Abstract
PURPOSE: To develop a clinical prediction model for the diagnosis of osteoporosis using lumbosacral X-ray images through radiomics analysis. METHODS: A total of 272 patients who underwent dual-energy X-ray absorptiometry (DXA) and lumbosacral X-ray examinations were categorized into two groups: (1) the training set (n = 191) and (2) the validation set (n = 81). Radiomic features were extracted using 3D Slicer software, and radiomic scores were calculated using the least absolute contraction and selection operator logistic regression, facilitating the generation of radiomic features. Subsequently, a clinical model, in conjunction with the radiomic features, was employed to develop a column-line diagram for the clinical and imaging feature prediction model. Performance evaluations for various models were conducted, encompassing recognition ability, accuracy, and clinical value, with the aim of identifying and optimizing prediction models. RESULTS: The 12 most optimal imaging features were identified. Upon comprehensive performance analysis across different models, the clinical and radiomics model emerged as the most effective. The training set and test set area under the curves (AUCs) were 0.818 and 0.740, respectively. Additionally, the model exhibited a sensitivity and specificity of 81.6%, 80.6% and 77.5%, 73.2%, respectively. CONCLUSION: In this study, we developed a column-line diagram that integrates clinical and radiomics feature, presenting a novel screening tool for osteoporosis in primary hospitals. This tool aims to enhance the efficiency of osteoporosis diagnosis in primary hospitals.