Dimension reduction methods, persistent homology and machine learning for EEG signal analysis of Interictal Epileptic Discharges

TL;DR

This paper explores combining dimension reduction techniques, topological data analysis, and machine learning to improve detection of epileptic discharges in EEG signals.

Contribution

It introduces a novel approach integrating DyCA, PCA, persistent homology, and SVM for EEG event classification.

Findings

DyCA and PCA effectively reduce EEG data dimensions.

Persistent homology provides meaningful features for classification.

SVM achieves high accuracy in distinguishing IEDs from background.

Abstract

Recognizing specific events in medical data requires trained personnel. To aid the classification, machine learning algorithms can be applied. In this context, medical records are usually high-dimensional, although a lower dimension can also reflect the dynamics of the signal. In this study, electroencephalogram data with Interictal Epileptic Discharges (IEDs) are investigated. First, the dimensions are reduced using Dynamical Component Analysis (DyCA) and Principal Component Analysis (PCA), respectively. The reduced data are examined using topological data analysis (TDA), specifically using a persistent homology algorithm. The persistent homology results are used for targeted feature generation. The features are used to train and evaluate a Support Vector Machine (SVM) to distinguish IEDs from background activities.