阅读说明：本技术 一种基于自监督学习的脑电波图异常检测的方法及装置 (Electroencephalogram abnormity detection method and device based on self-supervision learning ) 是由 王瑞轩 许俊杰 郑瑶嘉 于 2020-12-12 设计创作，主要内容包括：本发明涉及图像处理技术领域,公开了一种基于自监督学习的脑电波图异常检测的方法及装置,方法包括：根据脑电图(EEG)信号,生成自标签数据集；所述自标签数据集包括经过变换后的EEG数据及其变换对应的标签；根据所述自标签数据集,训练分类器；所述分类器由四个卷积模块和一个全连接层组成；根据所述分类器输出的k维向量V,对EEG数据异常值进行评分；k表示变换集的基数|T|。本发明不需要对EEG数据进行人工特征提取和过于复杂的预处理就能够对EEG数据进行异常检测,并达到较好的检测效果。(The invention relates to the technical field of image processing, and discloses a electroencephalogram abnormity detection method and device based on self-supervision learning, wherein the method comprises the following steps: generating a self-label dataset from an electroencephalography (EEG) signal; the self-label dataset comprises transformed EEG data and labels corresponding to the transformations; training a classifier according to the self-label data set; the classifier consists of four convolution modules and a full connection layer; according to the k-dimensional vector V output by the classifier, grading an EEG data abnormal value; k represents the radix | T | of the transform set. The method can detect the abnormity of the EEG data without artificial feature extraction and excessively complex preprocessing of the EEG data, and achieves better detection effect.)

1. A method for electroencephalogram anomaly detection based on self-supervised learning, comprising:

generating a self-label dataset from an electroencephalography (EEG) signal; the self-label dataset comprises transformed EEG data and labels corresponding to the transformations;

training a classifier according to the self-label data set; the classifier consists of four convolution modules and a full connection layer;

according to the k-dimensional vector V output by the classifier, grading an EEG data abnormal value; k represents the radix | T | of the transform set.

2. The method for electroencephalography abnormality detection based on self-supervised learning as recited in claim 1, wherein the generating a self-labeled data set from EEG raw data specifically comprises:

linear interpolation of different scales is carried out on EEG original data;

the data is cut from the interpolated data to the same dimensions as the original EEG data.

3. The method for electroencephalography abnormality detection based on self-supervised learning according to claim 1, wherein the self-label dataset is obtained by using the following formula:

S_T＝{T_i(x)，i)：x∈S，Ti∈T}，

wherein S is_TT {.. } is a set of transforms performed on EEG raw data, x is EEG data, from a labeled dataset.

4. The method for electroencephalography abnormality detection based on self-supervised learning as recited in claim 1, wherein the training of the classifier according to the self-labeled data set comprises:

using self-label data set as a multi-class classification model f_θAnd setting the cross entropy as a loss function to train a multi-class classifier.

5. The method for electroencephalogram abnormality detection based on self-supervised learning as recited in claim 1, wherein the k-dimensional vector V is obtained by using the following formula:

V＝softmax(f_θ(T_i(x)))，

wherein, T_i(x) Transforming T for EEG data x_iData obtained after scaling, V_i(0. ltoreq. i. ltoreq.k-1) represents the EEG data T to be transformed by the classifier_i(x) The probability of predicting as class i.

6. The method for electroencephalogram abnormality detection based on self-supervised learning as recited in claim 1, wherein the score of the EEG data abnormal value is obtained by using the following formula:

wherein, g_s(x) Denotes the degree of abnormality of x, Vi]Representing EEG data x by T_iAfter transformation, the classifier f_θAnd judging the probability that the label is i.

7. An electroencephalogram abnormality detection apparatus based on self-supervised learning, characterized by comprising:

a data transformation module to generate a self-label dataset from an electroencephalogram (EEG) signal;

the data training module is used for training a classifier according to the self-label data set;

and the data scoring module is used for scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier.

