A hybrid deep transformer model for epileptic seizure prediction

Abstract

The EEG is a structured and dependable approach for analysing epileptic seizures and capturing brain electrical activity. The physical effort of clinicians diagnosing epilepsy is decreased through automatic epilepsy screening employing data-driven algorithms. The latest algorithms are skewed toward signal processing or DL, each with its own set of benefits and drawbacks. The proposed hybrid framework is developed by hybridizing a feature extraction module, and deep transformer model. The fourier transform is utilized for the effective feature extraction, and deep transformer model is used for the seizure prediction. The proposed framework can interpret the hidden features to naturally select the interesting fields in EEG data for strong predictions. The proposed framework is validated using CHB-MIT database, and the performance is compared with different benchmark models. The proposed model achieved an average sensitivity of 95.2% with a false positive rate of 0.02, which is better compared to other comparative models. The proposed framework achieved excellent results on the test datasets, and can be used as a promising tool for the hospitals for examining the patients.