CADDementia results have been released last week during MICCAI 2014 conference in Boston. Deformetrica was among the 5 teams selected for an oral presentation. We approached the challenge by using the new types of multivariate markers of brain atrophy that Deformetrica provided, while most of our competitors used more standard descriptors in various statistical settings. Eventually, Deformetrica is ranked 21 among 29 algorithms, with an accuracy of 49%, best being 63% in a 3 classes classification tasks (Alzheimer’s Disease, Mild Cognitive Impairments and Cognitively Normals). It is not too bad for such an “out-of-box” evaluation of Deformetrica with simple statistics. Cross-validation for setting the thresholds should have help surely (but we didn’t have time…) It is clear also that Deformetrica was among the very few methods to find discriminative features which could be interpreted…

 

 

Leave a Reply

Your email address will not be published. Required fields are marked *