8. The apparatus for electroencephalogram abnormality detection based on self-supervised learning of claim 7, wherein the data transformation module comprises:

the linear interpolation unit is used for performing linear interpolation of different scales on the EEG original data;

a data interception unit for intercepting data with the same dimensionality as the original EEG data from the interpolated data; the interception position can be before and after the center of the data sequence after interpolation, and can also be the front end of the sequence.

9. A computer arrangement comprising a processor for executing a computer program stored in a memory to implement the method of electroencephalogram abnormality detection based on self-supervised learning of any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, wherein a processor is configured to execute the computer program stored in the storage medium to implement the method for self-supervised learning based electroencephalography abnormality detection according to any one of claims 1 to 6.

Technical Field

The invention relates to the technical field of image processing, in particular to a electroencephalogram abnormity detection method and device based on self-supervision learning.

Background

In recent years, the algorithm related to abnormality detection is also effectively applied to the field of medical images, and is of great importance for detecting abnormalities (diseases) in image data, which is beneficial to the diagnosis of diseases, the detection of disease progression, the observation of treatment response and the formulation of subsequent treatment schemes.

In the prior art, in the field of eeg (electroacoustical) data anomaly identification, there are many classifier algorithms based on manually selected features and machine learning. Among them, there are algorithms that first select certain features of the EEG data, such as frequency domain and time domain analysis of the EEG data to obtain features, or first perform fourier transform and Wavelet transform (Wavelet transform) feature extraction on the EEG signal. The resulting features are then classified using some sort algorithm. However, the above methods rely on a series of specific and limited manual operations to extract features from the raw brain wave data, and then characterize the raw brain wave data with the extracted features. Therefore, the performance of the model depends on the degree of understanding of brain wave signals, the threshold is high, certain specific parameters need to be manually selected as characteristic values, the characteristic extraction process is blind, and the workload is large. In addition, some algorithms automatically extract features of data by deep learning, so that a manual feature selection process is omitted, but normal and abnormal data are required in the training process of the algorithms, namely a large amount of labeled data is required, and the workload of doctors is increased.

Therefore, how to provide a electroencephalogram anomaly detection method based on self-supervised learning to improve the accuracy of image data anomaly detection is an urgent technical problem to be solved.

Disclosure of Invention

The technical problem to be solved by the invention is how to provide a electroencephalogram anomaly detection method based on self-supervised learning so as to improve the accuracy of image data anomaly detection.

To this end, according to a first aspect, an embodiment of the present invention discloses a method for electroencephalogram anomaly detection based on self-supervised learning, including: generating a self-label dataset from an electroencephalography (EEG) signal; the self-label dataset comprises transformed EEG data and labels corresponding to the transformations; training a classifier according to the self-label data set; the classifier consists of four convolution modules and a full connection layer; according to the k-dimensional vector V output by the classifier, grading an EEG data abnormal value; k represents the radix | T | of the transform set.

Optionally, the generating the self-label dataset from EEG data specifically comprises: linear interpolation of different scales is carried out on EEG original data; intercepting data with the same dimensionality as the original EEG data from the interpolated data; the interception position can be before and after the center of the data sequence after interpolation, and can also be the front end of the sequence.

Optionally, the self-tag data set is obtained by using the following formula: s_r＝{(T_i(x) I): x belongs to S and Ti belongs to T, wherein S belongs to T_TT {.. } is a set of transforms performed on EEG raw data, x is EEG data, from a labeled dataset.

Optionally, training a classifier according to the self-labeling dataset specifically includes: using each EEG data in the self-labeling dataset as a multiclass classification model f_θAnd setting the cross entropy as a loss function to train a multi-class classifier.

Optionally, the k-dimensional vector V is obtained by using the following formula: v ═ softmax (f)_θ(T_i(x) In a) of (D), wherein T)_i(x) Transforming T for EEG data x_iData obtained after scaling, V_i(0. ltoreq. i. ltoreq.k-1) represents the EEG data T to be transformed by the classifier_i(x) The probability of predicting as class i.

