Optimum Template Selection for Atlas-Based Segmentation
Wu M, Aizenstein H, LoSurdo J, Carter C, Lopez-Garcia P
University of Pittsburgh
Atlas-based segmentation typically uses a single atlas (e.g., MNI Colin27) for region identification. However, a single template is limited for structures, such as the anterior cingulate cortex (ACC), which have significant individual structural variations. An automated template selection method (with normalized mutual information metric) was developed to select the template (from multiple templates) optimized for ACC segmentation for each subject. The ACC segmentation overlap-ratio (9 subjects) using the optimum template selection approach outperformed the segmentation from individual templates (p=0.0024). These results highlight the importance of template selection for atlas-based segmentation and demonstrate an automated approach using a family of templates.