Segmentation of lung CT with pathologies based on adapt active appearance models

An adapt Active Appearance Model (AAM) which can initialize the contour automatically is proposed to segment lung CT images with pathologies. Current segmentation methods have failed because the traditional PCA is not suitable to handle outliers caused by update points of pathology area. A fast robust PCA is cited to design the active AAM which can effectively segment the pathology lungs. For the same 30-cases from Jilin Province Tumor Hospital, two other schemes and algorithm in the paper are compared. The experiment results confirm feasibility of the active AAM which achieved a 93.9% overall sensitivity per section.