AI-derived opportunistic screening using chest radiographs identifies age- and sex-associated bone mineral density patterns and supports earlier osteoporosis evaluation.
Authors
Affiliations (6)
Affiliations (6)
- Sinwu Branch, Taoyuan General Hospital, Ministry of Health and Welfare, Taoyuan City, Taiwan.
- Department of Orthopaedic Surgery, Taoyuan General Hospital, Ministry of Health and Welfare, No. 1492, Zhongshan Rd., Taoyuan Dist., Taoyuan City, 330215, Taiwan (R.O.C.).
- Sinwu Branch, Taoyuan General Hospital, Ministry of Health and Welfare, Taoyuan City, Taiwan. [email protected].
- Department of Orthopaedic Surgery, Taoyuan General Hospital, Ministry of Health and Welfare, No. 1492, Zhongshan Rd., Taoyuan Dist., Taoyuan City, 330215, Taiwan (R.O.C.). [email protected].
- Department of Biomedical Engineering, Chung Yuan Christian University, Taoyuan City, Taiwan. [email protected].
- Department of Medical Science, National Tsing Hua University, Hsinchu, Taiwan. [email protected].
Abstract
AI-derived opportunistic screening using routine chest radiographs was evaluated in 6,028 adults and internally validated against DXA in a subset of participants. AI-derived T-scores identified marked age- and sex-associated skeletal differences, particularly among women aged 50-59 years, a population often underrepresented in current DXA-based screening strategies. Osteoporosis is associated with substantial fracture-related morbidity, yet many individuals with low bone mineral density (BMD) remain undetected during midlife, when routine dual-energy X-ray absorptiometry (DXA) is not commonly performed. We evaluated whether AI-derived opportunistic screening using routine chest radiographs could identify age-associated skeletal patterns and potential high-risk periods for osteoporosis evaluation. In this retrospective cross-sectional study, 6,028 adults aged ≥ 50 years underwent chest radiographs with AI-derived lumbar spine BMD estimation using a validated deep learning model. Participants were stratified by sex and 5-year age cohorts to assess age-associated differences in AI-derived T-scores and osteoporosis prevalence. In an internal validation subset (n = 446), AI-derived T-scores were compared with DXA measurements using correlation, diagnostic performance, and Bland-Altman agreement analyses. The mean AI-derived T-score was - 1.64 ± 1.08; women exhibited significantly lower T-scores than men (- 2.15 ± 0.97 vs. - 1.19 ± 0.96; P < 0.001). Lower mean T-scores and higher osteoporosis prevalence were observed with advancing age in both sexes. Osteoporosis prevalence increased from 10.3% (ages 50-54) to 67.0% (≥ 80) in women, compared to 5.0% and 24.6% in men. A marked between-group difference in mean T-scores was observed in women between the 50-54 and 55-59 age groups, identifying a potentially important early postmenopausal age range. Female sex was independently associated with osteoporosis (OR 7.04; 95% CI 5.22-9.50). In the validation subset, AI-derived T-scores demonstrated strong correlation with lumbar spine DXA T-scores (r = 0.859) and moderate-to-strong correlation with total hip (r = 0.730) and femoral neck (r = 0.635) T-scores (all P < 0.001). AI-derived opportunistic screening using chest radiographs identified marked age- and sex-associated differences in skeletal status, particularly among women in the early postmenopausal age range. These findings suggest that AI-derived chest radiograph assessment may help identify individuals who could benefit from further osteoporosis evaluation. Prospective studies are needed to validate potential screening thresholds and clinical implementation strategies.