Automatic full femur segmentation from computed tomography datasets using an atlas-based approach

Abstract

Automatic segmentation of femurs in clinical computed tomography remains a challenge. Joints degraded by old age are a particularly challenging dataset to segment. The objective of this study is to evaluate existing methods and propose an alternative method for segmentation of femurs in clinical computed tomography datasets for joints degraded by old age. Bilateral hip computed tomography scans of three cadaveric specimens (six femurs) were available for this study. Deformable registration using an affine selection criterion was used for atlas-based segmentation. For comparison, the six femurs were also segmented with two graph-cut algorithms. An automatic graph-cut segmentation algorithm was only able to separate the femur from the pelvis in two of the six femurs due to a limitation of graph-cuts. The atlas-based method produced consistent automatic segmentations for all degraded joints. In conclusion, atlas-based femur segmentation performs considerably better than an automatic graph-cut algorithm when applied to degraded joints.