosis of depression, the precise frequency range and characteristic values ​​of the three core target frequency bands, namely theta low rhythmic band, alpha high rhythmic band, and beta1 low rhythmic band, are output. This ensures the universal consistency of the frequency band architecture and adapts to the differences in individual EEG characteristics. By linking and cooperating the dual-channel rhythmic spectrum calculation module and the individual frequency band automatic segmentation module, a specific rhythmic frequency band adapted to the individual subject is finally obtained.

[0032] The depression-specific feature extraction module is used to extract core biomarker features based on the target frequency band positioning results; Specifically, it includes an alpha rhythmic asymmetric index calculation module, an oscillation burst detection module within the theta band, and a beta1 average rhythmic characteristic calculation module; The alpha rhythmic asymmetry index calculation module is used to calculate the average value of rhythmicity values ​​at all frequency points within the alpha band of channels F3 and F4, and to calculate the asymmetry index based on the average value to obtain the alpha rhythmic asymmetry index. The specific input data includes: the frequency range of the individualized alpha high-rhythm band in the F3 (left frontal) and F4 (right frontal) dual-channel bands; and the rhythmicity value λ(f) corresponding to each frequency point within the alpha band.

[0033] Specifically, based on the alpha frequency band range of channel F3 defined by ABBA, all frequency points within this range are traversed, the corresponding λ(f) is read, summed, and then divided by the total number of frequency points to obtain the average rhythm value of channel F3 alpha. ; Calculate the average rhythm value of channel alpha in F4 according to the exact same rules. ; The formula for calculating the average of the rhythmic values ​​at all frequency points within the alpha band of channels F3 and F4 is as follows: , , in, This represents the average rhythm value of the alpha band in channel F3. This indicates the average rhythm value of the alpha band in channel F4; This represents the set of all frequency points within the alpha band of channel F3; This represents the set of all frequency points within the alpha band of channel F4; This indicates the total number of frequency points within the alpha band of channel F3; This indicates the total number of frequency points within the alpha band of channel F4; Substituting the two averages into the asymmetric exponent formula, we obtain the alpha rhythmic asymmetric exponent. , The oscillation burst detection module in the theta band is used to calculate the average duration (number of cycles) and burst rate (times / minute / Hz) of all bursts in the theta band of the F3 channel. In this embodiment, a signal segment with power exceeding the 90th percentile and a duration of ≥2 cycles is defined as a burst. Input data includes the frequency range of the individualized theta low-rhythm band in the F3 channel, raw EEG temporal power data, LAVI rhythm data, and the power sequence of the entire EEG signal; specifically including: Extract all timing power data within the theta band of channel F3, calculate the 90th percentile P90 of the power, and use it as the burst power threshold; The timing signal is traversed frame by frame, and signal segments with instantaneous power > P90 and duration ≥ 2 cycles are selected. All valid burst events are counted, and the total number of bursts M and the duration of each burst are recorded. : , in, Theta represents the average duration of the burst, expressed in oscillation periods. This represents the total number of valid bursts detected within the F3 theta band; The duration of the i-th burst is represented by the number of oscillation cycles contained in a single burst from start to finish. The average burst duration is obtained by summing the durations of all single bursts and dividing by the total number of bursts M.

[0034] Combined with the total signal acquisition time theta bandwidth Calculate the standardized burst rate : , in, This indicates that theta carries a burst rate. This indicates the total duration of the EEG signal acquisition, in minutes. This represents the total bandwidth of the theta low-rhythm band as defined by ABBA, in Hz.

