Effiformer: a unified data-efficient vision transformer-CNN framework for interpretable epileptic seizure detection

Area of Science:

• * Neuroscience
• * Artificial Intelligence
• * Medical Technology

Background:

• * Epilepsy affects many patients, with ~30% experiencing continued seizures despite therapy.
• * Current seizure detection methods require improvement for timely intervention and patient safety.
• * Accurate and rapid seizure detection is crucial for effective epilepsy management.

Purpose of the Study:

• * To develop a reliable and accurate seizure detection model using electroencephalogram (EEG) spectrograms.
• * To introduce EffiFormer, a hybrid Vision Transformer-CNN model for enhanced seizure detection.
• * To improve patient safety through prompt and precise seizure identification.

Main Methods:

• * Developed EffiFormer, a hybrid Vision Transformer-CNN model integrating EfficientNet and Data-Efficient Image Transformer.
• * Utilized a four-phase pipeline: normalization, STFT for spectrograms, SMOTE for synthetic data, and data augmentation.
• * Employed a 60-20-20 training-validation-testing split on the CHB-MIT EEG dataset.

Main Results:

• * EffiFormer achieved high accuracy in seizure detection using EEG spectrograms.
• * Demonstrated average sensitivity of 99.8% and average accuracy of 99.3%.
• * Explainable AI (XAI) methods were used to enhance model transparency and interpretability.

Conclusions:

• * EffiFormer offers a highly accurate and reliable solution for epileptic seizure detection.
• * The model's performance is robust, even with limited training data.
• * Explainable AI integration facilitates clinical review and trust in the detection system.