Explanatory Analysis of EEG Data to Detect Seizures

In this paper, we address one of the long-standing psychological disorders-seizure and the classification of patients as susceptible to seizure or healthy subjects. The classification is based on the EEG (electroencephalogram) recordings provided, by deploying the LSTM (long short-term memory) neural network architecture. The Temple University Hospital (TUH) EEG corpus dataset is selected as the dataset to be worked upon throughout this paper. The TUH EEG corpus is an open and large dataset, which undergoes frequent updating. Along with the EEG recordings, the session records also contain a detailed text report describing the medical history and the current condition of the patient as well. Another model, trained to utilize the NLP (natural language processing) features, approves the classification performed by the neural network on the EEG recordings. The proposed framework then undergoes evaluation on the validation dataset. The accuracy achieved by the framework is 73.16% on EEG data and 64.86% on text data.