Classifying epileptic EEG signals with delay permutation entropy and Multi-Scale K-means

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Supervised epileptic electroencephalogram (EEG) classification algorithms require extensive datasets, limiting real-time applications.
• Traditional K-means clustering can suffer from inaccurate clustering due to random initialization.

Purpose of the Study:

• To propose an unsupervised Multi-Scale K-means (MSK-means) algorithm for distinguishing epileptic EEG signals and identifying epileptic zones.
• To enhance the efficiency and accuracy of EEG signal classification for epilepsy diagnosis.

Main Methods:

• Developed an unsupervised MSK-means algorithm that initializes cluster centroids with a scale factor based on EEG signal characteristics.
• Theoretically proved the superior efficiency of MSK-means over the standard K-means algorithm.
• Compared K-means, MSK-means, and Support Vector Machine (SVM) classifiers using delay permutation entropy features for seizure identification and epileptogenic zone localization.

Main Results:

• The MSK-means algorithm demonstrated superior theoretical efficiency compared to the K-means algorithm.
• Identifying seizures using MSK-means and delay permutation entropy achieved 4.7% higher accuracy than K-means.
• This approach yielded 0.7% higher accuracy than the SVM classifier in seizure identification.

Conclusions:

• The unsupervised MSK-means algorithm offers an efficient and accurate method for epileptic EEG signal classification.
• MSK-means shows promise for real-time seizure detection and localization of epileptogenic zones.
• This method provides a viable alternative to supervised learning algorithms, especially when large training datasets are unavailable.