Depression identification device and electronic equipment

Abstract

The invention belongs to the technical field of medical equipment, and mainly provides a depression recognition device and electronic equipment. Electrocardiosignals and electroencephalogram signals of a to-be-detected object are input into a target electrocardiosignal recognition model and a target electroencephalogram recognition model which are obtained through pre-training respectively; and obtaining a first depression identification result and a second depression identification result, constructing a prediction matrix based on the first depression identification result and the second depression identification result, and inputting the prediction matrix into a target support vector machine model obtained by pre-training to obtain a depression target identification result. According to the method, whether the to-be-detected object suffers from depression or not is identified based on the target electrocardiogram identification model, the target electroencephalogram identification model and the target support vector machine model by using the electrocardiogram signal and the electroencephalogram signal of the to-be-detected object, so that the accuracy and the efficiency of depression identification are improved; the technical problems of low depression identification accuracy and low efficiency in the existing scheme are solved.

Description

Technical Field

[0001] The present application belongs to the technical field of medical equipment, and in particular, relates to a depression identification device and electronic equipment. Background Art

[0002] Depression is a syndrome characterized by low mood. Different from normal depression, it is characterized by significant and persistent depression as the basic clinical manifestation, accompanied by corresponding abnormal thinking and behavior. Patients are depressed, inferior, depressed, and even pessimistic.

[0003] In recent years, the incidence of depression has increased year by year, especially among adolescents. Existing diagnostic methods mainly rely on assessment scales and clinicians' subjective judgments, which are time-consuming and inefficient, and there is also the risk of misdiagnosis and missed diagnosis. Summary of the invention

[0004] The present application provides a depression identification device and electronic device, which can solve the technical problems of depression identification in existing solutions, such as long time consumption, low efficiency and low accuracy.

[0005] A first aspect of an embodiment of the present application provides a depression identification device, the depression identification device comprising:

[0006] A signal acquisition module, used to acquire the electrocardiogram (ECG) and electroencephalogram (EEG) signals of the subject to be tested;

[0007] A first recognition module, configured to input the electrocardiogram signal into a pre-trained target electrocardiogram recognition model to obtain a first depression recognition result output by the target electrocardiogram recognition model;

[0008] A second recognition module is used to input the EEG signal into a pre-trained target EEG recognition model to obtain a second depression recognition result output by the target EEG recognition model;

[0009] The third recognition module is used to construct a prediction matrix based on the first depression recognition result and the second depression recognition result, and input the prediction matrix as an input feature into a pre-trained target support vector machine model to obtain a depression target recognition result output by the target support vector machine model.

[0010] A second aspect of an embodiment of the present application provides an electronic device, comprising the depression identification device described in the first aspect above.

[0011] The depression identification device and electronic device proposed in the embodiments of the present application obtain a first depression identification result and a second depression identification result by inputting the ECG signal and EEG signal of the subject to be tested into a pre-trained target ECG identification model and a target EEG identification model respectively, and constructing a prediction matrix based on the first depression identification result and the second depression identification result, and inputting the prediction matrix into a pre-trained target support vector machine model to obtain a target depression identification result. The device and electronic device utilize the ECG signal and EEG signal of the subject to be tested, and identify whether the subject to be tested suffers from depression based on the target ECG identification model, the target EEG identification model and the target support vector machine model, thereby improving the accuracy and efficiency of depression identification, solving the technical problems of low accuracy and low efficiency of depression identification in the existing scheme, and providing an innovative and practical solution for the early screening and diagnosis of depression. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a first structural schematic diagram of a depression identification device provided in an embodiment of the present application.

[0013] Figure 2 This is a second structural schematic diagram of the depression identification device provided in an embodiment of the present application.

[0014] Figure 3 A schematic diagram of a preprocessed EEG signal provided in an embodiment of the present application.

[0015] Figure 4 A schematic diagram of a preprocessed ECG signal provided in an embodiment of the present application.

[0016] Figure 5 A schematic diagram of the structure of the initial EEG recognition model and the initial ECG recognition model provided in the embodiments of the present application.

[0017] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] It should be understood that in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0020] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0021] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the phrases "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. appearing in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways.

[0022] In recent years, the incidence of depression has increased year by year, especially among adolescents. Existing diagnostic methods mainly rely on assessment scales and clinicians' subjective judgment, which is time-consuming, inefficient, and has the risk of misdiagnosis and missed diagnosis.

[0023] Based on this, the present application provides a depression identification device and electronic device, which can solve the technical problem of low depression identification accuracy in existing solutions.

[0024] like Figure 1 As shown, it is a structural schematic diagram of a depression identification device provided in an embodiment of the present application. The depression identification device 100 can be implemented by software or hardware or a combination of software and hardware, and includes: a signal acquisition module 10, a first identification module 20, a second identification module 30 and a third identification module 40.

