Investigating osteoporosis diagnosis using clinical and hip X-ray data from the Syrian population

Osteoporosis diagnosis remains challenging due to variability in imaging quality, clinical factors, and population diversity. This study evaluated osteoporosis classification using clinical data and hip radiographs (HXR) from the Syrian population, with external validation on an independent Indian dataset. Among clinical models, the neural network achieved the highest testing accuracy of 84.04% and a balanced overall performance, while random forest and ensemble classifiers demonstrated competitive sensitivity. Deep learning using ResNet50 on contrast-enhanced HXR images reached 94.12% accuracy, outperforming EfficientNet-B0 (92.76%) and YOLOv9 (85.10%). A hybrid model combining clinical and image-derived features further achieved 93.21% accuracy, confirming the diagnostic value of multimodal integration. Cross-dataset validation performed on clinical data revealed reduced performance between the Syrian and Indian cohorts, reflecting the influence of diagnostic standards and demographic variability. These findings demonstrate that combining clinical and radiographic information can enhance osteoporosis detection while underscoring the need for standardized, population-specific validation of artificial intelligence models.