Li Wang1, Feng Shi1, John H.
Gilmore2, Weili Lin3, Dinggang Shen1
1IDEA Lab, Department of
Radiology & BRIC, University of North Carolina at Chapel Hill, Chapel
Hill, NC, United States; 2Department of Psychiatry, University of North
Carolina at Chapel Hill, Chapel Hill, NC, United States; 3MRI Lab,
Department of Radiology & BRIC, University of North Carolina at Chapel
Hill, Chapel Hill, NC, United States
Accurate segmentation of neonatal brain MR images in longitudinal MRI studies plays an important role in revealing neurodevelopmental disorders. Due to poor image quality, it still remains challenging to segment neonatal brain images. Most existing methods are voxel-based and work on only the single time-point image, and thus cannot benefit from the tissue distribution information which can be provided by the late-time-point images. In this paper, we propose a novel longitudinal guided level-sets method for consistent neonatal image segmentation, by combining intensity information, atlas prior, cortical thickness, and longitudinal information into a variational framework.