[0035] The beta1 average rhythmicity characteristic calculation module is used to calculate the average of the rhythmicity values ​​at all frequency points within the beta1 band of channels F3 and F4. Taking the average of both values ​​yields the beta1 average rhythmicity characteristic. Based on the beta1 frequency band range of channels F3 and F4 defined by ABBA, calculate the average value of all λ(f) within this frequency band. and : , , in, This indicates the average rhythm value of channel F3, beta1. This indicates the average rhythm value of channel F4, beta1. This represents the set of frequency points within the beta1 band of channel F3; This represents the set of frequency points within the beta1 band of channel F4; This indicates the total number of frequency points within the beta1 band of channel F3; This indicates the total number of frequency points within the beta1 band of channel F4; The arithmetic mean of the average values ​​of the two channels is taken to obtain the comprehensive rhythmic characteristics of the beta1 band. , in, This indicates the final average rhythmicity characteristic of the beta1 low-rhythm band. and This represents the average rhythm value of the left and right frontal beta1 bands in a single channel; The extracted four-dimensional feature vector is input into a pre-trained depression recognition model, which outputs the probability of depression (0-1) and the severity level of the condition. The depression identification model uses an XGBoost ensemble learning model. Based on the probability values ​​and HAMD scores output by the model, the condition is divided into three levels: mild (probability 0.5-0.7, HAMD 8-17), moderate (probability 0.7-0.9, HAMD 18-24), and severe (probability ≥0.9, HAMD ≥25). The host computer is connected to the data processing platform and is used to acquire and display the depression identification results.

[0036] Objective identification of depression based on resting-state EEG This embodiment uses the publicly available MODMA dataset for depression (containing 128 patients with depression and 128 age- and sex-matched healthy controls) to verify the effectiveness of the invention. All subjects underwent 5 minutes of resting-eye-closed EEG recordings with 128 channels. All patients with depression were diagnosed by two psychiatrists and had a HAMD-17 score ≥8.

[0037] Implementation steps: 1. Data Extraction: Extract the F3 and F4 dual-channel signals from the 128-channel data; 2. Signal preprocessing: Perform detrending, ICA artifact removal, spectral interpolation to remove power frequency, and resampling according to step S2; 3. Rhythmic spectrum calculation: The rhythmic spectrum in the range of 3-45Hz is calculated using the LAVI algorithm, with parameters set to a wavelet width of 5 periods and a lag of 1.5 periods; 4. Frequency band division: The ABBA algorithm was used to automatically divide individual frequency bands, and the alpha, theta, and beta1 bands were successfully detected in all subjects; 5. Feature extraction: Extract four-dimensional features according to step S5, including the alpha rhythmic asymmetry index, the average duration of bursts in the theta band of the F3 channel, the burst rate, and the average rhythmicity value of beta1. 6. Model Training and Validation: Using 10-fold cross-validation, the 256 subjects were divided into a training set (205 cases) and a test set (51 cases) to train the XGBoost binary classification model.

[0038] like Figure 2The diagram shows the LAVI rhythmic spectrum calculation structure. Using 3-45Hz as a uniform frequency axis, it covers all target frequency bands relevant to the diagnosis of depression. The diagram is divided into two sub-graphs, representing the output results of two different EEG analysis methods. The upper blue sub-graph, representing the traditional power spectrum, is based on signal energy calculation and exhibits typical 1 / f decay characteristics. It cannot separate real neural oscillations from background noise, nor can it distinguish between continuous oscillations and transient bursts—two completely different neurodynamic modes. It suffers from poor feature stability and susceptibility to artifacts. The lower red LAVI rhythmic spectrum is based on signal phase consistency calculation, naturally eliminating interference from 1 / f non-periodic noise. It does not exhibit an overall energy decay trend and only reflects the phase and temporal stability of the neural oscillations themselves. It clearly distinguishes three functional regions: a high-rhythmic band representing steady-state continuous neural activity, a low-rhythmic band representing transient burst neural activity, and a noise band corresponding to irregular interference signals. This visually demonstrates the LAVI rhythmic spectrum. The core innovation of the algorithm is to shift from traditional energy-based analysis to phase stability-based analysis, which provides a more accurate and robust analytical foundation for subsequent individualized frequency band division and extraction of biomarkers for depression.

