Ultrahigh-frequency electroencephalogram signal-based depression classification model and application thereof

Abstract

The invention discloses a depression classification model based on ultrahigh frequency electroencephalogram signals and application thereof, and relates to the technical field of signal processing, and the depression classification model comprises the following steps: filtering collected original electroencephalogram signals by using fast Fourier transform, only retaining a frequency band of 0.5-150Hz, and filtering signals higher than 150Hz as noise; performing frequency band division on the filtered signal to obtain 11 different frequency bands; performing frequency spectrum shift modulation on each frequency band; extracting an energy feature (relative energy), a complexity feature (Lempel-Ziv complexity) and an entropy feature (approximate entropy and sample entropy) from each frequency band signal; performing weighted fusion on the calculated features to generate a comprehensive feature vector; and training a classification model based on the comprehensive feature vector by using a machine learning algorithm of a random forest.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and more specifically, to a depression classification model based on ultra-high frequency electroencephalogram signals and its application. Background Art

[0002] Depression is a common mental illness that seriously affects the quality of life of patients. At present, the diagnosis of depression mainly relies on questionnaires and the judgment of clinicians, and lacks objective and rapid diagnostic tools. As a non-invasive means of monitoring brain function, EEG signals have been widely used in the study of neuropsychiatric diseases in recent years. In particular, ultra-high frequency EEG signals, with a frequency band above 50 Hz, are believed to provide physiological information closely related to mental state. However, how to extract effective features from ultra-high frequency EEG signals and use these features to classify depression remains a challenging problem.

[0003] Based on the above technical background, the present invention discloses a depression classification model based on ultra-high frequency EEG signals and its application, which relates to the field of signal processing technology. This is a method for screening and diagnosing depression, which solves the problems and pain points of traditional diagnostic methods and improves the efficiency of depression detection and depression research. The design of this model can be applied to different types of EEG signal data, especially in the field of mental illness diagnosis, and has great promotion potential. It is not limited to depression, and may also be applied to the screening and classification of other neurological and psychological diseases in the future. Summary of the invention

[0004] In order to overcome the defects of the above-mentioned prior art such as high possibility of missed diagnosis and high time cost, the present invention proposes to construct a depression classification model based on a machine learning algorithm, and provides a method that can conveniently, quickly and effectively reduce the risk of missed diagnosis.

[0005] The technical solution of the present invention: a depression classification model based on ultra-high frequency EEG signals, comprising the following steps:

[0006] Step 1: Use a single-channel headband as a collector to obtain the EEG signals of the subjects as a data set. One group of subjects are healthy people, and another group of subjects are patients with confirmed depression.

[0007] Step 2: Filter the signal through fast Fourier transform, retain only the frequency band from 0.5Hz to 150Hz, remove the spectrum above 150Hz, divide the preprocessed signal into 11 different frequency bands, and perform spectrum modulation on the signal of each frequency band to enhance the expression ability of the features;

[0008] Step 3: Construct, design and implement a comprehensive feature classification network, including preliminary feature extraction and feature weighted fusion; build a comprehensive feature classification network based on random forest, randomly shuffle the obtained comprehensive features as training sample data, normalize the data, map all features of the data to the same scale, and obtain training sets and validation sets;

[0009] Step 4: Use the training set to train the constructed comprehensive feature classification network, set the parameters of the random forest to 100 trees, start with the maximum depth of no limit, gradually try to increase the depth, and observe the model performance, try different minimum sample numbers to find the best value, and use grid search to find the best hyperparameter settings;

[0010] Step 5: Input the comprehensive feature vector in the test set into the trained random forest model for classification prediction and analyze the sensitivity, specificity, and accuracy of the model to evaluate the classification performance. Adjust and optimize the model as needed.

