Verifying the accuracy and precision of foot bone segmentation from fan beam and cone beam computed tomography scans

Abstract

Accurate segmentation of foot and ankle bones from computed tomography scans is an essential precursor to morphological comparison studies and model-based biplanar fluoroscopic tracking. Errors in modeling the bones may propagate into coordinate system definitions or affect arthrokinematic calculations. Cone beam computed tomography scanners with sensitive detectors offer a lower dose alternative to fan beam scanners, although with signal-to-noise ratio and tissue contrast tradeoffs. The primary aims of this study were to validate the accuracy of foot bone models (distal tibia, talus, calcaneus, navicular, first metatarsal, and proximal phalanx) generated using established semi-automatic segmentation techniques relative to ground truth laser scans of the dissected bone specimens and to compare the errors in models segmented from a fan beam scanner to those from a cone beam scanner. We found excellent between and within-operator intra-class correlation scores for segmentations (all > 0.90). Voxel-based metrics revealed small but significant differences between scanner types when comparing the Dice score and precision of segmentations, with cone beam having higher scores for nearly every metric type. Surface-based error data also revealed cone beam scans to be more accurate to the actual bone morphology despite having noisier images. The small isotropic voxels of the cone beam scanner produced bone models with errors averaging 0.43 ± 0.11 mm compared to the fan beam average of 0.74 ± 0.15 mm error across all bones. The findings suggest that voxel resolution is more critical than scanner type for modeling foot bones and that these segmentation techniques may overcome the limitations in cone beam scanners. These outcomes support the growing movement towards more widespread adoption of low-dose cone beam weight-bearing computed tomography imaging for 3D modeling of foot and ankle bones for clinical and research purposes.