An Explainable Model for EEG Seizure Detection based on Connectivity Features

TL;DR

This paper introduces an explainable deep learning approach utilizing connectivity features from EEG signals to detect seizures with high accuracy and interpretability.

Contribution

It combines undirected and directed connectivity features with advanced neural network architectures to improve seizure detection and explain feature relevance.

Findings

Achieved 97.03% accuracy on MITBIH dataset

Provided interpretability of feature contributions across patients

Validated scientific facts about seizure-related connectivity patterns

Abstract

Epilepsy which is characterized by seizures is studied using EEG signals by recording the electrical activity of the brain. Different types of communication between different parts of the brain are characterized by many state of the art connectivity measures which can be directed and undirected. We propose to employ a set of undirected (spectral matrix, the inverse of the spectral matrix, coherence, partial coherence, and phaselocking value) and directed features (directed coherence, the partial directed coherence) to learn a deep neural network that detects whether a particular data window belongs to a seizure or not, which is a new approach to standard seizure classification. Taking our data as a sequence of ten sub-windows, we aim at designing an optimal deep learning model using attention, CNN, BiLstm, and fully connected layers. We also compute the relevance using the weights of the learned model based on the activation values of the receptive fields at a particular layer. Our best model architecture resulted in 97.03% accuracy using balanced MITBIH data subset. Also, we were able to explain the relevance of each feature across all patients. We were able to experimentally validate some of the scientific facts concerning seizures by studying the impact of the contributions of the activations on the decision.