To better understand cerebellum-related diseases and functional mapping of the cerebellum, quantitative mea-surements of cerebellar regions in magnetic resonance (MR) images have been studied in both clinical and neu-rological studies. Such studies have revealed that different spinocerebellar ataxia (SCA) subtypes have differentpatterns of cerebellar atrophy and that atrophy of different cerebellar regions is correlated with specific functionallosses. Previous methods to automatically parcellate the cerebellum|that is, to identify its sub-regions|havebeen largely based on multi-atlas segmentation. Recently, deep convolutional neural network (CNN) algorithmshave been shown to have high speed and accuracy in cerebral sub-cortical structure segmentation from MRimages. In this work, two three-dimensional CNNs were used to parcellate the cerebellum into 28 regions. First,a locating network was used to predict a bounding box around the cerebellum. Second, a parcellating networkwas used to parcellate the cerebellum using the entire region within the bounding box. A leave-one-out crossvalidation of fifteen manually delineated images was performed. Compared with a previously reported state-of-the-art algorithm, the proposed algorithm shows superior Dice coefficients. The proposed algorithm was furtherapplied to three MR images of a healthy subject and subjects with SCA6 and SCA8, respectively. A Singularitycontainer of this algorithm is publicly available.
展开▼