Learning Directional Semantic Transitions for Longitudinal Chest X-ray Analysis

2026-06-14

AI Summary

Existing methods for longitudinal chest X-ray analysis struggle to capture subtle semantic changes associated with disease progression and often overlook the inherent directionality of such changes. To address this, this work proposes ProTrans, a novel framework that formalizes disease progression as a directional semantic transition between paired chest X-rays. ProTrans anchors interpretable disease states using radiology reports, introduces learnable progression feature maps to model semantic evolution between states, and incorporates reverse-temporal modeling with bidirectional reconstruction consistency constraints to disentangle directional semantics. Built upon vision–language pretraining, the framework enables report-guided semantic alignment and significantly outperforms existing approaches in tasks such as disease progression classification and descriptive report generation, establishing the first unified pretraining paradigm for longitudinal chest X-ray understanding.

Abstract

Chest X-ray (CXR) interpretation often requires longitudinal comparison to assess disease progression. Existing approaches typically rely on temporal feature fusion or inter-study discrepancy modeling, yet remain limited in capturing subtle progression semantics and overlook the inherently directional nature of disease trajectories. In this paper, we propose ProTrans, a novel vision-language pretraining framework that formulates disease progression as a directional semantic transition between paired CXR studies. ProTrans leverages radiology reports to anchor individual CXR representations within interpretable disease states, and introduces a learnable progression feature map to explicitly encode semantic shifts between states, aligned with report-derived progression descriptions. To enforce direction-aware perception, ProTrans incorporates a reversed temporal modeling process and imposes bidirectional reconstruction consistency across states and transitions, thereby disentangling directional semantics and promoting coherent trajectory modeling. Extensive experiments on longitudinal downstream tasks, including disease progression classification and progression captioning, demonstrate that ProTrans consistently outperforms existing methods, establishing a unified pretraining framework for longitudinal CXR understanding. https://github.com/RPIDIAL/ProTrans

Problem

• longitudinal CXR analysis
• disease progression
• directional semantics
• semantic transitions
• chest X-ray interpretation

Innovation

• directional semantic transition
• vision-language pretraining
• longitudinal CXR analysis
• progression feature map
• bidirectional reconstruction consistency