Research Area:  Machine Learning

Abstract:

It is critical to determine whether the brain state of an epilepsy patient is indicative of a possible seizure onset; thus, appropriate therapy or alarm may be delivered in time. Successful seizure prediction relies on the capability of accurately separating the preictal stage from the interictal stage of ictal electroencephalography (EEG). With the booming of brain e-health technologies, there exists a pressing need for an approach that provides accurate seizure prediction while operating efficiently on edge computing platforms with very limited computing resources in Internet of Things environments. This study proposes a lightweight solution to this problem based on synchronization measurement of multivariate EEG captured from multiple brain regions consisting of two phases, i.e., synchronization measurement and classification. For phase one, Pearson correlation coefficient is calculated to obtain the correlation matrices. For phase two, the correlation matrices are classified to distinguish the preictal states from the interictal ones with a simple CNN model, and seizure onset can then be predicted. Experiments have been performed to evaluate the performance of the lightweight solution on the CHB-MIT scalp EEG dataset. The experimental results indicate that: (1) the solution outperforms most of the state-of-the-art counterparts with a high accuracy of seizure prediction (89.98% for 15 mins alarm in advance) for all subjects, and (2) the solution incurs a very low computational overhead and holds potentials in brain e-health applications.

Keywords:  

Author(s) Name:  Shasha Zhang, Dan Chen, Rajiv Ranjan, Hengjin Ke, Yunbo Tang & Albert Y. Zomaya

Publisher name:  Springer

DOI:  10.1007/s11227-020-03426-4

Volume Information:  volume 77, pages: 3914–3932 (2021)

Paper Link:   https://link.springer.com/article/10.1007/s11227-020-03426-4