Longitudinal Diffeomorphic Fetal Brain Atlas Learning For Tissue Labeling Using Geodesic Regression And Graph Cuts

The human brain undergoes fundamental structural changes between the second and the thirdtrimester of pregnancy [1]. The most accurate non-invasive method for observing these events todate is the (ultra) fast magnetic resonance (MR) imaging technique. It allows to image a fetus ata satisfying resolution, despite its small size or varying position [2]. A problem of MR imaging is thelack of comparability and constancy of gray-values, which are mapped according to the proton(hydrogen) concentration. It differs among patients and results in varying gray-values for varyingproton density [3]. This motivates to build a fetal brain atlas to use it as a standard space. Brainstructures can be mapped according to marked anatomical locations, to make fetal brainscomparable for studying brain development, fetal pathology locations, fetal abnormalities oranatomy. !!The aim of the work is to provide an atlas of the developing fetal brain, consisting of acontinuous, quantifiable model of brain development derived by geodesic shooting regression[4,5] and an automated labeling procedure using a graph cut based segmentation approach [7].