Improving Diffusion Tensor Imaging Segmentation Through an Adaptive Distance Learning Scheme

Paulo Reis Rodrigues¹, Anna Vilanova¹, Thorsten Twellmann², Bart M. ter Haar Romeny¹

Similarity of diffusion tensors is a crucial point in several applications like segmentation, group statistical analysis, etc. The selection of the most suitable measure, for a given task, is not always clear and often done by trial and error. We present a proof of concept of an initially flexible learning scheme that infers the measure or combination of measures that achieves the best discrimination between a user selected Region of Interest and a representative set of tensors of the whole volume. The results demonstrate the methods potential to infer the ideal parameters for a task specific segmentation algorithm, region growing.