Presented at the Neonatal Society 2016 Spring Meeting.
Serag A1, Wilkinson AG2, Telford EJ1, Pataky R1, Sparrow S1, Anblagan D1,4, Macnaught G3, Semple S3, Boardman JP1,4
1 MRC Centre for Reproductive Health, University of Edinburgh, UK
2 Department of Radiology, Royal Hospital for Sick Children, Edinburgh, UK
3 Clinical Research Imaging Centre, University of Edinburgh, UK
4 Centre for Clinical Brain Sciences, University of Edinburgh, UK
Background: Quantitative volumes from brain magnetic resonance imaging (MRI) acquired across different stages of development offer the possibility of new insights into life course growth trajectories, and they provide a novel approach for evaluating long-term effects of early life neuroprotective treatments. Despite the large number of existing brain MRI segmentation methods, the literature presents a clear distinction between methods developed for different stages of development (1,2), and the use of automated segmentation tools for performing studies across the life course is lacking. Aim: To develop an automatic segmentation method for human brain MRI operable across the life course.
Methods: The study included 100+ MR brain images (3) at different stages of the life course (between the ages of 0 and 71 years), including neonatal period, childhood and adulthood. Framework: A Multi-atlas label propagation framework where a number of labelled images (i.e. atlases) is selected and propagated to the target image coordinate space, and a voxel-wise label fusion approach is then used to assign a label to each voxel. Accuracy between automatic and reference segmentations was measured using the Dice coefficient. The study was supported by Theirworld and NHS Research Scotland.
Results: Quantitative analyses indicated high accuracy for all tissues and structures with an average Dice coefficient between 0.82-0.91. High accuracies obtained for white matter and gray matter with average Dice coefficient of 0.92 and 0.91, respectively. The cerebrospinal fluid had an average Dice coefficient of 0.75. We tested the reproducibility of the proposed method across different MR modalities of the newborn brain (T1- and T2-weighted), and the method provided consistent segmentation results across different MRI modalities.
Conclusion: We present a method for segmentation of brain MRI across the life course that is robust and provides accurate and consistent results across different age groups. The idea of the proposed method is generic and could be applied to different populations and imaging modalities across the life course.
Corresponding author: email@example.com
1. Serag et al., NeuroImage, 2012.
2. Zikic et al., Med Image Anal, 2014.
3. Job et al., NeuoImage, 2016.