[0025] Among them, the signal acquisition module 10 is used to acquire the ECG signal and EEG signal of the object to be tested; the first recognition module 20 is used to input the ECG signal into a pre-trained target ECG recognition model to obtain a first depression recognition result output by the target ECG recognition model; the second recognition module 30 is used to input the EEG signal into a pre-trained target EEG recognition model to obtain a second depression recognition result output by the target EEG recognition model; the third recognition module 40 is used to construct a prediction matrix based on the first depression recognition result and the second depression recognition result, and input the prediction matrix as an input feature into a pre-trained target support vector machine model to obtain a depression target recognition result output by the target support vector machine model.

[0026] In this embodiment, the types of depression target recognition results may include two types: the subject to be tested suffers from depression and the subject to be tested does not suffer from depression.

[0027] In one embodiment of the present application, the depression target recognition results may also include recognition results of the subject to be tested suffering from mild depression, moderate depression, severe depression, etc. The present application does not limit the classification type of the depression target recognition results output by the target support vector machine model.

[0028] The depression identification device proposed in the embodiment of the present application obtains a first depression identification result and a second depression identification result by inputting the ECG signal and EEG signal of the subject to be tested into a pre-trained target ECG identification model and a target EEG identification model respectively, and constructs a prediction matrix based on the first depression identification result and the second depression identification result, and inputs the prediction matrix into a pre-trained target support vector machine model to obtain a target depression identification result. The device utilizes the ECG signal and EEG signal of the subject to be tested and uses a bimodal decision fusion algorithm based on the target ECG identification model, the target EEG identification model and the target support vector machine model to identify whether the subject to be tested suffers from depression, thereby improving the accuracy of depression identification, being able to effectively identify mild depression, solving the technical problem of low depression identification accuracy in the existing scheme, being able to effectively reduce the occurrence of misdiagnosis and missed diagnosis, and providing an innovative and practical solution for the early screening and diagnosis of depression.

[0029] It should be noted that the first depression recognition result output by the target ECG recognition model and the second depression recognition result output by the target EEG recognition model can be floating-point data ranging from 0 to 1. The target support vector machine model constructs a prediction matrix based on the first depression recognition result and the second depression recognition result, and uses the prediction matrix as a new input feature to identify whether the subject is suffering from depression. It realizes depression recognition based on the target ECG recognition model and the target EEG recognition model, improves the accuracy of depression recognition, can effectively identify mild depression, and solves the technical problem of low depression recognition accuracy in existing solutions. It can effectively reduce the occurrence of misdiagnosis and missed diagnosis, and provides an innovative and practical solution for the early screening and diagnosis of depression.

[0030] In one embodiment of the present application, the electrocardiogram signal and the electroencephalogram signal acquired by the signal acquisition module may both be single-channel signals.

[0031] In the relevant depression recognition schemes, most of them are based on unimodal data (i.e., a single type of data, such as a single ECG signal or a single EEG signal), and most of them rely on multi-channel EEG data for depression recognition. However, due to the large number of channels, high cost, and complex operation, it is not convenient for popularization and use in clinical applications. The present application realizes the use of a single-channel ECG signal and a single-channel EEG signal, a total of two channels of bimodal data to identify whether the subject suffers from depression, breaking the limitations of existing algorithms, reducing the cost of depression recognition, and having a high application value in clinical applications.

[0032] In an actual application, the signal acquisition module, the first recognition module, the second recognition module and the third recognition module can be implemented by the host computer. The signal acquisition module can be specifically used to receive the ECG signal and EEG signal of the object to be tested collected by the signal acquisition module of the lower computer to obtain the ECG signal and EEG signal of the object to be tested.

[0033] In one embodiment of the present application, the signal acquisition module of the lower computer can be implemented through the OpenBCI module, and the signal acquisition module of the upper computer can output signal acquisition instructions to the OpenBCI module of the lower computer, control and receive the ECG signals and EEG signals collected by the OpenBCI module of the lower computer, and transmit the ECG signals and EEG signals to the upper computer, and then the first recognition module, the second recognition module and the third recognition module of the upper computer automatically perform depression identification.

[0034] Among them, the OpenBCI module is mainly used to collect the electroencephalogram (EEG) signal and single-lead electrocardiogram (ECG signal) of the prefrontal lobe Fp1 or Fp2 of the subject to be tested.

[0035] In the embodiment of the present application, the core components of the OpenBCI module are the ADS1299 chip and the PIC32MX250F128B microprocessor. Among them, the high common mode rejection ratio of the ADS1299 chip effectively suppresses common mode interference and ensures the quality and stability of the signal; the adjustability of the data rate allows users to flexibly set the sampling frequency according to actual needs to adapt to different application scenarios; the low power consumption design makes it have a wide range of application prospects in portable devices.

[0036] Since the size of EEG and ECG signals is very weak, they are easily disturbed by external noise. Therefore, in order to improve the accuracy of depression recognition, the original EEG and ECG signals can be preprocessed, and then depression can be recognized based on the preprocessed EEG and ECG signals.

