Back to all papers

Computed Tomography-Verified Deep Learning Model for Rib Fracture Prediction and Complication Association from Chest Radiographs.

July 22, 2026pubmed logopapers

Authors

Chao TH,Wang CC,Lin C,Lin CS,Chen TW,Lin SH,Tsai DJ

Affiliations (6)

  • Department of Orthopedic Surgery, Tri-Service General Hospital, National Defense Medical University, Taipei, Taiwan, R.O.C.
  • Medical Technology Education Center, School of Medicine, College of Medicine, National Defense Medical University, Taipei, Taiwan, R.O.C.
  • Military Digital Medical Center, Tri-Service General Hospital, National Defense Medical University, Taipei, Taiwan, R.O.C.
  • Division of Cardiology, Department of Internal Medicine, Tri-Service General Hospital, National Defense Medical University, Taipei, Taiwan, R.O.C.
  • Division of General Surgery, Department of Surgery, Tri-Service General Hospital, National Defense Medical University, Taipei City, 114, Taiwan, R.O.C.
  • Division of Nephrology, Department of Internal Medicine, Tri-Service General Hospital, National Defense Medical University, Taipei, Taiwan, R.O.C.

Abstract

Chest radiography (CXR) is the first-line imaging for suspected rib fractures but has limited sensitivity, potentially increasing the risk of thoracic complications. This study aimed to develop and validate a deep learning model (DLM) based on CT-verified labels to predict acute rib fractures on CXRs and evaluate its association with thoracic complications. This retrospective study included 14,758 patients who underwent chest CT and CXR within seven days. We used 10,818 CXRs for development, 4,377 for tuning, and 3,020 for validation. The DLM employed a Vision Transformer (ViT-B/32) architecture. Performance was assessed using AUC, sensitivity, and predictive values. Risk stratification was performed via Youden index and F-score optimization. The DLM achieved an AUC of 0.874, sensitivity of 84.9%, and negative predictive value (NPV) of 99.4% in validation. Performance remained consistent across care settings (AUCs 0.819-0.853). Subgroup analysis showed the best performance in patients < 60 years (AUC 0.934), while comorbidities were associated with lower AUCs. High-risk patients exhibited significantly higher one-month and one-year incidences of pneumothorax (17.4% and 12.5%) and pulmonary contusion (9.8% and 8.0%) compared to low-risk patients (all p < 0.001). No significant difference was observed for pneumonia association. This CT-verified DLM demonstrated high diagnostic performance for predicting acute rib fractures and effectively stratified patients by associated complication risk. This model may assist clinicians with early prediction and triage, potentially enhancing clinical efficiency and patient outcomes in trauma workflows.

Topics

Rib FracturesDeep LearningTomography, X-Ray ComputedRadiography, ThoracicPneumothoraxJournal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.