MPD$^2$-Router: Mask-aware Multi-expert Prior-regularized Dual-head Deferral Router in Glaucoma Screening and Diagnosis
TLDR
MPD$^2$-Router is a mask-aware multi-expert AI system for glaucoma screening that intelligently defers difficult cases to available human experts.
Key contributions
- Introduces MPD$^2$-Router, a mask-aware multi-expert AI system for glaucoma screening.
- Employs a dual-head policy with Gumbel-sigmoid gating to route cases to available experts.
- Fuses multiple signals (uncertainty, morphology, OOD) for robust, context-aware deferral.
- Uses an asymmetric cost-sensitive objective to balance expert workload and prevent collapse.
Why it matters
Current AI deferral systems for glaucoma overlook practical issues like expert availability and workload. MPD$^2$-Router solves this by intelligently routing difficult cases to available human experts, improving safety and efficiency. This leads to lower clinical costs and better diagnostic performance, making AI more robust and practical for real-world healthcare deployment.
Original Abstract
Learning-to-defer (L2D) can make glaucoma screening safer by routing difficult/uncertain cases to humans, yet standard formulations overlook expert availability, heterogeneous readers behavior, workload imbalance, asymmetric diagnostic harm, case difficulty from morphology and deployment shift. We introduce MPD$^2$-Router, a mask-aware multi-expert deferral framework that recasts ophthalmic triage as constrained human--AI routing: whether to defer and to which available expert. It couples a dual-head deferral/allocation policy with mask-aware Gumbel--sigmoid gating that strictly enforces per-sample availability, and fuses uncertainty, morphology, image-quality, and OOD signals. Training uses an asymmetric cost-sensitive objective with an augmented-Lagrangian deferral budget, a group-specific distribution prior, and a rank-majorization JS regularizer that jointly prevent expert collapse without forcing uniform allocation. Across three cross-national glaucoma cohorts (REFUGE, CHAKSU, ORIGA) with a frozen REFUGE-trained backbone, MPD$^2$-Router substantially lowers clinical cost and improves MCC over AI-only at a moderate deferral rate. It is Pareto-optimal in F1--MCC--cost, robust under cross-domain shift, and yields balanced expert utilization.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.