A Convolutional Wasserstein Distance for Tractography Evaluation: Complementarity Study to State-of-the-Art Measures

Evaluation and comparison of tractograms are crucial and open problems that need to be solved in order to evaluate the false positives rate and variability of tractography algorithms. In this context, a lot of measures have been developed and are typically used to judge the quality of a tractogram. They however do not rely on the same quantities extracted from the tractograms to compare and may thus not evaluate the same aspects of tractogram quality. To evaluate this aspect and the measures redundancy or complementarity, we perform a quantitative analysis of the most common ones, both on simulated data and in real circumstances. We also propose a new evaluation measure based on optimal transport theory. We show that, when used in conjunction, these measures can provide a more in depth comparison of tractograms and thus a more complete evaluation.