Segmentation of the neonatal brain for MR Images: a review

Abstract

The need for a good segmentation method for neonatal brain MR images increased over the last decade due to a significant increase of MR examinations in this group of patients. Adult brain segmentation methods are not suitable, because the intensity representation and shape of the neonatal brain are very different compared to adults. Furthermore, neonatal brain scans have a lower signal to noise and contrast to noise ratio, because of a limited scan time. Several groups of researchers designed a segmentation method which overcomes these specific challenges. This review compares twelve published neonatal segmentation methods by nine different research groups. The segmentation approaches described in these articles are compared on several subjects, including usage of atlases, segmented tissues, scan parameters, age of neonates, pre and post-processing. The results described in the article are also compared. The key problem for validation of neonatal brain segmentations is the absence of a gold standard. The ground truth described in the publications range from a single manually segmented slice to an entire manually segmented brain. The results show that the boundary of cerebrospinal fluid and cortical grey matter and the boundary of myelinated and unmyelinated white matter are the main problem areas for neonatal segmentation methods. The majority of the methods show good results, but it depends on many factors which method performs best in a clinical research environment.