Three-dimensional magnetic resonance imaging-based registration techniques and statistical shape analysis for knee osteoarthritis

Abstract

The efficacy of three-dimensional (3D) magnetic resonance imaging (MRI)-based registration techniques in femur models of osteoarthritis (OA) with respect to OA severity was investigated in this study. MRI data of 58 OA femurs (23 Kellgren–Lawrence grade 2, 20 grade 3, and 15 grade 4) and 31 normal femurs were analyzed. Distal femurs were segmented and converted into 3D reconstructed models. Several registration techniques (fiducial registration and automated landmarking using point-cloud alignment and correspondence analysis [ALPACA]), were applied to OA femur models. Fit quality and volume differences between the reference and OA models were assessed with respect to OA severity. Generalized Procrustes analysis (GPA) and principal component analysis (PCA) explored important bone-shape features of OA femurs. Deformable ALPACA registration exhibited the best fit. Significant differences were observed in the quality of fit of our techniques and volume differences between the reference and OA models in the OA severity groups. The mean OA model demonstrated bony enlargement at the edges of the cartilage plate in 3D statistical shape analysis (SSA). This shape variation was a major component associated with OA severity in the GPA-aligned PCA. This novel 3D MRI-based registration technique and SSA is useful to differentiate OA severity grades.