Meeting Banner
Abstract #2890

Improved Skull Stripping Using Graph Cuts

Suresh Anand Sadananthan1, Weili Zheng1, Michael WL Chee2, Vitali Zagorodnov1

1Computer Engineering, Nanyang Technological University, Singapore, Singapore; 2Cognitive Neuroscience Laboratory, Duke-NUS Graduate Medical School, Singapore, Singapore

Many recent skull stripping approaches rely on iterative surface deformation to fit the brain boundary and tend to leave residual dura, which may cause cortical thickness overestimation. The approach proposed here is based on intensity thresholding followed by removal of narrow connections using graph theoretic image segmentation to position cuts that serve to isolate and remove dura. Relative to the Hybrid Watershed Algorithm (HWA), the novel approach achieved at least 10-30% reduction in dura mater without having to trade off for increased brain tissue erosion. However, the main value of the new approach is significant reduction in cortical thickness overestimation when combined with HWA.