Chest X-ray report generation improves by comparing with prior images

CPVB-RG: Cross-Prior Visual Bridging for Prior-Aware Chest X-Ray Report Generation

Computer Vision and Pattern Recognition

Summary

It is important for doctors to spot changes in chest X-rays by comparing current images with earlier ones to understand how a patient’s condition is evolving. Most computer systems that create X-ray reports either ignore earlier images or mix their information without telling what stayed the same and what changed. The authors propose a new method called CPVB-RG that focuses on highlighting differences between current and prior chest X-rays while keeping track of stable areas. This approach also uses medical text information and generates more accurate reports. Tests show this method better captures important changes and produces clearer reports than previous systems.

What this means in practice

  • For radiology software developers: Integrate method to enhance automatic chest X-ray report generation by distinguishing new and resolved findings from stable anatomy using prior images.
  • For hospital data teams: Use improved automated reports that track longitudinal changes in patient X-rays to support radiologists and reduce report errors.

Authors

Sike Xiang, Shuang Chen, Julie Cox, Amir Atapour-Abarghouei

Abstract

Radiologists interpret follow-up chest X-rays by comparing current and prior examinations to identify new, resolved or progressing findings. Most report generation systems either omit this longitudinal context or fuse prior features without distinguishing interval changes from anatomy that remains stable over time. We propose CPVB-RG, a prior-aware framework that combines current and prior radiographs with available clinical text. Its Cross-Prior Visual Bridge (CPVB) contains Progression-Guided Focus Attention (PGFA) and two Cross-Attention units. PGFA is designed to contrast progression-difference attention with shared-spatial attention, aiming to reduce invariant responses whilst retaining coarse spatial correspondence between examinations. The Cross-Attention units then connect the resulting longitudinal visual evidence with clinical semantics. CPVB-RG is trained in two stages, using prior-guided image-text alignment followed by autoregressive report generation with a lightweight language decoder. Experiments on MIMIC-CXR and IU X-Ray show consistent improvements across language and clinical metrics. On MIMIC-CXR, CPVB-RG achieves 0.2847 F1RadGraph and 0.6692 VLScore, with statistically significant gains over the closest decoder-matched prior-aware baseline. These results support the effectiveness of the proposed difference-aware and spatially guided fusion design for prior-aware chest X-ray report generation.
View PDFOpen arXiv