Optionally, the score of the EEG data outliers is obtained using the following formula:wherein, g_s(x) Denotes the degree of abnormality of x, Vi]Representing EEG data x by T_iAfter transformation, the classifier f_θAnd judging the probability that the label is i.

According to a second aspect, an embodiment of the present invention discloses an apparatus for electroencephalogram anomaly detection based on self-supervised learning, including: a data transformation module to generate a self-label dataset from an electroencephalogram (EEG) signal; the data training module is used for training a classifier according to the self-label data set; and the data scoring module is used for scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier.

Optionally, the data transformation module includes: the linear interpolation unit is used for carrying out linear interpolation of different scales on EEG original data; a data interception unit for intercepting data with the same dimensionality as the original EEG data from the interpolated data; the interception position can be before and after the center of the data sequence after interpolation, and can also be the front end of the sequence.

According to a third aspect, an embodiment of the present invention discloses a computer apparatus, comprising a processor for executing a computer program stored in a memory to implement the method for electroencephalogram abnormality detection based on self-supervised learning of any one of the above first aspects.

According to a fourth aspect, an embodiment of the present invention discloses a computer-readable storage medium, on which a computer program is stored, wherein a processor is configured to execute the computer program stored in the storage medium to implement the method for self-supervised learning based electroencephalography abnormality detection of any one of the above-mentioned first aspects.

The invention has the following beneficial effects: the embodiment of the invention discloses a electroencephalogram abnormity detection method based on self-supervision learning, which comprises the steps of generating a self-label data set according to EEG data; training a classifier according to the self-label data set; and scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier. Compared with the prior art, the method can detect the abnormity of the EEG data without artificial feature extraction and excessively complex preprocessing of the EEG data, and achieves better detection effect.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flowchart of a method for electroencephalogram anomaly detection based on self-supervised learning according to the present embodiment;

fig. 2 is a schematic structural diagram of an electroencephalogram anomaly detection apparatus based on self-supervised learning according to the present embodiment;

FIG. 3 is a schematic diagram of self-labeling dataset generation for a method for electroencephalogram anomaly detection based on self-supervised learning according to the present embodiment;

fig. 4 is a schematic diagram of an abnormality detection process of the electroencephalogram abnormality detection method based on self-supervised learning according to the embodiment;

fig. 5 is a schematic diagram comparing the application of the electroencephalogram anomaly detection method based on self-supervised learning disclosed in this embodiment with other detection methods.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The embodiment of the invention discloses a method for detecting electroencephalogram abnormity based on self-supervision learning, which comprises the following steps of:

step S101, generating a self-label data set according to an electroencephalogram (EEG) signal; the self-label dataset comprises transformed EEG data and labels corresponding to the transformations;

step S102, training a classifier according to the self-label data set; the classifier consists of four convolution modules and a full connection layer; mapping the averaged pooled output into k-dimensional space through the fully-connected layer;

step S103, according to the k-dimensional vector V output by the classifier, the abnormal value of the EEG data is scored; k represents the radix | T | of the transform set.

The embodiment of the invention discloses a electroencephalogram anomaly detection method based on self-supervised learning, which comprises the steps of generating a self-label data set according to EEG data; training a classifier according to the self-label data set; and (4) scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier. Compared with the prior art, the method can detect the abnormity of the EEG data without artificial feature extraction and excessively complex preprocessing of the EEG data, and achieves better detection effect.

In an implementation, generating the self-label data set from the EEG raw data specifically comprises: linear interpolation of different scales is carried out on EEG original data; the data is cut from the interpolated data to the same dimensions as the original EEG data. For example, a given EEG waveform has 400 data points, and the original 400 data points are interpolated to 2 times the original to obtain 800 data points, and then the middle 400 data points or the front 400 data points are cut from the middle part or the front end of the data sequence, so as to achieve the goal of amplifying the original EEG data by two times.

