Automated Segmentation of Cerebral Aneurysms Based on Conditional Random Field and Gentle Adaboost

摘要

Quantified geometric characteristics of cerebral aneurysms such as volume, height, maximum diameter, surface area and aspect ratio are useful for predicting the rupture risk. Moreover, a newly developed fluid structure interaction system requires healthy models generated from the aneurysms to calculate anisotropic material directions for more accurate wall stress estimation. Thus the isolation of aneurysms is a critical step which currently depends primarily on manual segmentation. We propose an automated solution to this problem based on conditional random field and gentle adaboost. The proposed method was validated with eight datasets and four-fold cross-validation, an accuracy of 89.63%±3.09% is obtained.