阅读说明：本技术 一种基于残差卷积神经网络的抑郁倾向评估系统 (Depression tendency evaluation system based on residual convolutional neural network ) 是由 李岱 郑芮 柏德祥 于 2019-11-20 设计创作，主要内容包括：本申请公开了一种基于残差卷积神经网络的抑郁倾向评估系统,包括采集单元、预处理单元、提取单元和评估单元；抑郁评估实验中,采集单元采集受试者的多导联脑电信号；预处理单元预处理脑电信号；提取单元基于残差卷积神经网络,提取预处理后脑电信号的特征；评估单元将提取的各个导联的脑电信号特征通过回归模型训练和验证,计算出受试者的抑郁评估分数,进而对其抑郁倾向进行评估。本申请通过卷积残差神经网络直接从原始脑电信号中提取特征,增大了卷积网络深度,最大程度的提取了脑电信号的特征。本申请对各个导联的脑电信息进行回归模型训练,使得模型可以充分考虑大脑各个区域的相关性,从而使得模型可以同时处理不同大脑区域的脑电信号。(The application discloses a depression tendency evaluation system based on a residual convolutional neural network, which comprises a collecting unit, a preprocessing unit, an extracting unit and an evaluating unit; in a depression evaluation experiment, a collecting unit collects multi-lead electroencephalogram signals of a subject; the preprocessing unit preprocesses the electroencephalogram signals; the extraction unit extracts the characteristics of the preprocessed electroencephalogram signals based on the residual convolutional neural network; the evaluation unit trains and verifies the extracted EEG signal characteristics of each lead through a regression model, calculates the depression evaluation score of the subject and further evaluates the depression tendency of the subject. The method and the device have the advantages that the convolution residual neural network is used for directly extracting the features from the original electroencephalogram signals, so that the depth of the convolution network is increased, and the features of the electroencephalogram signals are extracted to the greatest extent. The method and the device have the advantages that regression model training is carried out on the electroencephalogram information of each lead, so that the relevance of each brain region can be fully considered by the model, and therefore the electroencephalogram signals of different brain regions can be processed by the model at the same time.)

1. A residual convolutional neural network-based depression tendency assessment system, characterized in that:

the evaluation system comprises a collection unit, a preprocessing unit, an extraction unit and an evaluation unit;

the acquisition unit is used for acquiring multi-lead electroencephalogram signals of a subject in a depression assessment experiment;

the preprocessing unit is used for preprocessing the electroencephalogram signals;

the extraction unit is used for extracting the characteristics of the preprocessed electroencephalogram signals based on the residual convolutional neural network;

the evaluation unit is used for training and verifying the electroencephalogram signal characteristics of each lead extracted by the extraction unit through a regression model, calculating the depression evaluation score of the subject and further evaluating the depression tendency of the subject.

2. The residual convolutional neural network-based depression tendency assessment system according to claim 1, wherein:

the operation process of the evaluation system is as follows:

in a depression evaluation experiment, a collecting unit collects multi-lead electroencephalogram signals of a subject;

the preprocessing unit preprocesses the electroencephalogram signals;

the extraction unit extracts the characteristics of the preprocessed electroencephalogram signals based on the residual convolutional neural network;

the evaluation unit trains and verifies the extracted EEG signal characteristics of each lead through a regression model, calculates the depression evaluation score of the subject and further evaluates the depression tendency of the subject.

3. The residual convolutional neural network-based depression tendency assessment system according to claim 1 or 2, wherein:

in the depression assessment experiment, electroencephalogram signals are collected through portable multi-lead electroencephalogram equipment, a sensor is a contact type wet electrode, before the experiment begins, harmless conductive paste is squeezed into the wet electrode to enhance the contact between the electrode and the scalp, and the collected electroencephalogram signals are uploaded to a cloud processing platform in real time.

4. The residual convolutional neural network-based depression tendency assessment system according to claim 1 or 2, wherein:

the depression assessment experiment comprises emotional stimulation and a cognition experiment, which are both subjected to standardized assessment in advance, so that the arousal degree of the experimental stimulation faced by the subject is equivalent, and the cognition difficulty is equivalent;

the emotional stimulation experiment comprises static emotional stimulation and dynamic emotional stimulation;

the static emotional stimuli are composed of pictures, wherein one half of the pictures are positive pictures and the other half of the pictures are negative pictures;

the dynamic emotional stimulation is composed of short pieces, and the duration of each short piece is 20-100 seconds;

the cognitive experiment is as follows: different digit sequences are flashed and when the desired digit sequence occurs, the subject presses a button to examine the subject's short-term memory.

