Automatic Seizure Detection Employing Machine Learning-Based Late Fusion Techniques Over Behind-the-Ear and the Scalp EEG Signals

Differentiating the ictal state (seizure) from the interictal state (non-seizure) for epileptic patients over limited channels EEG wearables such as behind the ear (BTE) is a complex task. Exploring different features such as magnitude and phase of fast Fourier transform (FFT) and discrete wavelet transform (DWT) along with Youden's J statistic thresholding and late fusion over the International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2023 seizure detection challenge dataset gives satisfying results of 98% sensitivity, 86.96% specificity and 0.00069 of false alarm rate (FA/hr) over the development set and 90.32% sensitivity, 74.99% specificity and 0.00117 of false alarm rate (FA/hr) over the validation set. On analyzing this dataset, we acknowledge that the benchmark features such as FFT and DWT are giving good results and can be combined for better possible prediction metrics.