Mean-shifted surface curvature algorithm for automatic bone shape segmentation in orthopedic surgery planning: a sensitivity analysis.

The results of recent studies concerning statistical bone atlases and automated shape analysis are promising with a view to widening the use of surface models in orthopedic clinical practice, both in pre-operative planning and in the intra-operative stages. In this domain, automatic shape analysis is strongly advocated because it offers the opportunity to detect morphological and clinical landmarks with superior repeatability in comparison to human operators. Surface curvatures have been proposed extensively for segmentation and labeling of image and surface regions based on their appearance and shape. The surface curvature is an invariant that can be exploited for reliable detection of geometric features. In this paper, we investigate the potentiality of the algorithm termed mean-shift (MS), as applied to a non-linear combination of the minimum and maximum curvatures of a surface. We exploited a sensitivity analysis of the algorithm parameters across increasing surface resolutions. Results obtained with femur and pelvic bone surface data, reconstructed from cadaveric CT scans, demonstrated that the information content derived by the MS non-linear curvature overcomes both the mean and the Gaussian curvatures and the original non-linear curvature. By applying a threshold-based clustering algorithm to the curvature distribution, we found that the number of clusters yielded by the MS non-linear curvature is significantly lower (by a factor of up to 6) than that obtained by using the original non-linear curvature. In conclusion, this study provides valuable insights into the use of surface curvature for automatic shape analysis.