Abstract:
Epilepsy is one of the most prevalent neurological disorders in the world. It is estimated that between four and six children out of every thousand suffer from epilepsy. Visually diagnosing seizure activity can be challenging for medical professionals because doing so correctly requires a great deal of observation and practice. Recent develop ments in deep learning technology have made it so that erroneous diagnoses of seizure activity are occurring much less frequently and that significantly less time is needed to forecast when a seizure will occur. This study improves upon existing metrics by utilizing cross-frequency coupling gleaned from scalp electroencephalogram (EEG) and transfer learning inside deep neural network architectures. We conducted our study using data from the CHB-MIT dataset, which included 23 children as participants. In the first experiment, we culled all 18 channels from the dataset; in the second, we narrowed it down to just one. Fine-tuning boosted the accuracy of the first exper iment to 87.21%, and the accuracy of the second experiment to 96.82%. Images of the phase-amplitude coupling help predict the next seizure when fed into the VGG-19 architecture. Several tests are used to assess the models’ efficacy. In-depth analyses are performed and reported on.|Keywords : Scalp EEG, Pediatric Epileptic Seizures, Phase-Amplitude Coupling, Prediction, Transfer Learning.
