CRC-Router: Risk-Constrained Routing for Medical Agentic AI Systems

2026-09-24

AI Summary

This study addresses the high risk of error propagation and the absence of reliable routing mechanisms in autonomous medical AI decision-making by proposing CRC-Router. This method introduces a novel feature construction strategy that integrates multi-source uncertainty signals with predictive scores, employing a lightweight model to estimate false acceptance risk. Decision thresholds are subsequently calibrated via conformal risk control to ensure safe routing. The proposed architecture is model-agnostic and can be seamlessly integrated into both conventional predictive models and agentic AI systems. Experimental results demonstrate that CRC-Router achieves state-of-the-art risk-coverage trade-offs on the ChestX-ray14 dataset and substantially enhances the clinical safety of agent-based systems such as MedRAX. The source code has been made publicly available.

Abstract

Agentic AI systems are increasingly being explored in medical imaging to improve throughput and reduce clinician workload; however, safe deployment remains challenging because autonomous errors may propagate into downstream clinical decisions. A central requirement is therefore not only strong predictive performance, but also a reliable routing mechanism that determines when the system should proceed autonomously and when a case should be escalated for further review. To address this gap, we propose CRC-Router, a risk-constrained, uncertainty-aware routing module that is applicable to both conventional medical prediction models and agentic medical AI systems. CRC-Router combines multiple complementary uncertainty signals with the predictive score to construct a per-finding routing feature vector, maps this vector to an estimated wrong-accept risk using a lightweight per-finding risk model, and then applies Conformal Risk Control (CRC) to calibrate acceptance thresholds under a user-specified risk target. Instantiated on chest X-ray multi-finding triage using the NIH ChestX-ray14 dataset, CRC-Router achieves the strongest empirical risk--coverage trade-off among the evaluated baselines, both as a standalone routing layer and as a plug-in module integrated with the state-of-the-art MedRAX agent. These results demonstrate both the effectiveness of CRC-Router in selective medical automation and its modular, model-agnostic compatibility with existing predictive and agentic medical pipelines. Code is publicly available at https://github.com/XLIAaron/CRC-Router

Problem

• Agentic AI
• Medical Imaging
• Risk-Constrained Routing
• Uncertainty
• Clinical Safety

Innovation

• Risk-Constrained Routing
• Conformal Risk Control
• Uncertainty-Aware
• Agentic AI
• Model-Agnostic