Home  Details  Download  Rules  Evaluation  Reference  Submit  Results  Organizers 

Manual segmentation protocol

Selected scans were chosen from two sets of images of premature infants scanned at term equivalent age, and one set of images of preterm infants scanned at a gestational age around 30 weeks. No brain pathology was visible on MRI, additional 15 months follow-up corrected for gestational age was normal according to Griffith's assessment test. All scans were acquired on a Philips 3T system as follows: Images for the NeoBrainS12 were obtained from the scanner without any post-processing. Dicom format was transformed to mhd+raw data.

The following tissues classes were segmented: cortical grey matter (CoGM), unmyelinated white matter (UWM), brainstem, cerebellum, ventricles, cerebrospinal fluid in the extracerebral space (CSF), basal ganglia and thalami, myelinated white matter (MWM).
Each tissue class was segmented separately. Manual (reference) segmentations have been defined on T2-weighted images. They were generated using an in-house developed user interface by mouse painting of brain voxels in all slices. Each voxel was assigned to only one of the eight tissue classes. The labeling was indicated by color overlay where each tissue class was represented by one color. During manual segmentation, observers were free to adjust window and level settings and use zoom-in and zoom-out functions.

When segmenting CoGM, UWM and CSF, the outer border of each tissue was carefully identified, while the inner border was over-segmented. For example, when segmenting CoGM, the outer border was carefully identified along the CSF border, while CoGM was over-segmented on the inner border (with UWM). When segmenting UWM, its outer border, thus the voxels touching CoGM were carefully identified. Thereafter, final segmentation of CoGM was achieved by subtracting voxels identified as UWM from those defined as CoGM. Similarly, CSF was finally defined by subtracting segmentation of CoGM from it. The inner border of UWM was defined by basal ganglia or ventricles or MWM. Therefore, its final segmentation was obtained by subtracting manual segmentation of ventricles, basal ganglia and MWM from its initial manual segmentation.
After all tissues classes had been segmented, the individual segmentations were added to form a total brain segmentation. This image was formed to make sure all voxels were assigned to a single tissue type and no unassigned brain voxels remained. Manual segmentations were corrected accordingly, if needed.

The segmentations were performed as follows:
Large vessels (>2-3mm) such as basil artery or large sinuses, were not classified as such but were defined as background. Small vessels were included in the tissue where they were situated.

The segmentations were performed either by MDs who were working towards a PhD in neonatology, or by trained medical students. The segmentations were verified independently by three neonatologists with each at least seven years of experience in reading neonatal MRI scans. In case of disagreement, the decision on the segmentation was made in a consensus meeting.


Training data

Images of two patients scanned axially at 40 weeks corrected age and images of two patients scanned coronally acquired at 30 weeks corrected age are available for training. For each set T1-weighted and T2-weighted images as well as the corresponding reference segmentations are available.
To receive the data you need to sign a confidentiality agreement (available here). Please mail the signed agreement to neobrains12@isi.uu.nl, and you will receive instructions on how to obtain the data. We will use the email address you listed when registering your team.


Example segmentations

T2 Image T2 with segmentation overlayed

First     Previous

Next     Last 

T2 Image T2 with segmentation overlayed

First     Previous

Next     Last 



This website is copyright (c) 2012 Image Sciences Institute.