Real-Time Eeg-Based Epileptic Seizure Detection by One-Class Classification using CNN-LSTM and STFT

Epilepsy is a neurological disorder characterized by abnormal brain activity leading to seizures. Seizures are typically studied using Electroencephalograms (EEG). Automatically detecting seizure patterns in real-time using EEG data can greatly assist neurologists. But this task is challenging due to the patient-specific nature of EEGs, the scarcity of seizure data for each patient, and the complex spatiotemporal nature of EEG patterns. This paper presents a novel seizure detection method that combines a hybrid CNN-LSTM architecture for signal reconstruction with an auxiliary CNN that uses Short-Time Fourier Transform (STFT) for one-class classification (OCC), all trained on interictal samples only. On the CHB-MIT dataset, our method distinguished ictal and interictal samples with 90.83% sensitivity and 88.62% specificity. It was also able to detect seizure events with 91.56% sensitivity and a false alarm rate of 0.1 per hour. These results demonstrate that our approach outperforms current methods that also employ OCC. Our method is even comparable to supervised techniques, offering the additional advantage of resilience to the uneven distribution of seizure and non-seizure events.