Computer-aided diagnosis of small lesions and non-masses in breast MRI

Small and non-mass-enhancing lesions are diagnostically challenging and easily missed in a routine clinical diagnosis. Compared to mass-enhancing lesions, they show fundamentally different morphologies and kinetic characteristics. To overcome these limitations an automated analysis of such tumors is proposed to determine adequate shape and dynamical descriptors in order to capture this unique behavior. In the present paper, we evaluate several morphological and kinetic features as well combinations of those as potential shape and dynamic descriptors. We will show that for both types of lesions a combination of morphological and kinetic characteristics yields the highest AUC-values compared to dynamic or shape descriptors only. This suggests that for increasing diagnostic accuracy in breast MRI spatio-temporal descriptors for these lesions need to be included in an automated computer-aided system.