Abstract

We present an automated image-to-mesh workflow that meshes the cortical surfaces of the human femur, from clinical CT images. A piecewise parametric mesh of the femoral surface is customized to the in-image femoral surface by an active shape model. Then, by using this mesh as a first approximation, we segment cortical surfaces via a model of cortical morphology and imaging characteristics. The mesh is then customized further to represent the segmented inner and outer cortical surfaces. We validate the accuracy of the resulting meshes against an established semi-automated method. Root mean square error for the inner and outer cortical meshes were 0.74 mm and 0.89 mm, respectively. Mean mesh thickness absolute error was 0.03 mm with a standard deviation of 0.60 mm. The proposed method produces meshes that are correspondent across subjects, making it suitable for automatic collection of cortical geometry for statistical shape analysis.

Authors and Affiliations

Auckland Bioengineering Institute, The University of Auckland, New Zealand

Ju Zhang, Duane Malcolm & Poul Nielsen

Clinical Applications Research Center, Toshiba Medical, Sydney, Australia

Jacqui Hislop-Jambrich

The Melbourne Dental School, The University of Melbourne, Victoria, Australia

C. David L. Thomas

Department of Engineering Science, The University of Auckland, New Zealand

Poul Nielsen

Editors and Affiliations

Visual Computing Division, School of Computing, Clemson University, 100 Mc Adams Hall, 29634, Clemson, South Carolina, USA

Joshua A. Levine

DTU Informatics, Technical University of Denmark, Richard Petersens Plads, 2800, Kgs. Lyngby, Denmark

Rasmus R. Paulsen

Department of Mechanical Engineering, Carnegie Mellon University, 303 Scaife Hall, 5000 Forbes Avenue, 15213, Pittsburgh, Pennsylvania, USA

Yongjie Zhang