Figure 1.

Registration process with the different steps that were taken, starting with the shapes in the dataset towards the principal component analysis. (a) Template initialization for the non-rigid ICP algorithm. This template was obtained by first calculating the mean shape from the shapes in the dataset. The patient in the dataset who resembles the mean shape best was chosen as the template. (b) Registration process. All the hemipelves in the dataset were initially translated to the same position, after which affine registration and non-rigid ICP were performed, resulting in registered vertices. These were rescaled to the original size and used as input for the principal component analysis (PCA).