Early detection of osteoporosis is becoming imperative in an aging society, as low bone mineral density (BMD) poses an elevated risk of bone injury. We develop an optical bone densitometer (OBD) combined with deep learning to predict BMD in specific regions of human body. The OBD utilized a near-infrared light source to obtain photos with optical information from distal radius by emitting the light through the wrist, the relatively thin region of human body.To precisely capture the position of wrist, we employed U-net for biomedical image segmentation, which generates a mask for the original wrist image, leaving behind images without background noise for the subsequent deep learning analysis. The algorithm considers the preprocessed images and various physiological parameters to predict BMD in the target regions, thus providing a reliable result for both orthopedic surgeons and patients.
|