Detection of ADHD Disorder in Children using Layer-wise Relevance Propagation and Convolutional Neural Network: an EEG Analysis

Abstract

In this study, a deep learning-based model for ADHD diagnosis in children has been presented. For this purpose, the dataset of the ‘First_National_EEG_Data_Analysis_Competition_with_Clinical_Application’. After preprocessing, data was segmented to 3-second epochs and the frequency features of these epochs were extracted. The Fourier transform was applied to each of the channels separately and, finally, the two-dimensional matrix obtained (channel x frequency) for each epoch is considered to be the input of the Convolutional Neural Network (CNN). The CNN is made up of two convolutional layers, two max pooling layers and two fully connected layers as well as the output layer (totally 9 layers) for classification. In order to improve the performance of the method, the output of the classification of each input variable was examined. In other words, the role of each channel / frequency in the final classification has been investigated using the Layer-wise Relevance Propagation (LRP) algorithm. According to the results of LRP algorithm, only efficient channels have been used as the Convolutional Neural Network (CNN) inputs in the next step. The final accuracy obtained by this approach was 94.52% for validation data. The findings indicate that the proposed technique can be utilized for ADHD diagnosis.