Selected scans for training and testing were randomly chosen from the pilot study of Baby Connectome Project (BCP). All scans were acquired on a Siemens head-only 3 T scanners with a circular polarized head coil. During the scan, infants were asleep, unsedated, fitted with ear protection, and their heads were secured in a vacuum-fixation device.

  • T1-weighted images were acquired with 144 sagittal slices using parameters: TR/TE = 1900/4.38 ms, flip angle = 7º, resolution = 1×1×1 mm3;
  • T2-weighted images were obtained with 64 axial slices: TR/TE = 7380/119 ms, flip angle = 150º, resolution =1.25×1.25×1.95 mm3.

For image preprocessing, T2 images were linearly aligned onto their corresponding T1 images. All images were resampled into an isotropic 1 × 1 × 1 mm3 resolution. Next, standard image preprocessing steps were performed before manual segmentation, including skull stripping, intensity inhomogeneity correction, and removal of the cerebellum and brain stem by using in-house tools. The preprocessing was conducted to maximally eliminate the influences of different image registration and bias correction algorithms on infant brain segmentation.

To generate reliable manual segmentation, we first took advantage of the follow-up 24-month scans with high tissue contrast to generate an initial automatic segmentation for 6-month subjects [1,2] by using a publicly available software iBEAT (http://www.nitrc.org/projects/ibeat/). Second, based on the obtained initial automatic segmentation, manual editing was performed, under the guidance of an experienced neuroradiologist (Dr. Valerie Jewells, UNC), to correct segmentation errors (based on both T1- and T2-weighted MR images) and geometric defects by using ITK-SNAP, with the help of surface rendering. For example, if there is a hole/handle in the surface, the neuroradiologist will first localize the related slices, and then check the segmentation maps of both T1 and T2 images to determine whether to fill the hole or cut the handle. Generally, it took almost one week for correcting one subject.

[1]. Li Wang, Feng Shi, Pew-Thian Yap, John H. Gilmore, Weili Lin, Dinggang Shen. 4D Multi-Modality Tissue Segmentation of Serial Infant Images, PLOS ONE, 7(9), e44596, 2012.
[2]. Li Wang, Feng Shi, Pew-Thian Yap, Weili Lin, John H. Gilmore, Dinggang Shen. Longitudinally guided level sets for consistent tissue segmentation of neonates, Human Brain Mapping, 34(4), 956-972, 2013.