Towards Brain Big Data Classification: Epileptic EEG Identification With a Lightweight VGGNet on Global MIC

Abstract

Brain big data empowered by intelligent analysis provide an unrivalled opportunity to probe the dynamics of the brain in disorder. A typical example is to identify evolving synchronization patterns from multivariate electroencephalography (EEG) routinely superimposed with intensive noise in epilepsy research and practice. Under the circumstance of insufficient a priori knowledge of subject dependency on domain problem, it becomes even more important to adaptively classify the synchronization dynamics to accurately characterize the intrinsic nature of seizure activities represented by the EEG. This paper first measures the global maximal information coefficient (MIC) of all EEG data channels to form a time sequence of correlation matrices. A lightweight VGGNet (Visual Geometry Group) is designed to adapt to the need to prune massive EEG datasets. The $VGGNet$ characterizes the synchronization dynamics captured in the correlation matrices and then automatically identifies the seizure states of the EEG. Experiments are performed over the Children’s Hospital Boston-Massachusetts Institute of Technology (CHB-MIT) scalp EEG dataset to evaluate the proposed approach. Seizure states can be identified with an accuracy, sensitivity, and specificity of [98.13%±0.24%], [98.85%±0.51%], and [97.47%±0.36%], respectively; the resulting performance is superior to those of most existing methods over the same dataset. The approach directly applies to raw EEG analysis, which holds great potential for handling brain big data.