Abstract
Neonatal cerebral disorders such as hypoxic-ischemic encephalopathy might deform brain shape, and reduce cerebral function. For cerebral disorders diagnosis, it is effective to measure cerebral volume and surface area using head magnetic resonance (MR) image. The measurement requires a brain segmentation process. However, an automated segmentation algorithm has not been established. This study proposes a new brain segmentation method in newborn head MR images. The proposed method uses a fuzzy object shape model, which is produced from some learning datasets. It segments the brain region by maximizing a fuzzy degree of a fuzzy deformable contour model based on the fuzzy object shape model and MR signal. The fuzzy degree is estimated by using expert knowledge of the brain MR images. In order to validate segmentation accuracy of the proposed method, we applied the proposed method to 12 newborn subjects. Subject's revised ages were between -1 month and 1 month. In 12 subjects, 9 subjects were used for creation of the fuzzy object shape model. And, the remained subjects were used for evaluation. The segmentation accuracy has been evaluated by using sensitivity and false-positive ratio, which were calculated by comparing with delineation result (ground truth).
