A Sparsity-Based Atlas Selection Technique For Multiple-Atlas Segmentation: Application To Neonatal Brain Labeling

Quantitative brain tissue volumes from neonatal magnetic resonance imaging (MRI) offer the possibility of improved clinical decision making and diagnosis. However, the neonatal brain presents specific challenges to automated segmentation algorithms. We developed a new method for automatic labeling of neonatal brain MR images. The method uses a new sparsity-based atlas selection strategy that requires a very limited number of atlases 'uniformly' distributed in the low-dimensional data space, combined with a machine learning based label fusion technique. The performance of the method for brain labeling from data of 66 newborns is evaluated and compared with results obtained using majority vote. The proposed method provides accurate brain labeling results with a mean Dice coefficient of 91%. As the proposed method can learn from partially labeled datasets, it can be used to segment large-scale datasets efficiently.