Hardware Implementation of a Performance and Energy-optimized Convolutional Neural Network for Seizure Detection

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Neurology

Background:

• Epilepsy affects millions globally, necessitating advanced seizure detection methods.
• Current seizure detection systems often require significant power, limiting their use in wearable or implantable devices.

Purpose of the Study:

• To introduce a novel, ultra-low-power convolutional neural network (CNN) for real-time seizure detection.
• To evaluate the performance of this CNN on a patient dataset for clinical applicability.

Main Methods:

• Development of a μW-power convolutional neural network (CNN) architecture.
• Implementation and testing on a low-power microcontroller.
• Validation using a dataset comprising 22 epilepsy patients.

Main Results:

• The CNN achieved a median sensitivity of 100% across the patient cohort.
• A low false positive rate of 20.7 false positives per hour was recorded.
• A rapid detection delay of 3.4 seconds was observed.

Conclusions:

• The developed μW-power CNN is highly effective for seizure detection.
• Its low power consumption and high accuracy make it suitable for implantable closed-loop seizure management systems.
• This technology represents a significant advancement in wearable epilepsy monitoring.