[0037] like Figure 2As shown, in one embodiment of the present application, the above-mentioned depression identification device may further include a preprocessing module 11. Before the first identification module, the second identification module and the third identification module on the host computer automatically identify depression, the preprocessing module may be used to preprocess the ECG signal and the EEG signal to obtain a preprocessed ECG signal and a preprocessed EEG signal.

[0038] In one embodiment of the present application, the pre-processing module 11 may include a filtering unit, a wavelet decomposition and reconstruction unit, and an artifact checking unit.

[0039] Among them, the filtering unit is used to perform bandpass filtering and notch filtering on the ECG signal and the EEG signal respectively; the wavelet decomposition and reconstruction unit is used to perform wavelet decomposition on the ECG signal and the EEG signal after the bandpass filtering and the notch filtering, and perform reconstruction processing after removing artifacts; the artifact inspection unit is used to load the EEG signal after the wavelet decomposition and reconstruction processing using the first visualization tool, perform notch filtering on the EEG signal after the wavelet decomposition and reconstruction processing, and receive the user's artifact removal operation to obtain the preprocessed EEG signal; the artifact inspection unit is also used to load the ECG signal after the wavelet decomposition and reconstruction processing using the second visualization tool, and receive the user's high-frequency noise signal removal operation to obtain the preprocessed ECG signal.

[0040] Specifically, experiments have found that the EEG signal after wavelet decomposition and reconstruction is still likely to contain electrooculographic artifacts, while the ECG signal is only likely to contain high-frequency noise. Therefore, the artifact detection unit needs to perform notch filtering on the EEG signal after wavelet decomposition and reconstruction before receiving the user's artifact removal operation, while the ECG signal after wavelet decomposition and reconstruction does not need to be notch filtered.

[0041] In practical applications, EEG and ECG signal noise mainly comes from electromagnetic interference, other physiological signal interference and body movement, including: 50Hz power frequency noise, ECG signal and myoelectric signal interference, etc. In one embodiment of the present application, in the process of bandpass filtering and notch filtering of ECG and EEG signals, the ECG and EEG signals can be bandpass filtered at 2-45Hz and notch filtered at 49-51Hz respectively to remove some electrooculographic artifacts.

[0042] In one embodiment of the present application, the wavelet decomposition and reconstruction unit can use wavelet sym3 / db6 to perform 5-layer decomposition on the ECG signal and EEG signal that have been bandpass filtered and notch filtered, reconstruct the detail coefficients of layers 2 to 5, and perform wavelet decomposition and reconstruction processing on the ECG signal and EEG signal that have been bandpass filtered and notch filtered to further remove the electrooculogram artifacts of the ECG signal and EEG signal.

[0043] In one embodiment of the present application, the artifact detection unit can use Matlab's open source toolbox EEGLAB to load the EEG signal after wavelet decomposition and reconstruction, and perform notch filtering on the EEG signal after wavelet decomposition and reconstruction, and then receive the user's artifact removal operation to obtain the pre-processed EEG signal. That is, the electrooculographic artifacts are removed by manual recognition.

[0044] In one embodiment of the present application, the artifact detection unit can also use the single-lead electrocardiograph singleAnalyzer (second visualization tool) to load the ECG signal processed by wavelet decomposition and reconstruction, and receive the user's high-frequency noise signal removal operation to obtain the preprocessed ECG signal.

[0045] like Figure 3 and Figure 4 As shown, it is a schematic diagram of the preprocessed EEG signal and the preprocessed ECG signal. It can be seen that when the preprocessing module of the present application is used to preprocess the EEG signal and ECG signal collected by the signal acquisition module, a better preprocessing effect can be achieved.

[0046] In one embodiment of the present application, the depression identification device may also include a model training unit for training an initial ECG recognition model, an initial EEG recognition model, and a support vector machine model to be trained to obtain a target ECG recognition model, a target EEG recognition model, and a target support vector machine model.

[0047] Among them, the initial ECG recognition model and the initial EEG recognition model can be deep learning neural network models.

[0048] In one embodiment of the present application, the initial ECG recognition model and the initial EEG recognition model may be CNN_LSTM models and have the same model structure.

[0049] In one embodiment of the present application, the initial ECG recognition model and the initial EEG recognition model may include a feature extraction unit, a temporal learning unit and a classification unit; wherein the feature extraction unit includes six convolutional layers, a pooling layer and a batch normalization layer (BN layer) are arranged between two adjacent convolutional layers in the six convolutional layers, and a pooling layer is arranged between the last convolutional layer and the temporal learning unit; the temporal learning unit includes two long short-term memory network layers (LSTM_1 layer and LSTM_2 layer); the classification unit includes two fully connected layers and an activation function layer, and a random dropout layer (Dropout layer) is arranged between the two fully connected layers.

[0050] For the sake of distinction, the feature extraction unit, the time series learning unit and the classification unit of the initial EEG recognition model are defined as the first feature extraction unit, the first time series learning unit and the first classification unit. The feature extraction unit, the time series learning unit and the classification unit of the initial ECG recognition model are defined as the second feature extraction unit, the second time series learning unit and the second classification unit.

