基于神经网络的脑电信号体质检测研究

Research on EEG Signal Constitution Detection Based on Neural Network

针对神经学专家根据青少年脑电图信号判断癫痫状况耗时长、具有主观性且容易出现误诊的问题,本研究提出了一种基于注意力机制的青少年癫痫体质检测方法。首先在预处理阶段为了拥有更多的上下文信息,将一维信号转化为二维图像,然后以VGG网络为基础,将注意力机制引入网络,给重要特征信息赋予更多的权重,最后将处理后的脑电图输入到本研究所提出的网络中进行分类检测。结果表明,本研究算法在新德里HauzKhas的神经和睡眠中心收集的数据集上进行实验,得到了99.7%的分类准确率,并在CHB-MIT数据集上进行了验证。本研究算法在脑电图信号分类方面具有潜力,对青少年癫痫体质的检测具有一定的指导意义。

Aiming at the problem that neurologists judge the status of epilepsy according to teenagers’EEG signals,which is time-consuming,subjective and prone to misdiagnosis,a detection method of teenagers’epilepsy constitution was proposed based on attention mechanism in this study.Firstly,in the preprocessing stage,in order to have more context information,one-dimensional signals were transformed into two-dimensional images.Then,based on VGG network,attention mechanism was introduced into the network,which could give more weight to important feature information.Finally,the processed EEG was input into the network proposed in this study for classification and detection.The algorithm was tested on the data set collected by the nerve and sleep center in HauzKhas,New Delhi,and the classification accuracy was 99.7%,which was verified on the CHB-MIT data set.It shows that the algorithm has potential in EEG signal classification and has certain guiding significance for the detection of teenagers’epilepsy constitution.