Automated cerebellar lobule segmentation with application to cerebellar structural analysis in cerebellar disease

Abstract

The cerebellum plays an important role in both motor control and cognitive function. Cerebellar function is topographically organized and diseases that affect specific parts of the cerebellum are associated with specific patterns of symptoms. Accordingly, delineation and quantification of cerebellar sub-regions from magnetic resonance images are important in the study of cerebellar atrophy and associated functional losses. This paper describes an automated cerebellar lobule segmentation method based on a graph cut segmentation framework. Results from multi-atlas labeling and tissue classification contribute to the region terms in the graph cut energy function and boundary classification contributes to the boundary term in the energy function. A cerebellar parcellation is achieved by minimizing the energy function using the α-expansion technique. The proposed method was evaluated using a leave-one-out cross-validation on 15 subjects including both healthy controls and patients with cerebellar diseases. Based on reported Dice coefficients, the proposed method outperforms two state-of-the-art methods. The proposed method was then applied to 77 subjects to study the region-specific cerebellar structural differences in three spinocerebellar ataxia (SCA) genetic subtypes. Quantitative analysis of the lobule volumes shows distinct patterns of volume changes associated with different SCA subtypes consistent with known patterns of atrophy in these genetic subtypes.

Keywords: Cerebellar lobule segmentation; Cerebellum; Graph cuts; Magnetic resonance imaging; Multi-atlas labeling; Random forest classifier; Spinocerebellar ataxia.

Figure 1

Annotated examples of the cerebellar lobules in three different views. The top and bottom rows are MR images overlaid with expert delineated lobules from a healthy control and an ataxia patient respectively.

Figure 2

Diagram of the proposed algorithm.

Figure 3

Example training and prediction result of the boundary classifier. (a) Preprocessed MR image of a subject used for training. (b) Voxel class overlaid with the image in (a), where yellow indicates boundary voxels, blue indicates lobule voxels and otherwise non-cerebellar voxels. (c) Preprocessed MR image of a test subject. (d) Boundary probability output from the random forest classifier overlaid with the image in (c).

Figure 4

Box plots of Dice similarity coefficient comparing ACCLAIM, NL-STAPLE, and the proposed method.

Figure 5

Example lobule segmentation results. Each column contains the preprocessed MR image (coronal slice), the segmentation results and the expert delineation of a subject. From the top to the bottom row are in turn the preprocessed MR images, the segmentation results obtained by ACCLAIM, NL-STAPLE, the proposed method, and the expert delineations.

Figure 6

Box plots of relative lobe volumes for healthy controls and three SCA subtypes.

Figure 7

Box plots of relative lobule volumes for healthy controls and three SCA subtypes.