[0051] Among them, the first feature extraction unit is used to extract local features of sample EEG signals; the first time series learning unit is used to learn the relationship between the sample EEG signals at the previous moment and the current moment, and output feature information of the sample EEG signals based on the local features and the learned relationship; the first classification unit is used to perform classification based on the feature information output by the first time series learning unit to obtain the corresponding depression prediction result.

[0052] The second feature extraction unit is used to extract local features of the sample ECG signal; the second time series learning unit is used to learn the relationship between the sample ECG signal at the previous moment and the current moment, and output feature information of the sample ECG signal based on the local features and the learned relationship; the second classification unit is used to perform classification based on the feature information output by the second time series learning unit to obtain the corresponding depression prediction result.

[0053] like Figure 5 As shown, the first feature extraction unit 50 includes a first convolution layer 501, a second convolution layer 502, a third convolution layer 503, a fourth convolution layer 504, a fifth convolution layer 505, and a sixth convolution layer 506. A pooling layer 511 and a batch normalization layer 521 are arranged between the first convolution layer 501 and the second convolution layer 502, a second pooling layer 512 and a second batch normalization layer 522 are arranged between the second convolution layer 502 and the third convolution layer 503, a pooling layer 513 and a batch normalization layer 523 are arranged between the third convolution layer 503 and the fourth convolution layer 504, a pooling layer 514 and a batch normalization layer 524 are arranged between the fourth convolution layer 504 and the fifth convolution layer 505, and a pooling layer 514 and a batch normalization layer 524 are arranged between the fifth convolution layer 505 and the sixth convolution layer 506. A pooling layer 515 and a batch normalization layer 525, a pooling layer 516 is arranged between the sixth convolutional layer 506 and the first long short-term memory network layer (LSTM_1 layer) 507 of the first temporal learning unit 51, the first long short-term memory network layer (LSTM_1 layer) 507, the second long short-term memory network layer (LSTM_2 layer) 508, and the first fully connected layer 530, the random dropout layer (Dropout layer) 531, the second fully connected layer 532 and the activation function layer 533 of the first classification unit 52 are arranged in sequence.

[0054] The second feature extraction unit 60 includes a first convolution layer 601, a second convolution layer 602, a third convolution layer 603, a fourth convolution layer 604, a fifth convolution layer 605, and a sixth convolution layer 606. A first pooling layer 611 and a first batch normalization layer 621 are arranged between the first convolution layer 601 and the second convolution layer 602, a second pooling layer 612 and a second batch normalization layer 622 are arranged between the second convolution layer 602 and the third convolution layer 603, a third pooling layer 613 and a third batch normalization layer 623 are arranged between the third convolution layer 603 and the fourth convolution layer 604, a fourth pooling layer 614 and a fourth batch normalization layer 624 are arranged between the fourth convolution layer 604 and the fifth convolution layer 605, and a fifth pooling layer 615 and a sixth convolution layer 616 are arranged between the fifth convolution layer 605 and the sixth convolution layer 606. 606, a fifth pooling layer 615 and a fifth batch normalization layer 625 are arranged between the sixth convolutional layer 606 and the first long short-term memory network (LSTM_1 layer) 607 of the second temporal learning unit 61, a sixth pooling layer 616 is arranged between the first long short-term memory network 607 (LSTM_1 layer), the second long short-term memory network (LSTM_2 layer) 608, and the first fully connected layer 630, the Dropout layer 631, the second fully connected layer 632 and the activation function layer 633 of the first classification unit 62 are arranged in sequence.

[0055] In the embodiment of the present application, the convolution layer in the feature extraction unit (i.e., the first feature extraction unit and the second feature extraction unit) convolves the input sample EEG signal or sample ECG signal by moving the convolution kernel to realize the feature extraction of the sample EEG signal or sample ECG signal. Since too many convolution layers may cause the gradient explosion and gradient disappearance of the model, in the embodiment of the present application, after the experimental comparison test as shown in Table 1, 6 convolution layers are used to construct the initial ECG recognition model and the initial EEG recognition model, and the average of 10 evaluation indicators is used as the final result for each layer test. In the setting of the convolution kernel, the convolution kernel size of the first convolution layer is set to 1×64, and the number of convolution kernels is also set to 64, so as to extract the features of the input data (for example, sample EEG signal or sample ECG signal) to the greatest extent; the size of the convolution kernels of the second to sixth convolution layers are all set to 1×5, and the number is set to 64. During the model training process, useful information in the input data is obtained by continuously updating the convolution kernel weights.

[0056] Table 1:

[0057]

[0058] In an embodiment of the present application, the pooling layer in the feature extraction unit is used to reduce the size of the feature map obtained from the previous convolution layer while retaining the basic features. The present application chooses to use the maximum pooling to retain only the maximum value in the set pooling window of 1×2 size to reduce the number of parameters and computational overload. The present application adds the maximum pooling between consecutive convolutional layers to reduce the number of parameters and computational overload, thereby controlling the overfitting of the model. In an embodiment of the present application, after each convolution kernel pooling operation is completed, the batch normalization layer will normalize the input features to reduce changes in data distribution.

