Apr 10, 2025
WCO-IOF-ESCEO
Abstract
OSTEO
Osteoporosis Screening Using Chest Radiographs with a Deep Learning Model (PROS CXR: OSTEO) Among a Multi-Ethnic Population in Malaysia: An Interim Analysis
Nasir Muhammad Ridhwan ,1 Jack Kang Tan,2 Minjee Kim,3 Jeongmin Song,3 Wei-Lin Ng,1 Lee-Ling Lim2 1Department of Biomedical Imaging, Faculty of Medicine, Universiti Malaya, Kuala Lumpur, Malaysia 2Department of Medicine, Faculty of Medicine, Universiti Malaya, Kuala Lumpur, Malaysia 3Promedius Inc., 4F 35 Olympic-ro, 13 Gil, Songpa-gu, Seoul, South Korea
Objective(s): Osteoporosis, though common, is often underdiagnosed due to limited access to dual-energy X-ray absorptiometry (DXA) scans and reluctance to undergo screening among the public. A deep learning model, PROS CXR: OSTEO, has been developed to close these gaps using the chest radiograph (a cost-effective and widely accessible tool) in the South Korean population. Here, we aimed to externally validate PROS CXR: OSTEO in an urban, multi-ethnic population in Malaysia.
Materials and Methods: We recruited participants aged ≥50 years and osteoporotic treatment-naïve from Universiti Malaya Medical Centre, Kuala Lumpur, Malaysia from September 2024 to January 2025. We excluded those with either chest implants or active lung diseases. Eligible participants underwent chest radiographs (posteroanterior erect) and DXA with the bone mineral density (BMD) for lumbar spine and femur measured. We classified them into 1) osteoporosis (BMD T-score ≤ -2.5) or 2) non-osteoporosis (T-score > -2.5) using the lowest measured T-score on DXA. We evaluated the correlation between PROS CXR: OSTEO model predictions and actual BMD-derived categories for differentiating osteoporosis and non-osteoporosis. We also examined for sensitivity, specificity, and area under the curve (AUC) using a binary classification decision threshold of 0.2. Subgroup analyses by ethnicity were also conducted..
Results: This interim analysis involved 100 participants (44% Malays, 38% Chinese, and 18% Indians). On DXA, 35% of the participants were classified as having osteoporosis, 49% with osteopenia, and 16% with normal BMD. The sensitivity of PROS CXR: OSTEO was 70.6%, with a specificity of 81.8% and an AUC of 0.850. When stratified by ethnicity, the AUCs were 0.890, 0.812, and 0.850 for Malays, Chinese, and Indians, respectively.
Conclusion(s): The PROS CXR: OSTEO deep learning model shows broad applicability across diverse populations for early screening of osteoporosis.
Acknowledgments
We thank the study participants and staff at the Universiti Malaya Medical Centre for their involvement in this investigator-initiated study, funded by Promedius Inc.
Disclosures
LLL reports receiving honoraria for giving lectures from Amgen and Zuellig Pharma. Other co-authors have no conflicts of interest to disclose.
