DiffRGenNet: Difference-aware Medical Report Generation

Minghao Bian, Kun Zhang, Dexin Zhao, S Kevin Zhou

Keywords: Report Generation, Multimodal Learning.

TL;DR: Difference-aware Medical Report Generation

Abstract:

Medical report generation is a critical task in healthcare, aiming to automatically pro-
duce accurate diagnostic reports from medical images, thereby alleviating the burden on
radiologists. However, due to the high similarity among medical images of the same anatom-
ical region and the substantial variations captured from the same region across different time
points for individual patients, capturing these differences poses a significant challenge. We
propose a Difference-aware Report Generation Network (DiffRGenNet), which retrieves
similar reports through image search, identifies differences using the Feature Diff module,
and dynamically orchestrates global and local dependencies via the FlexiRoute Aggregation
Module to determine the optimal routing path for each sample, selecting the most suitable
report to describe the variations and connections. Finally, by leveraging the consistency
of classification information and the discrepancy information from the diff module, DiffR-
GenNet enhances the ability to learn differences in rare diseases, generating more precise
reports. Experiments demonstrate that DiffRGenNet outperforms existing methods on the
MIMIC-CXR and IU X-Ray datasets, confirming its effectiveness and potential.

Primary Subject Area: Generative Models

Secondary Subject Area: Detection and Diagnosis

Paper Type: Methodological Development

Registration Requirement: Yes

Reproducibility: https://github.com/1benv/DiffRGenNet

Visa & Travel: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 106