Estimating Bone Mineral Density Based on Age, Sex, and Anthropometric Measurements.

Osteoporosis is a global health problem characterized by low bone density and deterioration of bone tissue that increases the risk of fracture. Early identification of low bone mineral density (BMD) is crucial to reducing risks by providing correct treatment or prevention methods. Dual-energy x-ray absorptiometry (DXA) is often used to measure BMD. However, it is not affordable or accessible to some patients and is rarely suggested to non-risk groups. Alternatively, information such as age, sex, weight, height, and body circumferences have shown an association with BMD and are inexpensive and easy to obtain. Thus, this paper proposes a method to estimate BMD through anthropometric measurements, age, and sex. We also introduce BMD-10, a dataset containing 911 patients with their respective BMD values and 10 other features. Our approach evaluates the performance of different types of regression algorithms through nested cross-validation. A Least-Squares Support-Vector Machine achieves the lowest Mean Absolute Error: 0.0769 g/cm(2). Lastly, we interpret the model predictions with SHAP (SHapley Additive exPlanations), finding that weight is the most important feature for the estimation.