Automatic seizure detection based on imaged-EEG signals through fully convolutional networks.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy seizure detection is manual, time-consuming, and inefficient.
• Automating seizure detection can improve clinical practice and patient outcomes.

Purpose of the Study:

• To propose an automatic method for epileptic seizure detection using imaged-EEG signals.
• To evaluate the performance of fully convolutional neural networks (FCNNs) for seizure detection.

Main Methods:

• Analyzed EEG data from CHB-MIT and EPILEPSIAE databases.
• Utilized fully convolutional neural networks (FCNNs) for automated seizure detection.
• Compared performance using scalp and intracranial EEG recordings.

Main Results:

• Achieved high accuracy (up to 99.6%) and specificity (up to 99.6%) on both datasets.
• Demonstrated low false alarm rates (<1.0/h for 80% of EPILEPSIAE patients).
• The method requires no pre-selected features, offering a lightweight approach.

Conclusions:

• The proposed automatic imaged-EEG seizure detection method shows high performance and efficiency.
• This approach has promising potential for integration into clinical practice for epilepsy management.
• The lightweight, feature-free model offers a viable alternative to manual analysis.