A multi-channel threshold-based seizure detection algorithm for low-complexity hardware implementation

Roughly a quarter of patients affected by epilepsy experience drug-resistant seizures, which cannot be treated with pharmacological therapy. For these individuals, implanted neurostimulation devices, such as the ones used for Deep Brain Stimulation (DBS), have emerged as potentially effective solutions. However, their performance depends heavily on accurate and timely seizure onset identification to trigger the stimulation electrodes. This paper presents a low-complexity automated seizure detection algorithm based on the Teager Energy Operator (TEO), aimed at capturing abrupt power fluctuations in neural signals that precede the ictal state. Unlike traditional machine learning-based approaches, which rely on computationally intensive feature extraction and training processes, the proposed method exploits TEO's ability to detect instantaneous energy variations, enabling fast and accurate identification of seizures. The algorithm processes multi-channel neural recording data using a lightweight, adaptive thresholding strategy, with an integrated offline calibration process, achieving an accuracy of 99.23% and sensitivity of 96.75% on the clinical CHB-MIT dataset, with an average delay of 2.36 s. A comparative analysis demonstrates that the proposed approach is comparable with several state-of-the-art ML-based solutions in terms of both detection accuracy and latency, while being well-suited for real-time and resource-constrained applications such as implanted neuromodulation systems. To confirm the system's suitability, the Field-Programmable Gate Array (FPGA) implementation of the proposed architecture demonstrates that the algorithm can be deployed using a relatively small number of hardware resources.