Event-driven Processing of EEG using LIF-Based Flagging for Efficient Seizure Detection

The growing scale of EEG recordings used in seizure detection and prediction presents new challenges for conventional preprocessing pipelines, particularly in the context of long-term monitoring. Additionally, manual review of EEG segments by clinicians remains time-consuming and labor-intensive. In this work, we introduce a lightweight, neuromorphic-inspired flagging method designed to identify potentially seizure-relevant EEG segments before intensive model training or manual review. EEG signals are encoded into artificial spike trains using Leaky Integrate-and-Fire neurons, with encoding parameters optimized on seizure-containing EEG data. A spike count threshold is then defined based on the minimum spike activity observed in seizure-labeled windows. Using this threshold, candidate 10second EEG windows are flagged based on total spike count. We demonstrate that this method can reduce the amount of EEG requiring further processing by up to 97.07%. This method offers a biologically inspired, low-complexity strategy for filtering EEG data streams ahead of more computationally demanding seizure analysis models or before manual review from clinicians.