Semantically Calibrated Evidence Composition for CT Vision-Language Learning

2026-07-31

AI Summary

This work addresses the challenge that existing CT-report alignment methods struggle to effectively integrate local anatomical evidence with global semantic context, often resulting in findings disconnected from overall clinical meaning. To overcome this, the authors propose SCOPE, a novel framework that employs organ-specific masks to guide query-based extraction of context-aware local evidence, which is then coupled with unconstrained global queries. The combined representation is calibrated through diagnostic summaries to yield a unified CT embedding. Notably, SCOPE achieves progressive linkage between local evidence and study-level semantics without requiring a predefined set of organs. Trained end-to-end directly on raw 3D volumes, the method attains macro AUC scores of 85.0 and 72.2 on CT-RATE and RadChestCT, respectively—surpassing current state-of-the-art performance by 7.2 and 4.2 points—and demonstrates significant gains in linear probing and cross-modal retrieval tasks.

Abstract

Learning transferable representations from CT-report pairs requires combining whole-volume context with anatomy-specific evidence. Existing methods typically emphasize either global CT-report alignment or fine-grained anatomy-level correspondence. Global alignment preserves broad study context but leaves the contribution of localized evidence implicit, whereas anatomy-level alignment explicitly grounds local findings but does not specify how independently represented evidence should interact, acquire study-level meaning, and contribute to a global CT representation. To address this gap, we propose SCOPE (Semantic Calibration Of comPosed Evidence), a framework for semantically calibrated evidence composition in CT vision-language learning. Under organ-specific report supervision, mask-guided queries with fixed anatomical identities extract context-aware organ evidence from shared, uncropped volumetric features, while an unrestricted global query retains access to whole-volume context. The global query then drives Local-Global Coupling to compose the organ evidence into a unified evidence representation. The composed evidence is subsequently calibrated using the diagnostic summary, providing study-level semantic supervision beyond local organ descriptions, and is finally integrated as a controlled residual into a context-preserving whole-volume representation aligned with the complete report. This progressive pathway connects localized evidence with study-level semantics without reducing the CT representation to a predefined set of organs. On CT-RATE and RadChestCT, SCOPE achieves macro AUCs of 85.0 and 72.2, respectively, outperforming the previous SOTA by 7.2 and 4.2, while also yielding substantial gains in linear probing and cross-modal retrieval. These results demonstrate the effectiveness of semantically calibrated evidence composition.

Problem

• CT vision-language learning
• evidence composition
• semantic calibration
• anatomy-specific evidence
• global-local representation

Innovation

• evidence composition
• semantic calibration
• vision-language learning
• local-global coupling
• CT representation