Validation of an elastic matching algorithm based on complex wavelets for the realignment of dynamic MR breast images

In this paper, a novel approach for the registration of dynamic magnetic resonance (MR) breast images is validated. The approach is fully automatic and it performs a multi-resolution motion field estimation based on complex discrete wavelets transform (CDWT). The validation was designed to assess the registration quality in patient breast data. A set of diagnostic features of MR breast image was defined and two expert radiologists were asked to score the capability to detect these features in the subtraction images. Two registration methods were compared: a rigid registration algorithm and the proposed approach. Results show that the novel approach is superior in terms of both lesion detectability and lesion definition.