Jul 24, 2025
RSNA
Abstract
OSTEO
MULTINATIONAL, REAL WORLD, CONSECUTIVE DATA, EXTERNAL VALIDATION OF CHEST RADIOGRAPH-BASED AI MODEL IN OPPORTUNISTIC SCREENING OF OSTEOPOROSIS AND OSTEOPENIA
Anushree M. Burade, MBBS, Monica O. Bernardo, MD, Suman Singhal, MD,Madhuri Barnwal, Michael M. Maher, MD, FRCR, Matthias F. Froelich, MD, Parisa Kaviani,MD, Aaminah Kobeisy, MBBS, Emiliano Garza Frias, MD, Harsh Mahajan, MD, MBBS, Kabir Mahajan, MBA, Alok Kumar Singh, Antonio A. Moscatelli, MD, Friedrich Pietsch, Miso Jang,Gaeun Lee, Javier Contreras Yametti, MD, Elahe Hosseini, Shivangi Jha, Lina Karout, MD,Subba R. Digumarthy, MD, Mannudeep K.S. Kalra, MD
Title: MULTINATIONAL, REAL WORLD, CONSECUTIVE DATA, EXTERNAL VALIDATION OF CHEST RADIOGRAPH-BASED AI MODEL IN OPPORTUNISTIC SCREENING OF OSTEOPOROSIS AND OSTEOPENIA
Status: Accepted for Presentation
Activity: Scientific Presentations
Control Number: 16795
Control/Tracking Number: 2025-SP-16795-RSNA
Session Number: M3-SSCH03
Date/Time of Session: 12/1/2025 9:30:00 AM-12/1/2025 10:30:00 AM
Author Block: Anushree M. Burade, MBBS, Monica O. Bernardo, MD, Suman Singhal, MD,Madhuri Barnwal, Michael M. Maher, MD, FRCR, Matthias F. Froelich, MD, Parisa Kaviani,MD, Aaminah Kobeisy, MBBS, Emiliano Garza Frias, MD, Harsh Mahajan, MD, MBBS, Kabir Mahajan, MBA, Alok Kumar Singh, Antonio A. Moscatelli, MD, Friedrich Pietsch, Miso Jang,Gaeun Lee, Javier Contreras Yametti, MD, Elahe Hosseini, Shivangi Jha, Lina Karout, MD,Subba R. Digumarthy, MD, Mannudeep K.S. Kalra, MD
Abstract:
Purpose: To externally validate an AI tool for opportunistic osteopenia/osteoporosis detection on chest radiographs (CXR) with an international, multicenter real-world,consecutive data
Methods and Materials: The ongoing study included consecutive, real-world, 1039adult patients from US [685/1039 (65.9%)], India [251/1039 (24.1%)] and Brazil[103/1039 (9%)], who had both posterior-anterior CXRs and DEXA within 6 months. CXRswith any implanted devices, lines, tubes, prosthesis, moderate-severe scoliosis, or an interior-posterior projection were excluded. The AI tool (Promedius.Inc) processed thede-identified CXRs to output probability scores for three classes of BMD: normal,osteopenia, osteoporosis. We recorded spinal T and Z scores from DEXA reports.Statistical analyses included ROC AUC, sensitivity, specificity, PPV, NPV and accuracy along with 95% confidence interval for differentiating the three BMD classes.
Results: The mean age and SD for the cases was 65.5 ± 14.8 years (781 females, 258males). For osteoporosis vs non-osteoporosis groups, the AI model had ROC AUC of 0.836 (95% CI: 0.799 - 0.873), sensitivity of 85.7% (95% CI: 79% - 92.4%), specificity of 68.8% (95% CI: 65.8% - 71.8%), PPV of 23.6% (95% CI: 19.3% - 27.8%), NPV of 97.7% (95%CI: 96.5% - 98.8%) and accuracy of 70.5% (95% CI: 67.7% - 73.3%). For osteopenia vs normal groups, the AI model had ROC AUC of 0.719 (95% CI: 0.681 - 0.758). There was no significant difference in the model performance across the three countries.
Conclusions: Our ongoing study supports the generalizability of CXR-based BMD prediction AI model across data from the three countries with high AUC and NPV. Given the global underutilization of dedicated osteoporosis screening with DEXA, such AI-based opportunistic screening using CXRs may help bridge the diagnostic gap.
Clinical Relevance/Application: AI model can be used as an opportunistic screening tool for osteoporosis detection healthy population using CXRs.
Attachment File
Acceptance Letter ([Abstract Acceptance Letter] MGH_Multinational Study 3-class.pdf)
Abstract Summary([Abstract Summary] MGH_Multinational Study 3-class.pdf)
