Can magnetic resonance imaging-derived bone models be used for accurate motion measurement with single-plane three-dimensional shape registration?

Abstract

The purpose of this study was to compare three-dimensional (3D) kinematic measurements from single-plane radiographic projections using bone models created from magnetic resonance imaging (MRI) and computed tomography (CT). MRI is attractive because there is no ionizing radiation, but geometric field distortion and poor bone contrast degrade model fidelity compared to CT. We created knee bone models of three healthy volunteers from both MRI and CT and performed three quantitative comparisons. First, differences between MRI- and CT-derived bone model surfaces were measured. Second, shape matching motion measurements were done with bone models for X-ray image sequences of a squat activity. Third, synthetic X-ray images in known poses were created and shape matching was again performed. Differences in kinematic results were quantified in terms of root mean square (RMS) error. Mean differences between CT and MRI model surfaces for the femur and tibia were -0.08 mm and -0.14 mm, respectively. There were significant differences in three of six kinematic parameters comparing matching results from MRI-derived bone models and CT-derived bone models. RMS errors for tibiofemoral poses averaged 0.74 mm for sagittal translations, 2.0 mm for mediolateral translations, and 1.4 degrees for all rotations with MRI models. Average RMS errors were 0.53 mm for sagittal translations, 1.6 mm for mediolateral translations, and 0.54 degrees for all rotations with the CT models. Single-plane X-ray imaging with model-based shape matching provides kinematic measurements with sufficient accuracy to assess knee motions using either MRI- or CT-derived bone models. However, extra care should be taken when using MRI-derived bone models because model inaccuracies will affect the quality of the shape matching results.