부산대학교 도서관

According to the World Health Organization (WHO), around 70 million people are affected by Epilepsy. Accurate prediction of seizures well before its commencement alarms the epileptic patients to take necessary actions to reduce the consequence of these seizures. The major objective of this work is the real-time prediction of the epileptic seizures well before its onset. A smart, bio-inspired, self-cognizant and proactive seizure predictor called ForeSeiz is designed for the purpose of forecasting of seizures. This design is suitably constructed on the basis of the Internet of Medical Things (IoMT) framework. The front-end circuitry and Seizure Predictor Tag (SPT) are designed and integrated together into a seizure predictor headband, which totally weighing of 30 grams and dissipating 52.74 mW power dissipation. The proposed ForeSeiz predictor has an Enhanced Convolutional Neural Network (ECNN) classification model, optimized using Fletcher Reeves Algorithm (FRA) along with a Phase Transition Predictor (PTP) based on the Kullback-Leibler divergence for the prediction of real-time seizures. The model is trained and tested using CHB-MIT, NINC and SRM EEG recordings via transfer learning method and yielded 97% accuracy, 0.12 FP/h false prediction rate, along with Premium Seizure Prediction Horizon (PSPH) of 66.52 minutes, prior to the onset of seizures. A compatible Seizure Prediction mobile application (SeizPred APP) is designed for the interaction to the Firebase cloud, for recording the states of the epileptic patients for further reference for the doctors and if any onset of seizure is predicted, it is immediately informed to the caretakers for further intervention actions.