JANUS: Anatomy-Conditioned Gating for Robust CT Triage Under Distribution Shift

2026-05-13

AI Summary

This study addresses the limited sensitivity of CT-based automated triage models to quantitative imaging biomarkers—such as lesion size and density—under cross-institutional distribution shifts. To this end, the authors propose JANUS, a physiology-guided dual-stream Vision Transformer architecture that integrates macro-scale radiomic priors through an anatomy-guided gating mechanism and incorporates a physiological veto module to suppress high-confidence false positives. This design substantially enhances model robustness, discriminative performance, and calibration. Evaluated on the MERLIN test set (N=5,082), JANUS achieves a macro-AUROC of 0.88 and an AUPRC of 0.74; on an external cohort (N=2,000), it attains an AUROC of 0.87, with particularly pronounced improvements in detecting lesions defined by size and attenuation characteristics.

Abstract

Automated CT triage requires models that are simultaneously accurate across diverse pathologies and reliable under institutional shift. While Vision Transformers provide strong visual representations, many clinically significant findings are defined by quantitative imaging biomarkers rather than appearance alone. We introduce JANUS, a physiology-guided dual-stream architecture that conditions visual embeddings on macro-radiomic priors via Anatomically Guided Gating. On the MERLIN test set (N=5082), JANUS attains macro-AUROC 0.88 and AUPRC 0.74, outperforming all reproduced baselines. It generalizes to an external dataset N=2000; AUROC 0.87), with the largest gains on findings defined by size and attenuation as well as improved calibration on both datasets. We further quantify prediction suppression using the Physiological Veto Rate (PVR), showing that under domain shift JANUS reduces high-confidence false positives substantially more often than true positives. Together, these results are consistent with physically grounded conditioning that improves both discrimination and reliability in CT triage. Code is made publicly available at github repository https://github.com/lavsendahal/janus and model weights are at https://huggingface.co/lavsendahal/janus.

Problem

• CT triage
• distribution shift
• robustness
• automated diagnosis
• domain generalization

Innovation

• Anatomically Guided Gating
• Radiomic Priors
• Distribution Shift
• Physiological Veto Rate
• Dual-Stream Architecture