Seizure-Cluster-Inception CNN (SciCNN): A Patient-Independent Epilepsy Tracking SoC With 0-Shot-Retraining.

Area of Science:

• Biomedical Engineering
• Neurology
• Artificial Intelligence in Healthcare

Background:

• Epilepsy tracking Systems-on-Chips (SoCs) traditionally require patient-specific classification of electroencephalogram (EEG) signals.
• This patient-specific approach necessitates prior collection of patient EEG data, involving costly and time-consuming hospitalization.
• Variations in seizure patterns among patients pose a challenge for generic epilepsy detection systems.

Purpose of the Study:

• To develop and present a patient-independent epilepsy tracking SoC.
• To enable direct deployment of the SoC to target patients without prior data collection or patient-specific training.
• To improve the efficiency and accessibility of epilepsy monitoring.

Main Methods:

• The proposed SoC utilizes a Seizure-Cluster-Inception Convolutional Neural Network (SciCNN) Neural Processor (SNP).
• Kernel-Wise Pipeline (KWP) is implemented to significantly reduce the Static Random-Access Memory (SRAM) access rate by 179.05×.
• The 22-channel SoC was trained on pre-existing EEG databases.

Main Results:

• The patient-independent SoC achieved high performance on unseen patients from diverse datasets.
• Event-based sensitivity reached 90.3% (CHB-MIT), 90.4% (EU database), and 83.3% (local hospital).
• Event-based specificity was recorded at 93.6% (CHB-MIT), 95.7% (EU database), and 88.6% (local hospital).

Conclusions:

• The developed patient-independent epilepsy tracking SoC effectively detects seizures without requiring patient-specific training.
• The SciCNN Neural Processor with KWP demonstrates significant efficiency in reducing SRAM access.
• This technology offers a promising, cost-effective, and accessible solution for real-time epilepsy monitoring.