Scanner influence on the mechanical response of QCT-based finite element analysis of long bones

Patient-specific QCT-based finite element (QCTFE) analyses enable highly accurate quantification of bone strength. We evaluated CT scanner influence on QCTFE models of long bones. A femur, humerus, and proximal femur without the head were scanned with K₂HPO₄ phantoms by seven CT scanners (four models) using typical clinical protocols. QCTFE models were constructed. The geometrical dimensions, as well as the QCT-values expressed in Hounsfield unit (HU) distribution was compared. Principal strains at representative regions of interest (ROIs), and maximum principal strains (associated with fracture risk) were compared. Intraclass correlation coefficients (ICCs) were calculated to evaluate strain prediction reliability for different scanners. Repeatability was examined by scanning the femur twice and comparing resulting QCTFE models. Maximum difference in geometry was 2.3%. HU histograms before phantom calibration showed wide variation between QCT scans; however, bone density histogram variability was reduced after calibration and algorithmic manipulation. Relative standard deviation (RSD) in principal strains at ROIs was <10.7%. ICC estimates between scanners were >0.9. Fracture-associated strain had 6.7%, 8.1%, and 13.3% maximum RSD for the femur, humerus, and proximal femur, respectively. The difference in maximum strain location was <2 mm. The average difference with repeat scans was 2.7%. Quantification of strain differences showed mean RSD bounded by ∼6% in ROIs. Fracture-associated strains in "regular" bones showed a mean RSD bounded by ∼8%. Strains were obtained within a ±10% difference relative to the mean; thus, in a longitudinal study only changes larger than 20% in the principal strains may be significant. ICCs indicated high reliability of QCTFE models derived from different scanners.