[0059] Since EEG signals and ECG signals are complex, continuous, nonlinear, and have long signals, in order to extract more complete features, in addition to considering the features of the current moment, it is also necessary to consider the relationship between the sample EEG signals / sample ECG signals at the previous moment and the sample EEG signals / sample ECG signals at the current moment.

[0060] In the embodiment of the present application, in the initial ECG recognition model and the initial EEG recognition model, the 18th and 19th layers are set as long short-term memory network layers (LSTM layers), which include 64 units and 32 units respectively, to continue to find useful information from the previous feature map and learn the long-term dependencies in the signal. In the embodiment of the present application, the superposition of multiple long short-term memory network layers enables the model to learn more complex information in the time series signal, thereby improving the accuracy of depression prediction.

[0061] In the present application embodiment, Figure 5 As shown, in the initial ECG recognition model and the initial EEG recognition model, the 20th and 22nd layers are fully connected layers, which are used to perform actual classification according to the output of the long short-term memory network layer.

[0062] In order to prevent overfitting, in one embodiment of the present application, a Dropout layer is added as the 21st layer between the two fully connected layers of the initial ECG recognition model and the initial EEG recognition model.

[0063] In one embodiment of the present application, after the second fully connected layer, activation functions such as Softmax can be used to enable the neural network to perform a binary classification function, so that for a specific sample EEG signal or sample ECG signal, the initial ECG recognition model and the initial EEG recognition model can determine with a certain probability whether the subject corresponding to the sample EEG signal or the sample ECG signal suffers from depression.

[0064] In an embodiment of the present application, the model training unit is used to train the above-mentioned initial ECG recognition model and the initial EEG recognition model to obtain the target ECG recognition model and the target EEG recognition model, specifically based on the following steps A01 to A04.

[0065] Step A01, obtaining multiple groups of sample data and depression labels corresponding to each group of sample data.

[0066] In an embodiment of the present application, the aforementioned signal acquisition module can be used to collect the ECG signals and EEG signals of the subject to obtain sample ECG signals and sample EEG signals. The present application does not limit the method of acquiring the sample ECG signals and sample EEG signals and the method of determining the depression label.

[0067] Among them, the depression label may include two types of labels, namely, labels with depression and labels without depression, and may also include multiple types of labels such as mild depression, moderate depression, severe depression, and labels without depression, and this application does not limit this. For the convenience of description, this application takes the depression label including two types of labels, namely, labels with depression and labels without depression, as an example for illustration.

[0068] Step A02: preprocess the sample ECG signals and sample EEG signals in each group of sample data to obtain preprocessed sample ECG signals and sample EEG signals.

[0069] In an embodiment of the present application, in the process of preprocessing the sample ECG signals and sample EEG signals in each group of sample data respectively, the sample ECG signals and sample EEG signals can be preprocessed based on the aforementioned preprocessing module to obtain preprocessed sample EEG signals and preprocessed sample ECG signals.

[0070] In order to improve the generalization ability of the deep learning model, in one embodiment of the present application, after the model training unit preprocesses the sample ECG signals and sample EEG signals in each group of sample data respectively, it can also use data enhancement technology to perform data enhancement processing on the sample ECG signals and sample EEG signals in each group of sample data after preprocessing, so as to obtain multiple groups of sample data after data enhancement processing.

[0071] For example, each group of sample EEG signals and sample ECG signals is segmented without overlap using a 5-s time window, and the segmented sample EEG signals and sample ECG signals are labeled and used as input to the model.

[0072] Specifically, the sample EEG signal and sample ECG signal of each sample object can be divided into a sequence of segments {X1, X2, X3...X T}, where X t represents a certain segment sequence, and T represents the number of segments after the sample EEG signal or sample ECG signal of a sample object is segmented. The sub-signals segmented from the same sample EEG signal or sample ECG signal have the same label. t The label is denoted by L t, then the input data segment {X1,X2,X3......X T} has a corresponding label set {L1,L2,L3......L T}, thereby completing the segmentation of the entire sample EEG signal and sample ECG signal of a sample object and setting the label.

[0073] In one embodiment of the present application, when the data amount of sample EEG signals and sample ECG signals labeled as suffering from depression is less than the data amount of sample EEG signals and sample ECG signals labeled as not suffering from depression, and the ratio of the data amount of sample EEG signals and sample ECG signals labeled as not suffering from depression to the data amount of sample EEG signals and sample ECG signals labeled as suffering from depression is greater than a preset threshold, the synthetic minority over-sampling technique (Synthetic Minority Over-sampling Technique, SMOTE) can be used to generate multiple groups of new sample EEG signals and sample ECG signals labeled as suffering from depression to increase the number of minority class samples, thereby balancing the data set.