5. The residual convolutional neural network-based depression tendency assessment system according to claim 1 or 2, wherein:

in the preprocessing unit, through band-pass filtering and notch processing, electroencephalogram signals except voltage data of 1-70Hz are filtered, and meanwhile fundamental frequency interference of 50Hz is filtered;

the influence of noise in the signal is reduced by first-order smoothing processing.

6. The residual convolutional neural network-based depression tendency assessment system according to claim 1 or 2, wherein:

the operation process of the extraction unit is as follows:

extracting the electroencephalogram signal characteristics of each lead through one-dimensional convolution;

and constructing and training a residual convolutional neural network and carrying out deep extraction on the extracted electroencephalogram signal characteristics.

7. The residual convolutional neural network-based depression tendency assessment system according to claim 6, wherein:

the building and training of the residual convolutional neural network specifically comprises the following steps:

the output of the neural network of each layer is input to a deeper layer of the neural network backwards;

a number of residual blocks are stacked, and the residual blocks are used to train deeper networks, thereby forming a residual convolutional neural network.

8. The residual convolutional neural network-based depression tendency assessment system according to claim 7, wherein:

the residual block comprises two convolution layers, and the connection of the residual block is realized by the following formula:

wherein x is the value output to the neuron by the previous layer; w is the weight by which x passes to the neuron; y is the output value of x within a neuron as determined by the activation function.

9. The residual convolutional neural network-based depression tendency assessment system according to claim 8, wherein:

the two convolutional layers each have an activation function behind it.

10. The residual convolutional neural network-based depression tendency assessment system according to claim 1 or 2, wherein:

the operation process of the evaluation unit is as follows:

extracting the neuron weight of the brain electrical signal characteristic of each lead in the penultimate layer of the residual convolutional neural network;

inputting the extracted neuron weights into a regression model, and obtaining depression evaluation weights of the EEG signal characteristics of each lead through training and verifying the regression model;

linearly adding the products of the depression evaluation weights and the electroencephalogram signal characteristic values of all leads to obtain depression evaluation scores of the subjects;

regression model parameters are returned and a report of the subject's depression tendency is generated.

Technical Field

The invention belongs to the technical field of depression assessment, and relates to a depression tendency assessment system based on a residual convolutional neural network.

Background

Major depressive disorder (Major Depression) is a typical disease in depressive disorders. It characteristically manifests as a clear at least 2-week onset involving significant changes in emotional, cognitive, and autonomic neural functions. Studies have shown that major depressive disorder has a prevalence of about 7% at 12 months, which is one of the most common psychiatric disorders.

Major depressive disorder has long been a hot problem of concern in the field of mental hygiene, and a great deal of research has been conducted around the cause, treatment and prognosis of major depressive disorder. In these studies, early screening for depression predisposition is considered to be of great importance for the prevention and treatment of disease. However, early screening for depression is often difficult due to major depression often co-morbid with other physiological and psychological illnesses. Generally, the diagnosis of major depressive disorder is based on the description of the criteria for the disease on diagnostic and statistical manuals of mental disorders. However, it is difficult to completely avoid subjective influences due to human interference during the diagnosis process. Such as possible omissions in the collection of medical histories or an unintentional exaggeration of the severity of certain symptoms.

In recent years, with the maturity of machine learning algorithms, more and more researchers are trying to use physiological and behavioral data to early screen for major depressive disorder to reduce the influence of subjectivity and provide assistance for diagnosis of psychiatrists. Machine learning is a method for automatically mining deeper information of complex data by using a computer algorithm, and has been widely applied to various fields such as image recognition, voice recognition, disease diagnosis and the like.

Electroencephalogram data (EEG) contains rich physiological information of the brain, but brain telecommunication is difficult to be effectively processed and analyzed due to low signal-to-noise ratio, large data volume and the like.

Disclosure of Invention

In order to overcome the defects in the prior art, the application provides a depression tendency evaluation system based on a residual convolutional neural network, which is used for automatically extracting, analyzing and effectively training features of electroencephalograms based on the residual convolutional neural network, so that information capable of effectively reflecting brain features of a subject is mined, and therefore the depression degree of the subject is divided in a depression evaluation experiment.

In order to achieve the above objective, the following technical solutions are adopted in the present application:

a depression tendency evaluation system based on a residual convolutional neural network comprises a collecting unit, a preprocessing unit, an extracting unit and an evaluating unit;

the acquisition unit is used for acquiring multi-lead electroencephalogram signals of a subject in a depression assessment experiment;

the preprocessing unit is used for preprocessing the electroencephalogram signals;

the extraction unit is used for extracting the characteristics of the preprocessed electroencephalogram signals based on the residual convolutional neural network;

the evaluation unit is used for training and verifying the electroencephalogram signal characteristics of each lead extracted by the extraction unit through a regression model, calculating the depression evaluation score of the subject and further evaluating the depression tendency of the subject.

