基于深度学习的ADHD儿童和正常儿童脑电信号分类研究

Study of EEG signal classification based on deep learning for ADHD children and normal children

针对注意缺陷多动障碍(attention deficit hyperactivity disorder,ADHD)儿童和正常儿童的分类问题,实验采用经典干扰控制任务范式对两类儿童的事件相关电位(event-related potential,ERP)进行了研究,旨在通过ERP特征实现其分类。实验首次使用长短期记忆(long-short term memory,LSTM)方法分析两类儿童前额叶与顶枕叶脑区最佳电极(p <0. 05)潜伏期(200～450 ms)的脑电信号,并自动学习和分类其ERP特征。相比常规分类方法,LSTM方法的分类率略高,可达95. 78%。研究结果表明LSTM方法有助于ADHD儿童脑电信号的分类,为ADHD儿童个体诊断技术提供了一种新思路。

To solve the classification problem of attention deficit hyperactivity disorder(ADHD) children and normal children,it the experiment studied the event-related potential of them with the classical interference control task experimental paradigm in order to distinguish two categories of children through the ERP characteristics. In the experiment,firstly it used the long short-term memory(LSTM) method to analyze the EEG signals of two kinds of children’s the optimal electrodes(p <0. 05) in the frontal and parietal-occipital regions during the latency(200 ~ 450 ms),learn its characteristics automatically and realize classification. The classification rate was slightly higher than the conventional method,up to 95. 78%. The results show that the LSTM method is helpful to classify the EEG signals of children with ADHD,which provides a new idea for the individual diagnosis of ADHD children.