Algorithm for automatic EEG classification according to the epilepsy type: Benign focal childhood epilepsy and structural focal epilepsy. (February 2019)

• Record Type: Journal Article
• Authors: Misiukas Misiūnas, Andrius Vytautas; Meškauskas, Tadas; Samaitienė, Rūta
• Is Part Of: Biomedical signal processing and control. Volume 48(2019)
• Page Start: 118
• Page End: 127
• Publication Date: 2019-02
• Subjects: EEG -- Epilepsy -- Epileptiform discharge -- Spike -- Machine learning -- Artificial neural network; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.10.006
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: Highlights: A novel three-step algorithm for EEG classification by epilepsy type is proposed. Benign focal childhood epilepsy and structural focal epilepsy EEGs are dealt with. Classification is based on statistical differences of EEG spike geometric parameters. Artificial neural network well outperforms supported vector machine strategy. Classification accuracy crucially depends on spike detection reliability. Abstract: Rationale: It is still not clear if there are EEG parameters that may be related to the epilepsy etiology in epilepsies presenting with rolandic spikes. Rolandic spikes are not pathognomonic for rolandic epilepsy and could be related to the area of discharges itself. The initial hypothesis was that even visually identical spikes have some difference, because of the different etiology. Objective: The aim of the study was to find the differences in rolandic spike morphology in two epilepsy groups, different by etiology, but presenting with visually identical spikes. Methods: A novel algorithm for automatic classification of interictal electroencephalogram (EEG) rolandic spikes according to the epilepsy type (Group I – patients with benign focal childhood epilepsy, self-limiting, with no causal lesion in the brain, Group II – patients with structural focal epilepsy) is proposed. The algorithm consists of three stages: 1) EEG spike detection, 2) determination of EEG spike parameters, 3) classification of EEG by epilepsy type based on estimated spike parameters. Automatic classification method is defined by artificial neural network. The algorithm has been trained and tested on a large data sample provided by Children's Hospital, Affiliate of Vilnius University Hospital Santaros Klinikos. Only those EEGs that were visually identical and inaccessible for manual clustering to the groups according the visual spike morphology and contained 50 or more spikes have been analyzed. Training and testing pools have been selected as non overlapping (containing different patients) data sets. Results: The proposed methodology let us to achieve up to 75% of accuracy of classification of EEG.