[0011] The present invention has the following beneficial effects:

[0012] A depression classification model proposed in the present invention first uses modulated EEG signals for feature extraction and feature fusion to enhance feature expression capabilities and reduce the calculation time of the classification model, and then uses a comprehensive feature classification network for classification, which can effectively improve the recognition and sorting accuracy of depression EEG signals. The addition of high-frequency components allows multi-scale analysis of signals. By combining EEG signals of different frequencies, more comprehensive EEG activity information can be obtained. The model has excellent classification effect, can effectively distinguish patients with depression from healthy individuals, reduce false positives and false negatives, reduce the need for manual intervention, and can complete the screening task in a shorter time. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0014] Figure 1 It is a framework of a depression classification model based on ultra-high frequency EEG signals of the present invention;

[0015] Figure 2 This is a flow chart of a depression classification model based on ultra-high frequency EEG signals of the present invention;

[0016] Figure 3 This is a structural schematic diagram of a depression classification model based on ultra-high frequency EEG signals of the present invention. DETAILED DESCRIPTION

[0017] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0018] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0019] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0020] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0021] Example

[0022] The embodiment of the present invention provides a method for implementing a depression classification model based on ultra-high frequency electroencephalogram signals, such as Figures 2 to 3 shown.

[0023] This example includes the following steps:

[0024] Step 1: Use a portable and easy-to-operate EEG headband to collect EEG signals. The collected EEG signals are used as data sets, and the collected data are transmitted from the headband to the connected computer via WiFi.

[0025] Step 2: Preprocess the dataset. First, the raw EEG signal is converted from the time domain to the frequency domain using Fast Fourier Transform (FFT). This transformation allows us to analyze the spectral distribution of the signal. In the spectral domain, a bandpass filter is applied to retain only the frequency band from 0.5 Hz to 150 Hz to remove the spectral components above 150 Hz, which usually contain high-frequency noise. Next, the filtered signal is frequency-banded and divided into 11 different frequency bands. Each frequency band represents a different frequency range in the EEG signal. This division is based on the standard frequency bands of EEG waves, δ (0.5 Hz to 4 Hz), θ (4 Hz to 8 Hz), α (8 Hz to 12 Hz), SMR (12 Hz to 15 Hz), midβ (15 Hz to 20 Hz), highβ (20 Hz to 30 Hz), γ (30 Hz to 50 Hz), highγ (50 Hz to 80 Hz), UHF band 1 (80 Hz to 100 Hz), UHF band 2 (100 Hz to 125 Hz) and UHF band 3 (125 Hz to 150 Hz). The spectrum of each frequency band signal is modulated so that the starting frequency of each EEG wave in its signal band is moved to zero.

[0026] Step 3: Build, design and implement a comprehensive feature classification network. The core of this network includes two key stages: preliminary feature extraction and feature weighted fusion. In the preliminary feature extraction stage, various features are extracted from the EEG signal after spectral modulation processing, and the extracted features are weighted and fused. Next, a comprehensive feature classification network is built based on the random forest algorithm. During the network construction process, the comprehensive feature vectors obtained from the feature extraction and fusion stages are used as training sample data. In order to ensure the effectiveness of the model, the training sample data is first randomly shuffled, and finally, the processed data is divided into a training set and a validation set.

[0027] Step 4: Neural network training and parameter optimization. The number of trees in the random forest was set to 100 trees. The maximum depth parameter determines the maximum split depth of each decision tree. Initially, the maximum depth was set to unlimited, and then the depth was gradually increased for experimentation. The best setting was determined by experimenting with different minimum sample values. This process involves starting with a small number of samples, gradually increasing and observing the changes in model performance. In order to systematically find the best hyperparameter settings, a grid search method was used. Grid search evaluates the model performance of each parameter configuration by defining a range of values ​​for a parameter and cross-validating all possible parameter combinations.

[0028] Step 5: Perform classification prediction and classification performance evaluation. After completing the training and parameter optimization of the random forest model, the next step is to input the comprehensive feature vector in the test set into the trained model for classification prediction. By comparing the prediction results with the actual labels, calculate the accuracy index. Based on the calculation results of sensitivity, specificity, and accuracy, analyze the classification performance of the model. If the performance of the model does not meet expectations, further adjustments and optimization may be required.

