Optical bone densitometry robust to variation of soft tissue using machine learning techniques: validation by Monte Carlo simulation

SIGNIFICANCE: To achieve early detection of osteoporosis, a simple bone densitometry method using optics was proposed. However, individual differences in soft tissue structure and optical properties can cause errors in quantitative bone densitometry. Therefore, developing optical bone densitometry that is robust to soft tissue variations is important for the early detection of osteoporosis. AIM: The purpose of this study was to develop an optical bone densitometer that is insensitive to soft tissue, using Monte Carlo simulation and machine learning techniques, and to verify its feasibility. APPROACH: We propose a method to measure spatially resolved diffuse light from three directions of the biological tissue model and used machine learning techniques to predict bone density from these data. The three directions are backward, forward, and lateral to the direction of ballistic light irradiation. The method was validated using Monte Carlo simulations using synthetic biological tissue models with 1211 different random structural and optical properties. RESULTS: The results were computed after a 10-fold cross-validation. From the simulated optical data, the machine learning model predicted bone density with a coefficient of determination of 0.760. CONCLUSIONS: The optical bone densitometry method proposed in this study was found to be robust against individual differences in soft tissue.