Related Experiment Videos
Methods for robust clustering of epileptic EEG spikes
1Department of Electrical Engineering and Computer Science, Lund University, Sweden. pw@tde.lth.se
IEEE Transactions on Bio-Medical Engineering
|August 1, 2000
Summary
This study compares clustering algorithms for epileptic electroencephalogram (EEG) spikes. A graph-theoretic method outperformed fuzzy C-means (FCM) for analyzing EEG spike data, offering improved accuracy for neurophysiologists.
Area of Science:
- Computational neuroscience
- Signal processing
- Medical informatics
Background:
- Epileptic electroencephalogram (EEG) spike clustering is crucial for accurate analysis.
- Existing methods struggle with outlier data, necessitating robust algorithms.
- Identifying distinct spike classes aids in averaging and inverse computations.
Purpose of the Study:
- To compare the performance of fuzzy C-means (FCM) and a graph-theoretic algorithm for EEG spike clustering.
- To establish criteria for determining appropriate outlier contamination levels.
- To evaluate the utility of geometrically weighted feature extraction.
Main Methods:
- Comparative analysis of fuzzy C-means (FCM) and a graph-theoretic clustering algorithm.
- Simulations to assess algorithm performance under various conditions.
- Evaluation on seven real-life epileptic EEG datasets against manual neurophysiologist assessment.
Main Results:
- Both algorithms showed good performance in simulations across diverse circumstances.
- The graph-theoretic method demonstrated superior results compared to FCM on simulated and real-world EEG data.
- Geometrically weighted feature extraction proved beneficial as an additional dimension for clustering.
Conclusions:
- The graph-theoretic approach is more effective for clustering epileptic EEG spikes than FCM.
- A hybrid approach combining automatic clustering with human selection may mitigate discrepancies.
- Robust clustering algorithms are essential for accurate analysis of noisy biomedical signals.