ALARM-Net: An Event-Level False-Alarm Suppression Framework for Clinical EEG Seizure Detection on TUSZ v2.0.6

Automated electroencephalography (EEG) seizure detection systems support clinical monitoring through alarm-driven workflows, in which the practical utility of a detector is determined by its event-level false-alarm rate. We examine the false-alarm structure produced by a strong window-level seizure detector on the Temple University Hospital Seizure Corpus (TUSZ) v2.0.6 and find that the false-alarm burden is unevenly distributed across subjects, with worst-decile subjects carrying substantially higher FA/24h than the cohort median. We propose ALARM-Net (Alarm-Level Adaptive Rejection Module), a detector-agnostic event-level alarm-suppression framework. ALARM-Net treats the window-level detector as a frozen black box, generates high-recall event proposals from its per-second probability timeline, and applies a regularized CatBoost classifier over 14 causal features summarizing each proposal's probability morphology, local pre-context, and alarm history. Operating-point selection is governed by predefined clinical constraints on the dev split (strict sensitivity loss [≤] 8 percentage points; FA/24h relative reduction [≥] 30%); the locked operating point is applied without modification to the held-out eval split. ALARM-Net reduces false alarms per 24 hours from 30.75 to 12.63 on dev (a 58.9% reduction) and from 19.43 to 4.60 on eval (a 76.3% reduction), with strict sensitivity loss of 7.5 and 7.8 percentage points respectively. The detector-only false-alarm burden is heavily concentrated on a small subset of subjects (worst-decile mean: 79.55 FA/24h on dev, 52.77 on eval), motivating the event-level suppression approach. Ablations across feature design, suppressor family, and rule-based baselines suggest that event-level reformulation and proposal morphology features, rather than the classifier family alone, drive the observed false-alarm reduction. ALARM-Net complements window-level seizure detectors and highlights the importance of event-level evaluation for clinically oriented seizure detection.