High recall assisted detection of egg laying events in Caenorhabditis elegans using human in the loop validation

Accurate quantification of egg-laying in Caenorhabditis elegans from 4K video recordings remains a bottleneck due to the cost of manual annotation and, above all, the risk of missing events. We present a human-in-the-loop assistance method explicitly designed under a recall-first criterion: to avoid missed egg-laying events (recall close to 1.0) while reducing operator review time. The workflow combines two complementary, deliberately over-detecting approaches - appearance by inter-frame difference restricted to a worm-centred local mask, and trajectory-based sustained-change detection - with a minimalist GUI that allows events to be confirmed or discarded within seconds and exports traceable results. The method was validated on six videos (4K, 5 fps; reviewed at 25 fps; one worm per video). Using manual annotation as the ground truth, the system achieved a recall of 0.992 (95% CI: 0.955-0.999) with moderate precision (0.77), consistent with the recall-first strategy. The approach shifts operator effort from exhaustive frame-by-frame searching to rapidly confirming pre-filtered candidates, maintaining full traceability and reproducibility.