Abstract
Forecasting microsleep events from multimodal EEG/EOG signals is important for early vigilance assessment, but expert annotation is costly. This work investigates whether active learning (AL) can reduce labeling effort while preserving forecasting performance. Using the Maintenance of Wakefulness Test (MWT) dataset, we evaluate 1 s and 2 s ahead microsleep forecasting under strict Leave-One-Subject-Out validation. We benchmark four pool-based AL strategies and propose Multimodal Disagreement Active Learning (MD-AL), an interpretable acquisition strategy combining EEG–EOG disagreement, decision-boundary uncertainty, and rare-event prioritization. Results show that AL recovers approximately 95% of fully supervised performance using only 1–2% labeled samples. MD-AL performs competitively with established AL methods while providing a physiology-aware and explainable acquisition mechanism. Ablation analysis shows that combining multimodal disagreement with uncertainty and rare-event prioritization yields a more stable query signal than disagreement alone. These findings support active learning as a practical route toward label-efficient and interpretable microsleep forecasting.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EMERGE_039.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=adPYhf94dR
BibTex
@InProceedings{MohMar_LabelEfficient_MICCAISAT2026,
author = { Mohamed, Maram A. AND Redmond, Peter AND Mathur, Prateek AND Ward, Tomas E. AND O’Connor, Noel E.},
title = { { Label-Efficient Multimodal Microsleep Forecasting via Interpretable Disagreement-Driven Active Learning } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17266},
month = {pending},
page = {pending}
}