shape prediction improvements

[Stein]

Hi Scalismo community,

I made an algorithm to predict the shape of the distal radius, based on the proximal radius. However, my results are not sufficient and I wish to improve my results. I used 57 samples in my training set. Does anyone have any ideas or has anyone any experience in how to improve the prediction? Maybe I can get my correspondence slightly better, but other than that i'm out of ideas.

One of the best results in my database is shown in the picture below. Many others are worse: 

My method was roughly comparable to:

Mauler F, Langguth C, Schweizer A, et al. Prediction of normal bone anatomy for the planning of corrective osteotomies of malunited forearm bones using a three-dimensional statistical shape model. J Orthop Res. 2017;35(12):2630-2636. doi:10.1002/jor.23576

Kind regards,

Stein van der Heide