Convolutional Neural Network Classification of Topographic Electroencephalographic Maps on Alcoholism

Abstract

Alcohol use is a leading risk factor for substantial health loss, disability, and death. Thus, there is a general interest in developing computational tools to classify electroencephalographic (EEG) signals in alcoholism, but there are a limited number of studies on convolutional neural network (CNN) classification of alcoholism using topographic EEG signals. We produced an original dataset recorded from Brazilian subjects performing a language recognition task. Then, we transformed the Event-Related Potentials (ERPs) into topographic maps by using the ERP's statistical parameters across time, and used a CNN network to classify the topographic dataset. We tested the effect of the size of the dataset in the accuracy of the CNNs and proposed a data augmentation approach to increase the size of the topographic dataset to improve the accuracies. Our results encourage the use of CNNs to classify abnormal topographic EEG patterns associated with alcohol abuse.