OPXrayNet: a high-precision, explainable AI framework for transformative opportunistic osteoporosis screening utilizing multi-site X-ray imaging.
Authors
Affiliations (11)
Affiliations (11)
- Caoxian People's Hospital, Heze, China.
- School of Medicine at Nankai University, Tianjin, China.
- Quzhou People's Hospital, Quzhou, China.
- Xinjiang Medical University Affiliated First Hospital, Urumqi, China.
- School of Medicine at Nankai University, Tianjin, China. [email protected].
- Department of Orthopedic Surgery, Hebei Medical University Third Hospital, Shijiazhuang, China. [email protected].
- Key Laboratory of Biomechanics of Hebei Province and the Orthopaedic Research Institute of Hebei Province, Shijiazhuang, China. [email protected].
- National Health Commission Key Laboratory of Intelligent Orthopaedic Equipment at the Third Hospital of Hebei Medical University, Shijiazhuang, China. [email protected].
- Chinese Academy of Engineering, Beijing, China. [email protected].
- Xinjiang Medical University Affiliated First Hospital, Urumqi, China. [email protected].
- Caoxian People's Hospital, Heze, China. [email protected].
Abstract
This study aimed to develop and validate OPXrayNet, an automated deep-learning framework for opportunistic osteoporosis (OP) screening using multi-site skeletal X-ray images, to address the limitations of single-site analysis and poor interpretability in existing methods. This retrospective, multicenter diagnostic study utilized 16,347 annotated X-ray images from 2186 patients. The reference standard was OP diagnosis based on DXA T-score (≤ -2.5). OPXrayNet integrated YOLOv8 for automated region-of-interest detection, a spatial-frequency joint focusing architecture for feature enhancement, a feature pyramid network for multi-scale fusion, and an expert-guided loss module. Diagnostic accuracy was evaluated using accuracy and AUC, comparing performance against models including DeepSeek-VL2 and DCNN via statistical tests on internal and external validation sets. The study included 2186 participants (mean age 69.0 ± 10.5 years; 72.3% female). On the internal test set (n = 3892), OPXrayNet achieved an accuracy of 80.1%; observed accuracy values for baseline models were lower. On the external validation set, it attained an AUC of 0.844 (95% CI: 0.821-0.867); a numerical difference of 3.9% was observed when compared with other models. Feature heatmaps suggest that the model focuses on clinically critical bone structures. OPXrayNet effectively provides accurate, automated, and interpretable opportunistic OP screening from multi-site X-ray images, showing promising performance for scalable clinical application in primary care settings. Question There is a notable absence of automated, interpretable, and highly accurate deep learning methodologies for conducting opportunistic OP screening using multi-site X-ray images. Findings The OPXrayNet model achieved an accuracy of 80.1% and an AUC of 0.844 in automated OP screening across multi-site X-rays in internal and external validation cohorts; numerical differences in performance metrics were observed when compared with baseline models. Clinical relevance This framework offers an accurate, automated, and interpretable screening tool for OP using routine X-rays, presenting substantial potential for scalable clinical application in primary care. It enables early detection without incurring additional costs or radiation exposure.