Optimized dense back-projection joint attention network for automatic seizure detection from EEG signal

Seizure identification using EEG is crucial but traditionally relies on specialists, making it time-consuming and variable. Existing systems suffer from limited feature extraction and inefficient optimization. To overcome this, Optimized Dense Back-Projection Joint Attention Network for Automatic seizure recognition from EEG signal (DBPJAN-ASD-EEG) is proposed. Statistical features are extracted via Two-sided Offset Quaternion Linear Canonical Transform (TOQLCT). Extracted features are given to seizures detection and classification as normal, pre-ictal, and ictal using DBPJAN optimized with Giza Pyramids Construction Optimization Algorithm (GPCOA). DBPJAN-ASD-EEG is implemented in MATLAB and evaluated using various performance metrics achieving superior accuracy and recall over existing techniques.