[0039] like Figure 3 The diagram shows the ABBA automatic band boundary detection algorithm. The horizontal axis represents the frequency range of 3-45Hz, covering the entire target frequency band for depression analysis; the vertical axis represents the rhythmicity value λ(f). The red curve in the diagram is the rhythmicity spectrum output by the LAVI algorithm, which serves as the input data for this algorithm, reflecting the phase stability of neural oscillations at different frequencies. The core statistical benchmark of the algorithm is constructed using 200 sets of IAAFT (Iterative Amplitude Adjusted Fourier Transform) phase randomization substitution data: This method completely scrambles the signal phase while preserving the original signal amplitude distribution and power spectrum characteristics, generating 200 sets of random substitution signals without real neural rhythms. After simultaneously calculating the LAVI rhythmicity value for each substitution signal, a 95% confidence interval is taken for each frequency point, forming three horizontal boundaries: the upper limit of the noise band, the baseline of the median rhythmicity, and the lower limit of the noise band. The gray area between the upper and lower limits is the noise interval. The rhythmic fluctuations within this interval are at the level of random noise, lack statistical significance, and do not correspond to real stable neural rhythms.

[0040] The algorithm scans the rhythmicity curve point by point along the frequency axis, combining the noise band threshold and the median baseline to determine the effective frequency band: continuous intervals where the rhythmicity value exceeds the upper limit of the noise band and is generally above the median baseline are determined as high rhythmicity bands, corresponding to continuous neural activity with highly stable phase; continuous intervals where the rhythmicity value is below the lower limit of the noise band and is generally below the median baseline are determined as low rhythmicity bands, corresponding to neural activity with weak phase stability and existing in the form of instantaneous bursts. The figure marks three core target frequency bands corresponding to the diagnosis of depression: the 4~7Hz theta low rhythmicity band corresponds to instantaneous burst theta neural activity and is the core target for extracting abnormal electrophysiological features of depression; the 6~14Hz alpha high rhythmicity band is the high rhythmicity band with the highest rhythmicity peak and the strongest statistical significance, serving as the anchor point for full-band labeling and the basis for calculating the alpha rhythm asymmetry index; the 14.7~17.8Hz beta1 low rhythmicity band corresponds to burst beta neural activity and is one of the core feature frequency bands. This method does not require pre-setting a universal fixed frequency band boundary. It can complete individualized automatic segmentation based on the subject's own neural rhythm characteristics. At the same time, it ensures that the segmentation results correspond to real neural activity through statistical significance testing of alternative data, thus solving the problems of traditional fixed frequency band segmentation ignoring individual EEG differences and insufficient accuracy of feature extraction. Figure 4 This is a schematic diagram illustrating the core biomarkers and diagnostic results for depression, such as... Figure 4 As shown in the figure, three sets of bar charts present the quantitative differences of three types of rhythmic EEG biomarkers between healthy individuals and patients with depression. The alpha rhythmic asymmetry index was 0.4 in the healthy group and increased to 0.7 in the depressed group, corresponding to the hemispheric functional imbalance of the prefrontal cortex emotion regulation circuit in patients with depression, consistent with the research conclusion that frontal alpha asymmetry is a stable biomarker for depression. The mean duration of theta bursts was 1.2 in the healthy group and prolonged to 2.8 in the depressed group, reflecting transient neural activity abnormalities caused by prefrontal cortex neuroregulation dysfunction, consistent with the neurophysiological mechanism of theta oscillations participating in emotional processing and cognitive control. The mean beta1 rhythmicity value was 0.3 in the healthy group and increased to 0.6 in the depressed group, reflecting pathological changes in the synchronicity of beta1 band neural oscillations, consistent with the research finding of abnormally elevated beta1 band activity in depressed patients. Figure 4 The right side shows the diagnostic output of the XGBoost machine learning model for a single sample, including a quantified probability of 92.3%, a severity rating of moderate depression, and a traceable feature vector, which can achieve interpretable and quantifiable clinical auxiliary diagnosis. Figure 4The bottom label shows an accuracy of 91.5%, an AUC of 0.96, a sensitivity of 90.6%, and a specificity of 92.2%, representing the overall validation performance of the model on the dataset. This confirms that the depression recognition scheme based on rhythmic features has excellent sample discrimination, with both the false negative and false positive rates at a low level, thus fully validating the effectiveness and reliability of the scheme.

