SPA-IoT with MCSV-CNN: a novel IoT-enabled method for robust pre-ictal seizure prediction

Area of Science:

• Medical Technology
• Artificial Intelligence
• Signal Processing

Background:

• Epileptic seizures require accurate and timely prediction for effective patient management.
• Existing seizure prediction methods often lack the efficiency for real-time application in wearable devices.

Purpose of the Study:

• To introduce a novel, lightweight Convolutional Neural Network (CNN) for real-time epileptic seizure prediction.
• To develop a model suitable for integration into Internet of Things (IoT) enabled wearable technology.
• To improve early detection of seizures using electroencephalogram (EEG) data.

Main Methods:

• Developed the Multiresolution Critical Spectral Verge CNN (MCSV-CNN) architecture.
• Employed multiresolution feature extraction and spatial feature learning on EEG segments.
• Evaluated the model on clinical EEG recordings and the TUH-EEG corpus with a 5-minute pre-ictal window and 10-minute seizure occurrence prediction (SOP) horizon.

Main Results:

• Achieved an average prediction accuracy of 99.5% and sensitivity of 98.3%.
• Demonstrated a low false prediction rate of 0.045 and a high Area Under the Curve (AUC).
• Outperformed existing CNN-based seizure prediction techniques.

Conclusions:

• The MCSV-CNN model shows significant potential as a dependable, real-time seizure prediction tool.
• Its lightweight architecture is well-suited for practical application in wearable medical technology.
• The technology may enable early clinical intervention and continuous at-home monitoring for epilepsy management.