Output details

Improving intersubject image registration using tissue-class information benefits robustness and accuracy of multi-atlas based anatomical segmentation

Automated anatomical labelling of human brain images by fusion of labels propagated from atlases is a powerful method pioneered by our group. It can fail if the atlases are too different from the target brain, such as around the ventricles in Alzheimer’s cases. This paper added local tissue class information to resolve the problem, achieving substantial performance gains. Two variants of this method were ranked 3 &5 out of 25 in a recent MICCAI brain segmentation grand challenge. The first author was Hajnal’s PhD student and last authorship acknowledges the originator of the brain atlases used to demonstrate the technique.