Intracranial Electroencephalogram Based Epilepsy Seizure Onset Detection

Epilepsy is a neural disorder affecting 39 million people around the world. Recently, with the rapid growth of applying machine learning models to biomedical data, Electroencephalogram has proven capable of detecting epilepsy seizure onset. In this work, we introduce an intracranial Electroencephalogram (iEEG)-based seizure detection algorithm. This method relies on exploits the advantage of the high-frequency components embedded in iEEG signals. We select the optimum classifier for this task. We test the algorithm performance on three patients. On average, we achieve over 99% accuracy on 1s-long windows and a F1 score of 94% (for measuring imbalanced data). We also analyzed the complexity of different classifiers. Taking the classification performance and power consumption into consideration, gradient boosted decision trees present the best classifier for this task.