Deep learning based efficient epileptic seizure prediction with EEG channel optimization

A seizure is an unstable situation in epilepsy patients due to excessive electrical discharge by brain cells. An efficient seizure prediction method is required to reduce the lifetime risk of epilepsy patients. In state-of-the-art works, either the prediction accuracy is low or the number of EEG channels used is more, but the use of 22 channels in seizure prediction is not efficient in terms of complexity, comfortability, and cost. This article depicts an efficient seizure prediction technique using a Convolutional Neural Network (CNN) with minimizing the channels. CNN has been used for automatic feature extraction and classification of states of epilepsy patients. The proposed method is capable to achieve an average classification accuracy of 99.47 % (approximate to 100 %) by optimizing the EEG channels to 6 from 22, i.e., 72.73 % of channel reduction. The proposed method claims a satisfactory result of 97.83 % and 92.36 % in terms of average sensitivity and specificity respectively with a false positive rate of 0.0764. The aforesaid results have been obtained with a prediction of ten minutes in advance. The experimental results demonstrate that the proposed method is better than the state-of-the-art works. The work can be extended to design a transportable seizure prediction device to use in real-time.

Automated summary

This study introduces an efficient seizure prediction method using Convolutional Neural Network with optimized EEG channels, achieving an average classification accuracy of 99.47%. The method demonstrates excellent performance in terms of sensitivity, specificity, and false positive rate.