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

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence in Medicine

Background:

• Electroencephalogram (EEG) based seizure identification is critical for neurological disorder management.
• Traditional methods are labor-intensive, time-consuming, and prone to inter-observer variability.
• Existing automated systems often lack robust feature extraction and efficient optimization strategies.

Purpose of the Study:

• To develop an advanced, automated system for accurate seizure recognition from EEG signals.
• To enhance feature extraction and optimization techniques for improved seizure detection performance.
• To classify EEG signals into normal, pre-ictal, and ictal states.

Main Methods:

• Statistical features were extracted using the Two-sided Offset Quaternion Linear Canonical Transform (TOQLCT).
• An Optimized Dense Back-Projection Joint Attention Network (DBPJAN) was developed for seizure detection and classification.
• The DBPJAN model was optimized using the Giza Pyramids Construction Optimization Algorithm (GPCOA).

Main Results:

• The proposed DBPJAN-ASD-EEG system demonstrated superior accuracy and recall compared to existing techniques.
• The system successfully detected and classified EEG signals into normal, pre-ictal, and ictal categories.
• Implementation in MATLAB facilitated comprehensive evaluation using various performance metrics.

Conclusions:

• The DBPJAN-ASD-EEG system offers a promising, highly accurate, and efficient solution for automatic seizure recognition.
• The integration of TOQLCT for feature extraction and GPCOA for optimization significantly enhances performance.
• This automated approach can reduce the burden on specialists and improve diagnostic consistency.