OPERA: Multi-Agent Framework for Biomedical Image Analysis Without Retraining

UW · hf · 2026-07-30

Introduces OPERA (Offline Policy-guided Expert Routing and Adaptation), a multi-agent ensemble framework designed to tackle severe distribution shifts in real-world biomedical image analysis (e.g., across different scanners, protocols, and patient populations).

Instead of costly domain-specific fine-tuning, OPERA treats expert weight assignment as an offline policy learning problem. A routing policy is learned from a small validation set without gradient updates to any expert agent, which is then deployed with test-time adaptation to handle distribution shifts.

Evaluated on 9 datasets covering fundus photography, X-ray, CT, and MRI, OPERA outperforms 30+ baselines across classification, segmentation, and multimodal settings, proving to be a practical path to deployable biomedical AI without retraining.

Original post →

More from Research

Research channel →