Chest-OMDL: Organ-specific Multidisease Detection and Localization in Chest Computed Tomography using Weakly Supervised Deep Learning from Free-text Radiology Report

Xuguang Bai, Mingxuan Liu, Yifei Chen, Hongjia Yang, Qiyuan Tian

Keywords: Radiology report, Chest computed tomography, Weakly supervised learning, Multidisease detection

TL;DR: Chest-OMDL is a weakly supervised framework for chest CT that reduces annotation costs in multidisease detection and localization.

Abstract: Deep learning (DL) models designed to detect abnormalities in chest computed tomography (CT) reduce radiologists’ workload. However, training multidisease diagnostic models requires large expert-annotated datasets, significantly increasing model development cost. To address this challenge, we propose a weakly supervised learning (WSL) framework entitled Chest-OMDL for Organ-specific Multidisease Detection and Localization in chest CT. Chest-OMDL trains DL models using disease labels extracted by RadBERT from free-text radiology reports and multi-organ segmentation masks generated by the Segment Anything by Text (SAT) model, therefore reducing the need for manual annotation. Specifically, Chest-OMDL employs a Y-shaped Mamba model (Y-Mamba), comprising a feature extractor, an organ segmentation decoder, and a disease anomaly map generator. By incorporating multidisease anatomical knowledge, Y-Mamba is trained with a multi-task loss for organ-level weak supervision. Chest-OMDL was trained and validated on the large-scale CT-RATE dataset (25,692 non-contrast 3D chest CT scans from 21,304 patients) and tested on the external RAD-ChestCT dataset (3,630 scans), outperforming CT-CLIP (contrastive language-image pre-training) and CT-Net (full supervision). Code: \url{https://github.com/JasonW375/Chest-OMDL}

Primary Subject Area: Detection and Diagnosis

Secondary Subject Area: Application: Radiology

Paper Type: Both

Registration Requirement: Yes

Reproducibility: https://github.com/JasonW375/Chest-OMDL

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: 37