In the specific implementation process, the self-tag data set is obtained by adopting the following formula:

S_T＝{(T_i(x) I): x belongs to S and Ti belongs to T, wherein S belongs to T_TT {.. } is a set of transforms performed on EEG raw data, x is EEG data, from a labeled dataset.

In a specific implementation process, training the classifier according to the self-labeling data set specifically includes: using each EEG data in the self-labeling dataset as a multiclass classification model f_θAnd setting the cross entropy as a loss function to train a multi-class classifier.

In the specific implementation process, a k-dimensional vector V is obtained by using the following formula:

V＝softmax(f_θ(T_i(x) In a) of (D), wherein T)_i(x) Transforming T for EEG data x_iData obtained after scaling, V_i(0. ltoreq. i. ltoreq.k-1) represents the EEG data T to be transformed by the classifier_i(x) The probability of predicting as class i.

In a specific implementation, the score of the outlier of the EEG data is obtained using the following formula:

wherein, g_s(x) Denotes the degree of abnormality of x, Vi]Representing EEG data x channelsPassing T_iAfter transformation, the classifier f_θAnd judging the probability that the label is i.

The embodiment of the invention discloses a device for detecting electroencephalogram abnormity based on self-supervision learning, which comprises the following components as shown in figure 2: a data transformation module 201 for generating a self-label data set from an electroencephalogram (EEG) signal; a data training module 202, configured to train a classifier according to the self-labeling data set; and the data scoring module 203 is used for scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier.

In a specific implementation process, the data transformation module 201 includes: the linear interpolation unit is used for carrying out linear interpolation of different scales on EEG original data; a data interception unit for intercepting data from the interpolated data having the same dimensionality as the original EEG data. In a specific implementation process, the interception position may be before or after the center of the interpolated data sequence, or may be the front end of the sequence.

In addition, an embodiment of the present invention further provides a computer apparatus, where a processor executes computer instructions, so as to implement the following method:

generating a self-label dataset from an electroencephalography (EEG) signal; the self-label dataset comprises transformed EEG data and labels corresponding to the transformation of the transformed EEG data; training a classifier according to the self-label data set; the classifier consists of four convolution modules and a full connection layer; and (4) scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier.

FIG. 3 is a schematic diagram of a self-labeling dataset generated according to a method for electroencephalogram anomaly detection based on self-supervised learning disclosed in the present embodiment; fig. 4 is a schematic diagram of an abnormality detection process of the electroencephalogram abnormality detection method based on self-supervised learning according to the embodiment; fig. 5 is a schematic diagram comparing the application of the electroencephalogram anomaly detection method based on self-supervised learning disclosed in this embodiment with other detection methods.

As shown in fig. 3, the left part of the diagram is a multi-sequence electroencephalogram data; in the middle of the figure, each sequence is scaled in the time dimension by a different scaling transform; the right part of the diagram is to select a part of each sequence (around the center of the sequence) to be truncated to form a plurality of new self-labeling electroencephalogram data, with different labels corresponding to different scaling transformations.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a Random Access Memory (RAM), or the like. The computer processor is used to execute a computer program stored in a storage medium to implement the following method:

generating a self-label dataset from an electroencephalography (EEG) signal; the self-label dataset comprises transformed EEG data and labels corresponding to the transformation of the transformed EEG data; training a classifier according to the self-label data set; the classifier consists of four convolution modules and a full connection layer; and (4) scoring the EEG data abnormal value according to the k-dimensional vector V output by the classifier.

It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. These should also be construed as the scope of the present invention, and they should not be construed as affecting the effectiveness of the practice of the present invention or the applicability of the patent. And are neither required nor exhaustive of all embodiments. The scope of the claims of the present application shall be determined by the contents of the claims, and the description of the embodiments and the like in the specification shall be used to explain the contents of the claims. And obvious variations or modifications therefrom are within the scope of the invention.