Osteoporosis Prediction by Machine Learning Techniques Using Optically Measured Bone Density Data

Abstract

In order to realize the early detection of osteoporosis, the development of a device that enables simple screening is desirable. We have focused on a simple bone density evaluation method using near-infrared light and have studied its reliability. In this study, we propose a bone mineral density (BMD) prediction model, which uses optically-measured bone density data and machine learning techniques. For 177 participants, we measured with the optical bone densitometer that we have developed before and dual-energy x-ray absorptiometry (DXA). A BMD prediction model is created by using optically-measured bone density, age, weight data as features, and BMD value determined by DXA as target data. We evaluate the generalization performance of the model using cross-validation. As a result, The BMD predicted by the machine learning model shows a strong correlation with the BMD determined by DXA. Besides, the area under the curve (AUC) of the receiver operating characteristic value for evaluating the osteoporosis discrimination performance is equivalent to that of the equipment currently used for osteoporosis screening (osteoporosis: AUC = 0.856, osteopenia + osteoporosis: AUC = 0.747).