Automatic Registration-Based Segmentation for Neonatal Brains Using ANTs and Atropos

We implemented an automatic open-source processing pipeline for neonatal brain tissue segmentation. The framework makes use of N4 to correct bias field, the deformable registration SyN to warp a public neonatal template to the test image, and multivariate MRF-based segmentation Atropos to perform segmentation. The pipeline does not need training and runs efficiently. It can achieve satisfactory results for most classes of a neonatal brain MR image. 1 A Prior-Based Segmentation Algorithm We contribute a template-based method for the challenging problem of tissue segmentation in the neonatal brain. The task is difficult because of intensity inversion of unmyelinated white matter and low resolution (or large slice thickness) in standard neonatal T1 or T2-weighted MRI. Our framework employs SyN diffeomorphic registration and Atropos prior-based multivariate segmentation methods available within the open-source, multi-platform Advanced Normalization Tools (ANTs). SyN and Atropos are general-purpose tools, which were not designed specifically for neonatal brain processing. However, with the assistance of a neonatal brain template, the general purpose methods are adapted for a task-specific algorithm. There are four components in the proposed framework: registration, bias correction, segmentation and post-processing (see Fig. 1). 1.1 Registration Assuming there is a proper template available for each test image to be segmented, we aim to align the template with the test image and thus transform the priors associated with the template to the subject space. The registration is performed via the diffeomorphic mapping methods available within ANTs software [1], which rank among top performers in terms of registration accuracy [2]. ANTs is freely available at http://picsl.upenn.edu/ANTS/download.php. We use the neighborhood cross correlation as similarity measure for the mono-modal registration and mutual information for intra-subject affine registration between T1 and T2 weighted MR images. Registration between T1 and T2 weighted modalities enables us to employ a multivariate data-term in the segmentation 2 Jue Wu and Brian Avants Fig. 1. Overview of the proposed algorithm formulation. The brain mask from the template was transformed to subject space and helped remove non-brain tissues. The line of code for registration between template and test image is: ANTS 3 -m CC[t2template-GA.nii.gz, test_image.nii.gz, 1, 2] -r Gauss[3,0] -t SyN[0.25] -i 90x100x80 -o output.nii.gz