Seizure Detection Based on Lightweight Inverted Residual Attention Network.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Accurate seizure detection is critical for epilepsy patient care.
• Current electroencephalography (EEG) seizure detection models are often computationally intensive.
• There is a need for efficient and lightweight seizure detection methods that consider key EEG signal characteristics.

Purpose of the Study:

• To develop a lightweight EEG-based seizure detection model.
• To enhance feature extraction and discrimination in EEG signals.
• To improve the efficiency and accuracy of epilepsy seizure detection.

Main Methods:

• Proposed a lightweight inverted residual attention network (LRAN) for EEG seizure detection.
• Utilized four-stage inverted residual mobile blocks (iRMB) for hierarchical feature extraction.
• Incorporated the convolutional block attention module (CBAM) to focus on salient channel and spatial information.

Main Results:

• Achieved 99.25% accuracy in segment-based and 0.36/h false detection rate in event-based intra-subject detection.
• Obtained 84.32% accuracy in inter-subject detection.
• The LRAN model is computationally efficient with 25.86 M MACs and 0.57 M parameters.

Conclusions:

• The proposed LRAN model provides a lightweight and effective solution for EEG-based seizure detection.
• LRAN demonstrates high accuracy and efficiency, outperforming existing complex models.
• This model has the potential to significantly aid in the diagnosis and treatment of epilepsy.