Atlas selection strategy for automatic segmentation of pediatric brain MRIs into 83 ROIs

Registration algorithms can facilitate the automatic anatomical segmentation of pediatric brain MR data sets when segmentation priors (atlases) are in hand. Automatic segmentation can be achieved through label propagation and label fusion in target space. We investigated the performance of different...

Full description

Saved in:
Bibliographic Details
Main Authors: Gousias, I S, Hammers, A, Heckemann, R A, Counsell, S J, Dyet, L E, Boardman, J P, Edwards, A D, Rueckert, D
Format: Conference Proceeding
Language:English
Subjects:
Biomedical imaging
Brain
brain atlases
brain segmentation
Educational institutions
Image databases
Image segmentation
Inspection
Magnetic resonance imaging
non-rigid registration
parameters
Pediatric brain atlasing
Pediatrics
priors
Tellurium
Testing
validation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Registration algorithms can facilitate the automatic anatomical segmentation of pediatric brain MR data sets when segmentation priors (atlases) are in hand. Automatic segmentation can be achieved through label propagation and label fusion in target space. We investigated the performance of different age cohorts used as prior atlases for the segmentation of 13 MRIs of 1-year-olds. Thirty adults and 33 2-year-olds (including the 13 1-year olds, scanned a year later) served as priors for label propagation and fusion. In addition, we tested the accuracy of a single propagation step of the atlas of the same subject scanned at 2 years of age. Pediatric priors performed better than adult priors on visual inspection as well as manual validation of the caudate nucleus (Dice index = 0.89 ± 0.02 vs. 0.86 ± 0.03). Corresponding single atlases at the age of 2 performed better than the fusion of 30 adult priors (83 ROIs / average Dice=0.87 ± 0.05 vs. 0.84 ± 0.07).
ISSN:1558-2809
2832-4242
DOI:10.1109/IST.2010.5548493