Predicting the shapes of bones at a joint: application to the shoulder

Abstract

This paper presents a novel method to explore the intrinsic morphological correlation between the bones of a shoulder joint (humerus and scapula). To model this correlation, canonical correlation analysis (CCA) is used. We also propose a technique to predict a three-dimensional (3D) bone shape from its adjoining segment at a joint based on partial least squares regression (PLS). The high dimensional 3D surface information of a bone is represented by a few variables using principal component analysis, which also captures the pattern of variability of the shapes in our datasets. Our results show that the humerus set and scapula set have highly linear morphological relationship and that the correlation information can be used as a classifier. In this study, primate shoulder bone datasets were categorised into two clusters: great apes (including humans) and monkeys. A leave one out experiment was performed to test the robustness of this prediction method. The prediction behaviour using this method shows statistically significantly better results than using the mean shape from the training set.

Keywords:

• Canonical correlation analysis
• Partial least squares regression
• Principal component analysis
• Statistical shape model
• Shoulder