The invention further comprises the following scheme:

preferably, the operation process of the evaluation system is as follows:

in a depression evaluation experiment, a collecting unit collects multi-lead electroencephalogram signals of a subject;

the preprocessing unit preprocesses the electroencephalogram signals;

the extraction unit extracts the characteristics of the preprocessed electroencephalogram signals based on the residual convolutional neural network;

the evaluation unit trains and verifies the extracted EEG signal characteristics of each lead through a regression model, calculates the depression evaluation score of the subject and further evaluates the depression tendency of the subject.

Preferably, in the depression assessment experiment, the portable multi-lead electroencephalogram equipment is used for collecting electroencephalogram signals, the sensor is a contact type wet electrode, before the experiment begins, harmless conductive paste is injected into the wet electrode to enhance the contact between the electrode and the scalp, and the collected electroencephalogram signals are uploaded to the cloud processing platform in real time.

Preferably, the depression assessment experiment comprises emotional stimulation and cognition experiment, which are all standardized in advance to ensure that the arousal degree of the experimental stimulation and the cognition difficulty of the subject are equivalent;

the emotional stimulation experiment comprises static emotional stimulation and dynamic emotional stimulation;

the static emotional stimuli are composed of pictures, wherein one half of the pictures are positive pictures and the other half of the pictures are negative pictures;

the dynamic emotional stimulation is composed of short pieces, and the duration of each short piece is 20-100 seconds;

the cognitive experiment is as follows: different digit sequences are flashed and when the desired digit sequence occurs, the subject presses a button to examine the subject's short-term memory.

Preferably, in the preprocessing unit, through band-pass filtering and notch processing, electroencephalogram signals except for voltage data of 1-70Hz are filtered, and meanwhile, fundamental frequency interference of 50Hz is filtered;

the influence of noise in the signal is reduced by first-order smoothing processing.

Preferably, the operation process of the extraction unit is as follows:

extracting the electroencephalogram signal characteristics of each lead through one-dimensional convolution;

and constructing and training a residual convolutional neural network and carrying out deep extraction on the extracted electroencephalogram signal characteristics.

Preferably, the constructing and training of the residual convolutional neural network specifically includes:

the output of the neural network of each layer is input to a deeper layer of the neural network backwards;

a number of residual blocks are stacked, and the residual blocks are used to train deeper networks, thereby forming a residual convolutional neural network.

Preferably, the residual block includes two convolutional layers, and the connection of the residual block is implemented by the following formula:

wherein x is the value output to the neuron by the previous layer; w is the weight by which x passes to the neuron; y is the output value of x within a neuron as determined by the activation function.

Preferably, there are two convolutional layers, each followed by an activation function.

Preferably, the operation process of the evaluation unit is as follows:

extracting the neuron weight of the brain electrical signal characteristic of each lead in the penultimate layer of the residual convolutional neural network;

inputting the extracted neuron weights into a regression model, and obtaining depression evaluation weights of the EEG signal characteristics of each lead through training and verifying the regression model;

linearly adding the products of the depression evaluation weights and the electroencephalogram signal characteristic values of all leads to obtain depression evaluation scores of the subjects;

regression model parameters are returned and a report of the subject's depression tendency is generated.

The beneficial effect that this application reached:

(1) the method and the device have the advantages that the convolution residual neural network is used for directly extracting the features from the original electroencephalogram signals, so that the depth of the convolution network is increased, and the features of the electroencephalogram signals are extracted to the greatest extent.

(2) The brain electrical information of each lead is subjected to regression training, so that the correlation of each brain region can be fully considered by the model, and the brain electrical signals of different brain regions can be processed by the model simultaneously.

Drawings

FIG. 1 is a block diagram of a depression tendency assessment system based on a residual convolutional neural network according to the present application;

fig. 2 is a flowchart illustrating the operation of a residual convolutional neural network-based depression tendency estimation system according to the present application.

Detailed Description

The present application is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present application is not limited thereby.

As shown in fig. 1, the present application provides a depression tendency evaluation system based on a residual convolutional neural network, which includes an acquisition unit, a preprocessing unit, an extraction unit, and an evaluation unit;

the acquisition unit is used for acquiring multi-lead electroencephalogram signals of a subject in a depression assessment experiment;

the preprocessing unit is used for preprocessing the electroencephalogram signals;

the extraction unit is used for extracting the characteristics of the preprocessed electroencephalogram signals based on the residual convolutional neural network;

the evaluation unit is used for calculating depression evaluation scores of the subjects by performing logistic regression training and verification on the electroencephalogram signal characteristics of each lead extracted by the extraction unit, and further evaluating the depression tendency of the subjects.

As shown in fig. 2, the operation process of the evaluation system is as follows:

in a depression evaluation experiment, a collecting unit collects multi-lead (multichannel) electroencephalogram signals of a subject;