[0029] The EEG signal data in the present invention comes from 9 subjects. The ages of these subjects are between 18 and 24 years old. Among these 9 subjects, 4 of them have been medically diagnosed with depression before, and the remaining 5 have been certified as healthy subjects by medical staff. The subjects performed 6 activities in a highly concentrated state when collecting EEG. Each activity lasts about 10 to 15 minutes. During the activities, all subjects maintained a good state. This classification model can effectively classify these 9 subjects and correctly classify them into the corresponding categories, reflecting the robustness and classification ability of the model when processing small sample data.

[0030] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0031] Obviously, the above embodiments of the present invention are examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A depression classification model based on ultra-high frequency electroencephalogram signals and its application, characterized in that: The following steps are involved: S1. Use fast Fourier transform to filter the collected raw EEG signals, and only keep the spectrum of 0.5Hz-150Hz; S2. Divide the filtered signal into frequency bands to obtain 11 different frequency bands; S3. Spectrum modulate each frequency band; S4. Extract the energy characteristics, complexity characteristics and entropy characteristics of each frequency band signal; S5. Perform weighted fusion on the calculated features to generate a comprehensive feature vector; S6. Use the random forest algorithm to train a classification model based on the comprehensive feature vector.

2. The depression classification model based on ultra-high frequency EEG signals and its application according to claim 1, characterized in that: In the step S1, the collected original EEG signal is filtered using fast Fourier transform, and only the spectrum of 0.5 Hz-150 Hz is retained. Specifically, let x(n) be the real value of the EEG signal at time index n, let N be its length, and X(k) be the kth coefficient of discrete Fourier transform; let When , the discrete Fourier transform of x(n) is defined as: Fast Fourier transform is an effective implementation of discrete Fourier transform. Fast Fourier transform divides the EEG signal represented in the time domain into an even time index sequence and an odd time index sequence. Assuming N is an even number, let x odd (n) = x(2n+1), where Let x be an odd time index sequence. even (n) = x(2n), where is an even time index sequence, let the kth discrete Fourier coefficient odd time index sequence be F odd (k) namely: Then the even time index sequence of the kth discrete Fourier coefficient is Then, X(k) can be obtained as the Fast Fourier coefficient as follows: In order to perform filtering, we first calculate the coefficients of the fast Fourier transform of the EEG signal and represent it in the frequency domain. Here, only the fast Fourier transform coefficients corresponding to the frequency band between 0.5 Hz and 150 Hz are retained, and the remaining coefficients are set to zero.

3. A depression classification model based on ultra-high frequency EEG signals and its application according to claim 2, characterized in that: In the step S2, the filtered signal is divided into frequency bands to obtain 11 different frequency bands, specifically, the obtained frequency band between 0.5Hz-150Hz is segmented; the EEG signal frequency band includes δ (0.5Hz to 4Hz), θ (4Hz to 8Hz), α (8Hz to 12Hz), SMR (12Hz to 15Hz), midβ (15Hz to 20Hz), highβ (20Hz to 30Hz), γ (30Hz to 50Hz), highγ (50Hz to 80Hz), ultra-high frequency band 1 (80Hz to 100Hz), ultra-high frequency band 2 (100Hz to 125Hz) and ultra-high frequency band 3 (125Hz to 150Hz).

4. The depression classification model based on ultra-high frequency EEG signals and its application according to claim 3, characterized in that: In step S3, spectrum modulation is performed on each frequency band. Specifically, modulation is to move the signal from its signal frequency band to the zero frequency band. In this paper, amplitude modulation is performed so that the starting frequency of each brain wave in its signal band is moved to zero. Specifically, x(n) is the brain wave signal represented in the time domain. First, the starting frequency and the ending frequency of x(n) are estimated; let f c is the starting frequency of x(n). The second step is to multiply x(n) point by point in the time domain by cos(2πf c n); get x mod (n) = 2x(n)cos(2πf c n), there are extra harmonics in the frequency band; mod (n) Perform fast Fourier transform, Let L be X mod (k), where L is assumed to be an even number, and p is X mod (k) The number of points of the fast Fourier transform coefficients; Let Z(k)=[X mod (1:p)0 L-2p+1 X mod (L-p+2:L)], at this time, the harmonics generated by the modulation are filtered out, and only the spectrum coefficient with the starting frequency as the zero point is retained; the inverse fast Fourier transform is performed on Z(k) to obtain the frequency band after the spectrum modulation, that is:

