Seizure type classification algorithm based on multi-dimensional brain network feature selection

Area of Science:

• Neuroscience
• Computational Biology
• Signal Processing

Background:

• Epilepsy seizure type classification is challenging due to complex brain network dynamics.
• Identifying relevant features from electroencephalogram (EEG) signals is crucial for accurate classification.

Purpose of the Study:

• To propose a multi-dimensional brain network feature selection (MBNFS) algorithm for improved seizure type classification.
• To enhance the performance of seizure detection models by selecting significant brain network features.

Main Methods:

• Wavelet packet transform (WPT) applied to EEG signals to define network nodes.
• Mutual information (MI) calculated for edge weights, constructing brain networks using a leave-one-out approach.
• Random forest (RF) used for feature evaluation and multi-dimensional iterative selection based on channel, sub-band, and network feature contributions.

Main Results:

• The MBNFS algorithm effectively selected discriminative brain network features.
• The RF model trained on the refined feature set achieved high classification performance.
• Test set metrics included 99.86% accuracy, specificity, and sensitivity, with an F1 score of 0.9986 and kappa of 0.9984.

Conclusions:

• The proposed MBNFS algorithm offers a robust method for seizure type classification.
• Accurate seizure detection can be achieved by leveraging multi-dimensional feature selection from EEG data.
• This approach holds potential for improving clinical diagnosis and management of epilepsy.