Deep learning based epileptic seizure detection with EEG data

Epilepsy is one kind of life frightening and exigent intellect mayhem in which affected patients endure recurrent seizures. Large numbers of people are affected by this chaos worldwide. However, there is a lack of proper decision-making systems to predict the disease in its earlier stage. Electroencephalography (EEG) is the general clinical approach used for seizure detection where the electrical bustle of the brain is retrieved as signals. Identifying the seizures on-time is important for every patient to provide medications and protect the patients from adverse effects. The manual examination of the EEG signals takes more time and it is an arduous process which may lead to less performance sometimes. Developing automatic patient-specific seizure-detection system can help in intimating the seizure occurrence to the patients and the neurologists. Numerous automatic seizure-detection systems are implemented based on the conventional approaches and Deep learning approaches. Most of the available DL methods focus on cross-patient seizure detection only. Only few deep learning approaches were implemented for patient- specific seizure-detection and provide less performance only. In this work two different DL models are implemented for patient-specific seizure detection using CHB-MIT data and it provides better results than existing DL model. The first model focus on the one-dimensional CNN and the second model focus on hybrid architecture of CNN and LSTM. The prediction accuracy is 94.83%, sensitivity is 90.18%, 99.48% specificity, 99.43% precision, 94.5% F1-score, FPR is 0.5, FNR is 0.9, MCC is 90 for 11 epochs. Finally, it is proved that these two models provide different performance for different patients.