Epileptic Seizure Classification Based on Gramian Angular Field Transformation and Deep Learning
Anand Shankar, Hnin Kay Khaing, Samarendra Dandapat, Shovan Barma
Abstract
This work proposes a new method to classify epileptic seizures based on a well-known deep learning technique named convolutional neural network (CNN), where the input images are generated by Gramian angular field (GAF) transformation. For this purpose, the EEG signals have been assumed as time series data. Certainly, two different signals such as the EEG signal and its instantaneous power have been used for image generation by two different ways - Gramian angular summation field (GASF) and Gramian angular difference field (GADF). The generated images are directly fed into multilayer CNN having multiple hidden layers. For experimental validation, EEG dataset from Bonn University has been considered. The experimental results exhibit the classification accuracy up to 98%. The efficiency of the proposed method has been evaluated by measuring sensitivity and specificity of 99% and 98.9% respectively. In a comparative study, the proposed idea displays significant improvement in seizure classification. Thus, the proposed idea reveals the usefulness of GAF in deep learning framework for epileptic seizure classification.