Can low-frequency guided waves at the tibia paired with machine learning differentiate between healthy and osteopenic/osteoporotic subjects? A pilot study

Abstract Purpose Axial transmission quantitative acoustics (ax-QA) has shown to be a promising tool for assessing bone health and properties in a safe, inexpensive, and portable manner. This study investigated the efficacy of low-frequency ax-QA measured at the tibia, paired with a support vector machine (SVM) approach for combining multiple acoustic indicators, to diagnose osteoporosis as defined by bone mineral density. Methods This pilot study measured 41 female subjects using ax-QA (flexural mode, 3 kHz) at the tibia and using dual X-ray absorptiometry (DXA) at the lumbar spine, femoral neck, and distal radius. For each location, a threshold classifier and SVM were trained to differentiate between healthy and non-healthy subjects based on the phase velocity at different frequencies. Receiver Operating Characteristics and area under curve values (AUC) were used to assess the classifiers’ performances for various thresholds and class-weights. Results The SVM outperformed the threshold classifier for all three bone locations at low false positive rates. While differentiation between healthy and non-healthy bone states was poor for the spine (AUC: 0.56 ± 0.04), good to moderate performances were observed for the radius (AUC: 0.83 ± 0.03) and hip (AUC: 0.71 ± 0.04). Conclusions Low-frequency ax-QA has demonstrated potential for complementing DXA in screening for osteoporosis at the radius and hip. Through further addition of acoustic indicators ax-QA could provide a diagnostic alternative in third-world countries, and bring bone health screening and monitoring into the hands of clinicians and general health practitioners everywhere.