[0074] Specifically, we can randomly select minority class samples (i.e., sample EEG signals and sample ECG signals labeled as those suffering from depression) and find their K nearest neighbor samples. By calculating the differences between the samples, we can generate new synthetic samples to make the data set more balanced.

[0075] For example, assume that the sample data obtained includes 30 groups of data, of which 22 groups are data without depression and 8 groups are data with depression. Since the data with depression is relatively small, the synthetic minority oversampling technique can be used to generate more groups of sample EEG signals and sample ECG signals labeled as those with depression to balance the data set.

[0076] In practical applications, using the SMOTE algorithm to generate samples may increase the risk of model overfitting, so only 7 groups of depression sample data can be selected from the 8 groups of depression patient data, and 7 groups of new sample data labeled as suffering from depression are generated through the SMOTE algorithm, thereby obtaining 37 groups of sample data, and 14 groups of sample data are labeled as suffering from depression, so that the data volume of sample data of different categories in the sample data set is more balanced and the risk of model overfitting can be reduced.

[0077] Step A03, dividing the preprocessed sample ECG signals and the preprocessed sample EEG signals into a training set and a validation set.

[0078] In this embodiment, the training set is used for model training, and the test set is used to evaluate the generalization ability of the model and verify the model performance.

[0079] Taking the above 37 groups of sample data as an example, the 37 groups of sample data can be divided into training set and test set in a ratio of 29:8. This data division process can be repeated 10 times to generate 10 different folds, so that each group of sample data has the opportunity to appear in the test set. In addition, in order to avoid the contingency of a single experimental result, the average of the 10 experimental results can be taken to evaluate the overall performance of the model. For example, the performance of the model is comprehensively evaluated using multiple evaluation indicators such as accuracy, precision, recall, and F1 score to ensure the effectiveness and reliability of the model.

[0080] Step A04, using the training set to train the initial ECG recognition model and the initial EEG recognition model, and using the verification set to verify the trained initial ECG recognition model and the initial EEG recognition model, to obtain the target ECG recognition model and the target EEG recognition model.

[0081] In one embodiment of the present application, the cross entropy loss function can be used to calculate the loss function values ​​of the initial ECG recognition model and the initial EEG recognition model, and the parameters of the initial ECG recognition model and the initial EEG recognition model can be adjusted based on the loss function value using the Adam optimizer. The training set is reused to train the initial ECG recognition model and the initial EEG recognition model with adjusted parameters. After completing one round of training, the trained initial ECG recognition model and the initial EEG recognition model are verified using the validation set, and the loss function value of the validation set is calculated until the rate of change of the loss function value of the validation set is less than the preset rate of change threshold or the number of training times reaches the maximum number of training times. The training of the initial ECG recognition model and the initial EEG recognition model is terminated, and the target ECG recognition model and the target EEG recognition model are obtained.

[0082] After experiments, the initial ECG recognition model and the initial EEG recognition model proposed in the embodiment of the present application have an accuracy rate of 98.15%, a precision rate of 96.15%, a recall rate of 100%, and an F1 score of 0.98 after being trained using sample data, which have good discrimination and classification effects.

[0083] It should be noted that the above is merely an example of the training method of the initial ECG recognition model and the initial EEG recognition model, and is intended to limit the scope of protection of the present application. It can be understood that in the embodiments of the present application, other methods can also be used to train the initial ECG recognition model and the initial EEG recognition model to obtain the target ECG recognition model and the target EEG recognition model, and the present application does not impose any restrictions on this.

[0084] In one embodiment of the present application, after obtaining the target ECG recognition model and the target EEG recognition model, the model training unit is also used to train the support vector machine model to be trained to obtain the target support vector machine model. So that after the target ECG recognition model, the target EEG recognition model and the target support vector machine model are obtained through training, the prediction matrix constructed based on the first depression recognition result and the second depression recognition result of the object to be tested output by the target ECG recognition model and the target EEG recognition model can be used to predict whether the object to be tested suffers from depression, and the final depression target recognition result can be output.

[0085] Specifically, the model training unit is used to input each group of sample data into the target EEG recognition model and the target ECG recognition model after obtaining the target ECG recognition model and the target EEG recognition model, to obtain a first depression prediction result corresponding to each group of sample data output by the target EEG recognition model, and a second depression prediction result corresponding to each group of sample data output by the target ECG recognition model; based on the first depression prediction result corresponding to each group of sample data, the second depression prediction result corresponding to each group of sample data, and the depression label corresponding to each group of sample data, the support vector machine model to be trained is trained to obtain a target support vector machine model.

[0086] That is, the first depression prediction result corresponding to each group of sample data and the second depression prediction result corresponding to each group of sample data are used as training samples, and the depression label corresponding to each group of sample data is used as the label of the training sample. The support vector machine model to be trained is trained to obtain the target support vector machine model.

[0087] In the embodiment of the present application, the training principle of the support vector machine model (SVM) is to find a hyperplane so that data points of different categories can be separated to the greatest extent.

