Spatio-Temporal Feature Extraction For Differentiation Of Non-Mass-Enhancing Lesions In Breast Mri

Spatio-temporal feature extraction represents a challenge however critical step for the differential diagnosis of non-mass-enhancing lesions. The atypical dynamical behavior of these lesions paired with non well-defined tumor borders requires novel approaches to obtain representative features for a subsequent automated diagnosis. We evaluate the performance of mappings of pixelwise kinetic features within a tumor, morphological descriptors based on Minkowski functionals and a novel technique, the Zernike velocity moments, to capture the joint spatio-temporal behavior within an image sequence. The highest sensitivity is achieved by the Zernike velocity moments proving thus that dynamical and morphological behavior can not be separately analyzed based on features extracted only for a distinct behavior or as a feature combination of these two but has to be a simultaneous measure of these. The present paper provides the most detailed automated diagnosis of non-mass-enhancing lesions so far in the literature.