[0041] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the following steps of a method for an objective identification system for depression based on rhythmic analytic spectrum: Dual-channel EEG signals were collected from subjects at rest and transmitted to a data processing platform. The system acquires raw dual-channel EEG signals and obtains core biomarker features based on the raw dual-channel EEG signals and a biomarker extraction model. It then combines the core biomarker features with a pre-trained depression recognition model to output depression recognition results. Based on the preprocessed EEG signals, the rhythmic values ​​of the two channel signals at each frequency point within the specified range were calculated respectively; Based on noise bands generated from phase randomization substitution data, each channel is automatically divided into individual-specific high-rhythmic bands and low-rhythmic bands according to the rhythmicity value of each frequency point, and multiple target frequency bands are located. Core biomarker features are extracted based on the target frequency band localization results.

[0042] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the following steps of a method for an objective identification system for depression based on rhythmic spectral analysis: Dual-channel EEG signals were collected from subjects at rest and transmitted to a data processing platform. The system acquires raw dual-channel EEG signals and obtains core biomarker features based on the raw dual-channel EEG signals and a biomarker extraction model. It then combines the core biomarker features with a pre-trained depression recognition model to output depression recognition results. Based on the preprocessed EEG signals, the rhythmic values ​​of the two channel signals at each frequency point within the specified range were calculated respectively; Based on noise bands generated from phase randomization substitution data, each channel is automatically divided into individual-specific high-rhythmic bands and low-rhythmic bands according to the rhythmicity value of each frequency point, and multiple target frequency bands are located. Core biomarker features are extracted based on the target frequency band localization results.

[0043] Example 5 This embodiment provides a program product, which is a computer program product, including a computer program. When the computer program is executed by a processor, it implements the following steps of the method for the objective identification system of depression based on rhythmic analytic spectrum: Dual-channel EEG signals were collected from subjects at rest and transmitted to a data processing platform. The system acquires raw dual-channel EEG signals and obtains core biomarker features based on the raw dual-channel EEG signals and a biomarker extraction model. It then combines the core biomarker features with a pre-trained depression recognition model to output depression recognition results. Based on the preprocessed EEG signals, the rhythmic values ​​of the two channel signals at each frequency point within the specified range were calculated respectively; Based on noise bands generated from phase randomization substitution data, each channel is automatically divided into individual-specific high-rhythmic bands and low-rhythmic bands according to the rhythmicity value of each frequency point, and multiple target frequency bands are located. Core biomarker features are extracted based on the target frequency band localization results.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0045] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0047] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An objective identification system for depression based on rhythmic analytical spectrum, characterized in that, include: EEG acquisition equipment is used to acquire dual-channel EEG signals from subjects in a resting state and transmit them to a data processing platform; The data processing platform is connected to the EEG acquisition device to acquire raw dual-channel EEG signals and obtain core biomarker features based on the raw dual-channel EEG signals and the biomarker extraction model. The core biomarker features are combined with the pre-trained depression recognition model to output the depression recognition result. The biometric extraction model includes a dual-channel rhythmic spectrum calculation module, an individual frequency band automatic division module, a depression-specific feature extraction module, and a depression recognition module. The dual-channel rhythmicity spectrum calculation module is used to calculate the rhythmicity values ​​of the two channels at each frequency point within a specified range based on the preprocessed EEG signals. The individual frequency band automatic segmentation module is used to generate noise bands based on phase randomization substitution data. It automatically segments each channel into individual-specific high-rhythm bands and low-rhythm bands according to the rhythmicity value of each frequency point, and locates multiple target frequency bands. The depression-specific feature extraction module is used to extract core biomarker features based on the target frequency band positioning results; The host computer is connected to the data processing platform and is used to acquire and display the depression identification results.

2. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 1, characterized in that, The dual-channel rhythmic spectrum calculation module includes a time-frequency decomposition module and a rhythmic value calculation module; The time-frequency decomposition module is used to perform time-frequency decomposition on the preprocessed dual-channel EEG signal to obtain the complex wavelet coefficient matrix corresponding to each time sampling point and each analysis frequency. The rhythmicity value calculation module is used to combine the complex wavelet coefficient matrix of each time sampling point and each analysis frequency to calculate the mean value of the phase difference vector between the original signal and its 1.5-period replica, obtain the rhythmicity value of each frequency point, and generate the full-band rhythm spectrum.

3. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 1, characterized in that, In the dual-channel rhythmicity spectrum calculation module, the formula for calculating the rhythmicity value at each frequency point within a specified range is as follows: ， in, For frequency f The corresponding rhythmicity value ranges from [0,1]. A higher value indicates a more stable signal phase and a stronger oscillation duration. f The target analysis frequency is defined within the target analysis frequency band; t represents the time sampling point of the EEG signal, and T represents the total number of sampling points for the entire signal. This represents the number of sampling points contained in 1.5 oscillation cycles corresponding to frequency f; These are the complex wavelet coefficients at time point t and frequency f after complex Morlet wavelet time-frequency decomposition; These are the complex wavelet coefficients at time point t+L(f) and frequency f after complex Morlet wavelet time-frequency decomposition. Lagging The complex conjugate of the complex wavelet coefficients at the corresponding time and frequency points of the EEG signal for each cycle.

4. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 1, characterized in that, The individual frequency band automatic segmentation module includes a noise band generation module, an individual-specific preliminary segmentation module, and a target frequency band positioning module. The noise band generation module is used to generate phase randomization substitution data, calculate the rhythmic spectrum of each substitution data, and determine the noise band within a set confidence interval; The individual-specific preliminary segmentation module is used to classify regions that are above the baseline and exceed the upper limit of the noise band as high-rhythm bands, and regions that are below the baseline and exceed the lower limit of the noise band as low-rhythm bands, using the median of the rhythmicity values ​​of all frequency points as the baseline. The target frequency band positioning module is used to automatically label each frequency band with the alpha high rhythmic band within a set range as the anchor point, and to locate the alpha high rhythmic band, the theta low rhythmic band, and the beta1 low rhythmic band of the two channels respectively.

5. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 1, characterized in that, The dual-channel rhythmic spectrum calculation module calculates the rhythmic value of each frequency point in the range of 3~45Hz for the two channel signals.

6. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 1, characterized in that, The depression-specific feature extraction module includes an alpha rhythmic asymmetric index calculation module, a theta band oscillation burst detection module, and a beta1 average rhythmic feature calculation module. The alpha rhythmic asymmetry index calculation module is used to calculate the average value of rhythmicity values ​​at all frequency points within the alpha band of channels F3 and F4, and to calculate the asymmetry index based on the average value to obtain the alpha rhythmic asymmetry index. The oscillation burst detection module in the theta band is used to calculate the average duration and burst rate of all bursts in the theta band of the F3 channel; The beta1 average rhythmicity characteristic calculation module is used to calculate the average value of the rhythmicity values ​​of all frequency points within the beta1 band of channels F3 and F4. The average value of both values ​​is taken to obtain the beta1 average rhythmicity characteristic.

7. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 6, characterized in that, The oscillation burst detection module within the theta band is used to calculate the average duration and burst rate of all bursts in the theta band of channel F3, including: Extract all time-series power data within the theta band of channel F3, calculate the 90th percentile P90 of the power, and use it as the burst power threshold; The timing signal is traversed frame by frame, and signal segments with instantaneous power > P90 and duration ≥ 2 cycles are selected. All valid burst events are counted, and the total number of bursts M and the duration of each burst are recorded. ; Combined with the total signal acquisition time theta bandwidth Calculate the standardized burst rate .

8. The objective identification system for depression based on rhythmic analytical spectrum as described in claim 1, characterized in that, The depression identification model uses a gradient boosting tree ensemble learning model to classify the condition into three levels: mild, moderate, and severe based on the probability values ​​output by the model and the HAMD score.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method steps corresponding to the objective identification system for depression based on rhythmic analytic spectrum as described in any one of claims 1-8.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method steps corresponding to the objective identification system for depression based on rhythmic analytical spectrum as described in any one of claims 1-8.

Images