A deep convolutional neural network for brain tissue segmentation in Neonatal MRI
Murphy, Keelin, Boylan, Geraldine B., Smeaton, Alan F.ORCID: 0000-0003-1028-8389 and McGuinness, KevinORCID: 0000-0003-1336-6477
(2017)
A deep convolutional neural network for brain tissue segmentation in Neonatal MRI.
In: The 10th International Conference on Brain Monitoring and Neuroprotection in the Newborn, 5-7 Oct 2017, Killarney, Ireland.
Brain tissue segmentation is a prerequisite for many subsequent automatic quantitative analysis techniques. As with many medical imaging tasks, a shortage of manually annotated training data is a limiting factor which is not easily overcome, particularly using recent deep-learning technology. We present a deep convolutional neural network (CNN) trained on just 2 publicly available manually annotated volumes, trained to annotate 8 tissue types in neonatal T2 MRI. The network makes use of several recent deep-learning techniques as well as artificial augmentation of the training data, to achieve state-of-the- art results on public challenge data.