Automatic seizure detection in ECoG by differential operator and windowed variance.
Kaushik Kumar Majumdar1, Pratap Vardhan
1Systems Science and Informatics Unit, Indian Statistical Institute, Bangalore 560059, India. kmamjumdar@isibang.ac.in
Summary
A new Differential Windowed Variance (DWV) algorithm effectively detects seizure onset in electroencephalography (EEG) signals. This rapid, simple method achieves over 91% accuracy, aiding epilepsy diagnosis.
Area of Science:
- Neuroscience
- Signal Processing
- Medical Technology
Background:
- Differential operators are crucial for detecting significant changes in image and signal processing.
- The windowed variance method has shown success in identifying seizure onset in brain electrophysiological signals.
- Electrophysiological signals like ECoG (electrocorticography) are often contaminated by noise and artifacts, complicating accurate analysis.
Purpose of the Study:
- To enhance features in noisy electrophysiological brain signals for efficient change detection.
- To develop and validate a novel algorithm for automatic, real-time seizure onset detection in epileptic patients.
- To combine differential operators and windowed variance methods into a new algorithm named Differential Windowed Variance (DWV).
Main Methods:
- Developed the Differential Windowed Variance (DWV) algorithm by integrating differential operators with the windowed variance method.
- Tested the DWV algorithm on 369 hours of non-seizure and 59 hours of seizure ECoG data from 15 epileptic patients.
- Introduced eight novel empirical measures to mitigate false detections and employed quasi-ROC (qROC) curve analysis for reliability assessment.
Main Results:
- The DWV algorithm achieved 91.525% accuracy, detecting all but six seizures.
- Detection occurred with an average delay of 9.2 seconds post-onset.
- A maximum of three false detections were recorded in 24 hours of non-seizure data; performance was compared against a sharp transient (ST) detection algorithm.
Conclusions:
- The Differential Windowed Variance (DWV) algorithm offers a simple, fast, and accurate method for real-time seizure onset detection in continuous ECoG signals.
- The novel qROC analysis provides a reliable way to ascertain the performance of seizure detection methods.
- DWV demonstrates significant potential for improving the clinical management of epilepsy through automated seizure monitoring.