5. The depression classification model based on ultra-high frequency EEG signals and its application according to claim 4, characterized in that: In step S4, the energy feature, complexity feature and entropy feature of each frequency band signal are extracted. Specifically, the energy feature uses the frequency domain relative energy, and the frequency domain relative energy It is defined as the ratio of the signal energy in the [k1, k2] frequency band to the total energy in the entire frequency domain: Where X(k) is the discrete Fourier transform of signal x(n), k1 is the starting index of the frequency segment, k2 is the ending index of the frequency segment, and N is the total length of the signal; Lempel-Ziv complexity converts the signal x(n) into a binary sequence, sets a threshold θ to set the part of the signal value greater than the threshold to 1, and the part less than or equal to the threshold to 0; approximate entropy is a statistical indicator used to measure the degree of regularity or predictability in a time series or data series, and its formula is: apEn(m,r,N)=φ m (r)-φ m+1 (r), where m is the length of the subsequences to be compared, r is the tolerance parameter that determines the maximum difference between two data points for them to be considered similar, and φ m (r) and φ m+1 (r) is the average similarity function when the embedding dimension is m and m+1.

6. The depression classification model based on ultra-high frequency EEG signals and its application according to claim 5, characterized in that: In step S5, the calculated features are weighted and fused to generate a comprehensive feature vector. Specifically, the energy feature obtained is E = [E1, E2, ..., E N ], let the complexity feature be C = [C1, C2, ..., C N ], let the entropy feature be H = [H1, H2, ..., H N ], then the weighted feature is: The final fusion feature F fusion For: F fusion =α·E+β·C+γ·H, and the cross-validation technique is used to determine the optimal weight coefficient. The selected weights of α, β, and γ are 0.45, 0.23, and 0.32, respectively.

7. The depression classification model based on ultra-high frequency EEG signals and its application according to claim 6, characterized in that: In the step S6, the random forest algorithm is used to train the classification model based on the comprehensive feature vector. Specifically, the EEG data in each activity is divided into 70% and 30% ratios, 70% of the data is used for model training (training set), and 30% of the data is used for model evaluation (test set); this division method ensures that the model can learn enough information during the training process, while retaining a portion of the data for independent testing to verify the generalization ability of the model, and the feature vectors in the test set are used to evaluate the performance of the proposed method, and the classification results are obtained by inputting the test data into the trained model; in order to quantify the performance of the model, several common evaluation indicators are usually used, such as accuracy, sensitivity and specificity.

8. An application of a depression classification model based on ultra-high frequency electroencephalogram signals, characterized in that: The application can be used for early screening, diagnosis and monitoring of depression. The application includes an implementation program of a depression classification model based on ultra-high frequency electroencephalogram signals. The application of a depression classification model based on ultra-high frequency electroencephalogram signals is used for early screening of depression to implement the following steps: S1. Use fast Fourier transform to filter the collected raw EEG signal, and only keep the spectrum of 0.5Hz-150Hz; S2. Divide the filtered signal into frequency bands to obtain 11 different frequency bands; S3. Spectrum modulate each frequency band; S4. Extract the energy characteristics, complexity characteristics and entropy characteristics of each frequency band signal; S5. Perform weighted fusion on the calculated features to generate a comprehensive feature vector; S6. Use the random forest algorithm to train a classification model based on the comprehensive feature vector.

9. The application of a depression classification model based on ultra-high frequency EEG signals according to claim 8, characterized in that: This model can be applied to the early screening, diagnosis and monitoring of depression. Specifically, the EEG signals of the subjects are first collected, then filtered using the fast Fourier transform EEG signals and the signals are divided into 11 frequency bands; then, the spectrum of each frequency band is modulated to extract features such as energy, complexity and entropy; then, these features are weighted and fused to generate a comprehensive feature vector; finally, the classification model is trained using the random forest algorithm to achieve accurate classification and diagnosis of depression.

Images