Evaluation of machine learning methods for seizure prediction in epilepsy

Matthias Eberlein,Jens Müller,Georg Leonhardt,Ortrud Uckermann,Simon Walz,Hongliu Yang,Ronald Tetzlaff,Susanne Creutz,Janina Schreiber

doi:10.1515/cdbme-2019-0028

Abstract

Abstract Epilepsy affects about 50 million people worldwide of which one third is refractory to medication. An automated and reliable system that warns of impending seizures would greatly improve patient’s quality of life by overcoming the uncertainty and helplessness due to the unpredicted events. Here we present new seizure prediction results including a performance comparison of different methods. The analysis is based on a new set of intracranial EEG data that has been recorded in our working group during presurgical evaluation. We applied two different methods for seizure prediction and evaluated their performance pseudoprospectively. The comparison of this evaluation with common statistical evaluation reveals possible reasons for overly optimistic estimations of the performance of seizure forecasting systems.

Highlights

Affecting about 1 % of the world population, epilepsy is one of the most common neurological diseases
We recently proposed three Convolutional Neural Networks (CNN) topologies for seizure forecasting [3], showing that these networks produce promising results on different patients for several long-term data sets
Some studies show that the time of day can even be used as a feature for seizure prediction [7]

Summary

Introduction

Affecting about 1 % of the world population, epilepsy is one of the most common neurological diseases. Seizures cover relatively short periods in a patient’s life, the uncertainty when the seizure will occur can produce a high level of anxiety [4]. For 70 % of the patients, medication can reduce the frequency of seizures or even abolish them. Patients report that unwanted side effects of the medication as well as the unpredictability of seizures are the severest handicaps of this disease [13]. A mobile system with the ability to predict seizures can help to relief the patients’ anxiety related to the uncertainty of events by enabling them to seek shelter, apply a short acting drug or inform the treating physician about. The exact position of the electrodes could be identified in a postoperative 3-D T1-weighted MRI. The seizure onsets zone could be identified. Dysfunctional channels (identified by visual inspection) were excluded from this study

Objectives

Results

Conclusion