Tag
OPERA proposes a multi-agent ensemble framework that treats expert weight assignment as an offline policy learning problem for universal biomedical image analysis, enabling test-time adaptation without retraining and consistently improving performance across 9 datasets and 30+ baselines.