[0088] In one embodiment of the present application, the training process of the support vector machine model can be converted into a convex optimization problem, and the optimal hyperplane is found by solving the Lagrange multiplier. The training process may include: dividing the first depression prediction result corresponding to each group of sample data output by the target EEG recognition model and the second depression prediction result corresponding to each group of sample data output by the target ECG recognition model into a training set and a test set, finding the hyperplane that maximizes the interval of the training set by solving the convex quadratic programming problem or the Lagrange multiplier method, using the test set to evaluate the performance of the support vector machine model, and adjusting the parameters of the support vector machine model to optimize the model performance, so that the target support vector machine model can find the optimal hyperplane, so that the sample data labeled as suffering from depression and the sample data labeled as not suffering from depression can be separated to the greatest extent, thereby achieving efficient classification.

[0089] In an embodiment of the present application, a target ECG recognition model, a target EEG recognition model and a target support vector machine model are obtained by training an initial ECG recognition model, an initial EEG recognition model and a support vector machine model to be trained, and when it is necessary to identify whether the subject to be tested suffers from depression, a first depression recognition result and a second depression recognition result are obtained by inputting the ECG signal and the EEG signal of the subject to be tested into the pre-trained target ECG recognition model and the target EEG recognition model respectively, and a prediction matrix is ​​constructed based on the first depression recognition result and the second depression recognition result, and the prediction matrix is ​​input into the pre-trained target support vector machine model to obtain a target depression recognition result, thereby realizing the use of the ECG signal and the EEG signal of the subject to be tested, and identifying whether the subject to be tested suffers from depression based on a dual-modal decision fusion algorithm of the three models of the target ECG recognition model, the target EEG recognition model and the target support vector machine model.

[0090] This application improves the accuracy of depression recognition by combining the recognition results of the target ECG recognition model and the target EEG recognition model, and uses the target support vector machine model for final recognition. It can effectively identify mild depression and solve the technical problem of low depression recognition accuracy in existing solutions. It can effectively reduce the occurrence of misdiagnosis and missed diagnosis, and provides an innovative and practical solution for the early screening and diagnosis of depression.

[0091] The embodiment of the present application also provides an electronic device, which may include the depression identification device of any one of the above embodiments.

[0092] In one embodiment of the present application, the electronic device may be a computer, an embedded processor or other intelligent device. When the depression recognition device is a virtual device implemented by a computer program, such as Figure 6 As shown, the electronic device may include a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and executable on the processor 70. When the processor 70 executes the computer program 72, the functions of each module in the depression identification device are realized.

[0093] The processor 70 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0094] The memory 71 may be an internal storage unit of the electronic device, such as a hard disk or a memory. The memory 71 may also be an external storage device for the electronic device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 71 may also include both an internal storage unit of the electronic device and an external storage device. The memory 71 is used to store the above-mentioned computer program and other programs and data required by the electronic device.

[0095] The computer program can be divided into a signal acquisition module, a first identification module, a second identification module and a third identification module, wherein the signal acquisition module, the first identification module, the second identification module and the third identification module are stored in the memory 71 and executed by the processor 70 to complete the present application. The first identification module, the second identification module and the third identification module are a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the function of the computer program executing the depression identification device in the electronic device.

[0096] Specifically, the signal acquisition module is used to acquire the ECG signal and EEG signal of the subject to be tested; the first recognition module is used to input the ECG signal into a pre-trained target ECG recognition model to obtain a first depression recognition result output by the target ECG recognition model; the second recognition module is used to input the EEG signal into a pre-trained target EEG recognition model to obtain a second depression recognition result output by the target EEG recognition model; the third recognition module is used to construct a prediction matrix based on the first depression recognition result and the second depression recognition result, and input the prediction matrix as an input feature into a pre-trained target support vector machine model to obtain a depression target recognition result output by the support vector machine model.

[0097] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0098] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0099] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0100] In the embodiments provided in the present application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the electronic device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, systems or units, which can be electrical, mechanical or other forms.

[0101] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0102] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0103] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or recording medium that can carry computer program code, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0104] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A depression identification device, characterized in that: The depression identification device comprises: A signal acquisition module, used to acquire the electrocardiogram (ECG) and electroencephalogram (EEG) signals of the subject to be tested; A first recognition module, configured to input the electrocardiogram signal into a pre-trained target electrocardiogram recognition model to obtain a first depression recognition result output by the target electrocardiogram recognition model; A second recognition module is used to input the EEG signal into a pre-trained target EEG recognition model to obtain a second depression recognition result output by the target EEG recognition model; The third recognition module is used to construct a prediction matrix based on the first depression recognition result and the second depression recognition result, and input the prediction matrix as an input feature into a pre-trained target support vector machine model to obtain a depression target recognition result output by the target support vector machine model.

