Fuzzy Object Shape Model for Newborn Brain MR Image Segmentation

In magnetic resonance (MR) images, finding a small change of parenchyma in newborn babies' brain significantly helps physicians to diagnose suspicious hypoxic-ischemic encephalopathy patients. However, there are no computer-aided methods because an automated segmentation algorithm has not been established yet. This paper proposes a new image segmentation method for parenchyma segmentation in T2-weighted MR images. The proposed method introduces a fuzzy object model, which has a fuzzy boundary and MR signal learned from training data. It segments the parenchyma by maximizing a fuzzy degree of deformable surface model. The fuzzy degree is estimated by using the fuzzy object model. To validate the proposed method, we recruited 12 newborn babies whose revised ages were -1 month to 1 month. 9 subjects were used to generate the fuzzy object model, and the remained subjects were used to evaluate the segmentation accuracy. The segmentation accuracy has been evaluated by using sensitivity and false-positive ratio, which were calculated by comparing delineation result (ground truth).