Absence Seizure Detection Based on Embedded Machine Learning

Epilepsy affects about 50 million people worldwide, making it one of the most common neurological diseases. One type of seizure is absence seizures, which are typically short, sudden lapses of consciousness, often going unnoticed and delaying diagnosis. Most absence seizure detection methods involve machine learning and deep learning techniques; however, their algorithms are not implemented on an ultra-low-power microcontroller as an edge device. In this paper, we present the implementation of absence seizure detection algorithms on an ultra-low-power STM32L432KC microcontroller based on the Long Short-Term Memory network using the absence seizure dataset. The algorithm reaches a sensitivity of 84.6%, specificity of 95%, and F1-score of 79.4% for time-based evaluation and a sensitivity of 100% and F1-score of 85.7% for episode-based evaluation. Our proposed network implemented on STM32L432KC microcontroller performs 720,021 Multiply-Accumulate (MACC) operations and requires 32.89 KB RAM and 35.04 KB Flash, suitable for low-power microcontrollers due to their low computational and memory use.