2. The depression identification device according to claim 1, characterized in that: The depression identification device further comprises a preprocessing module, which comprises a filtering unit, a wavelet decomposition and reconstruction unit, and an artifact checking unit; The filtering unit is used to perform bandpass filtering and notch filtering on the electrocardiogram signal and the electroencephalogram signal respectively; The wavelet decomposition and reconstruction unit is used to perform wavelet decomposition on the ECG signal and the EEG signal after bandpass filtering and notch filtering, and perform reconstruction processing after removing artifacts; The artifact inspection unit is used to load the EEG signal after wavelet decomposition and reconstruction using the first visualization tool, perform notch filtering on the EEG signal after wavelet decomposition and reconstruction, and receive a user artifact removal operation to obtain a preprocessed EEG signal; The artifact checking unit is further used to use a second visualization tool to load the ECG signal after wavelet decomposition and reconstruction processing, and receive a high-frequency noise signal removal operation of a user to obtain a preprocessed ECG signal.

3. The depression identification device according to claim 1, characterized in that: The depression identification device also includes a model training unit for training an initial ECG recognition model, an initial EEG recognition model and a support vector machine model to be trained to obtain the target ECG recognition model, the target EEG recognition model and the target support vector machine model.

4. The depression identification device according to claim 3, characterized in that: The initial electrocardiogram recognition model has the same model structure as the initial electroencephalogram recognition model, including a feature extraction unit, a time series learning unit and a classification unit; The feature extraction unit is used to extract local features of the sample electrocardiogram signal / sample electroencephalogram signal; The time series learning unit is used to learn the relationship between the sample ECG signal / sample EEG signal at the previous moment and the current moment, and output feature information of the sample ECG signal / sample EEG signal based on the local features and the learned relationship; The classification unit is used to perform classification based on the feature information output by the time series learning unit to obtain a corresponding depression prediction result.

5. The depression identification device according to claim 4, characterized in that: The feature extraction unit includes six convolutional layers, a pooling layer and a batch normalization layer are arranged between two adjacent convolutional layers in the six convolutional layers, and a pooling layer is arranged between the last convolutional layer and the temporal learning unit; The temporal learning unit includes two long short-term memory network layers; The classification unit includes two fully connected layers and an activation function layer arranged in sequence, and a random dropout layer is arranged between the two fully connected layers.

6. The depression identification device according to claim 3, characterized in that: When training the initial electrocardiogram recognition model and the initial electroencephalogram recognition model, the model training unit is specifically used to: Obtain multiple groups of sample data and depression labels corresponding to each group of sample data; wherein each group of sample data includes a sample electrocardiogram signal and a sample electroencephalogram signal; Preprocessing the sample electrocardiogram signal and the sample electroencephalogram signal in each group of sample data respectively to obtain the preprocessed sample electrocardiogram signal and the sample electroencephalogram signal; Dividing the preprocessed sample electrocardiogram signals and the preprocessed sample electroencephalogram signals into a training set and a validation set; The initial ECG recognition model and the initial EEG recognition model are trained using the training set, and the trained initial ECG recognition model and the initial EEG recognition model are verified using the verification set to obtain the target ECG recognition model and the target EEG recognition model.

7. The depression identification device according to claim 6, characterized in that: The model training unit is used to train the initial ECG recognition model and the initial EEG recognition model using the training set, and to verify the trained initial ECG recognition model and the initial EEG recognition model using the verification set. The cross entropy loss function is used to calculate the loss function values ​​of the initial ECG recognition model and the initial EEG recognition model, and the parameters of the initial ECG recognition model and the initial EEG recognition model are adjusted using the Adam optimizer based on the loss function value. The training set is used again to train the initial ECG recognition model and the initial EEG recognition model after the parameter adjustment. After completing one round of training, the trained initial ECG recognition model and the initial EEG recognition model are verified using the validation set, and the loss function value of the validation set is calculated. When the rate of change of the loss function value of the validation set is less than the preset rate of change threshold or the number of training times reaches the maximum number of training times, the training of the initial ECG recognition model and the initial EEG recognition model is terminated, and the target ECG recognition model and the target EEG recognition model are obtained.

8. The depression identification device according to claim 6, characterized in that: After preprocessing the sample ECG signals and sample EEG signals in each group of sample data respectively, the model training unit is further used to: perform data enhancement processing on the preprocessed groups of sample data to obtain the data enhanced groups of sample data.

9. The depression identification device according to any one of claims 6 to 8, characterized in that: When training the support vector machine model to be trained, the model training unit is specifically used to: After obtaining the target ECG recognition model and the target EEG recognition model, inputting each group of sample data into the target EEG recognition model and the target ECG recognition model to obtain a first depression prediction result corresponding to each group of sample data output by the target EEG recognition model, and a second depression prediction result corresponding to each group of sample data output by the target ECG recognition model; The support vector machine model to be trained is trained based on the first depression prediction result corresponding to each group of sample data, the second depression prediction result corresponding to each group of sample data, and the depression label corresponding to each group of sample data to obtain the target support vector machine model.

10. An electronic device, characterized in that: The electronic device is provided with a depression identification device as described in any one of claims 1-9.

Images