A novel method of determining bone mineral density of the knee from a pre-surgical CT scan to aid in surgical planning

Patients with reduced bone mineral density (BMD) may have poorer outcomes after cementless joint replacement surgery. BMD phantoms may be used to calibrate Hounsfield units (HU) to BMD values, but BMD assessment with phantoms is not routine prior to surgery. We describe automated BMD calibration using air, fat and an aluminum motion-detection rod in Mako robotic-assisted knee arthroplasty 3D CT image volumes. Using 6,111 CT knee arthroplasty planning image volumes, including 133 with a BMD phantom (Mindways, TX), mean BMD values were determined for air, fat, and bone equivalent rods as calibration standards. Image voxels containing air were identified using histogram analysis. The rod surface was identified using an Active Appearance Model (AAM) search. At each point on the AAM-segmented femoral bone surface, lines perpendicular to the surface (“normal rays) were sampled and gaussian mixture model histogram analysis was used to identify fat. Ten iterations of 10-fold cross-validation were performed to compare calibration using either the phantom or the air, fat, and rod standards in tibial and femoral trabecular bone. Trabecular bone BMD from phantom calibrations were highly correlated with BMD values calculated using the air, fat, and rod standards for femur and tibia (r≈0.98). Bland-Altman analysis showed negligible bias of ∼0.3 mg/cm3. The standard deviation of differences was 9.9. mg/cm3, suggesting an upper bound of precision of 9 mg/cm3 (precision error for Mindways phantom is 3.6 mg/cm3), which produces a coefficient of variation of 7.7% at mean BMD=117 mg/cm3. Accurate volumetric BMD values may be derived at the knee from CT image volumes routinely acquired before robotic-assisted arthroplasty. This information may be useful in informing surgical decisions regarding the use of cementless implants, particularly in patients at risk of osteoporosis.