Optimizing Osteoporosis Detection: AutoML -Radiomics Approach Combining Imaging and Demographic Data

Motivation: To develop a machine learning model utilizing demographic data and lumbar fat-water MRI as an opportunistic screening for osteoporosis. Goal(s): To compare the performance of AutoML in predicting osteoporosis using demographic features, radiomics from lumbar fat and water MRI, and their combination. Approach: A TPOP-radiomics-based classification model was trained using demographic features and radiomic data from lumbar fat-water MRI, differentiating between normal and osteoporosis as identified by DEXA. Results: The combined model emerged as the most effective from the AutoML process, achieving a mean sensitivity of 0.783 and a mean specificity of 0.867 in distinguishing between normal and osteoporosis. Impact: Because osteoporosis is often considered a 'silent' disease, routine IDEAL-IQ lumbar scans have the potential to serve as an opportunistic screening tool for reducing the risk of fragility fractures, which are associated with morbidity and mortality.