Automatic Pathological Lung Segmentation Based On Dictionary Learning And Sparse Composition In Low-dose CT Image

Segmentation of lungs with severs pathology is a nontrivial problem in clinical application.Due to complex structures,pathological changes,individual differences and low image quality,accurate lung segmentation in clinical 3D CT images is still a challenging task.To overcome these problems,a novel dictionary-based approach is introduced to automatically segment pathological lungs in 3D low-dose CT images.Sparse shape composition is integrated with eigenvector space shape prior model,called eigenspace sparse shape composition(ESSC),to reduce local shape reconstruction error caused by weak and misleading appearance prior information.To initialize the shape model,vertex localization algorithm based on threshold segmentation image can obtain initial shape quickly.The initial shape is closer to the target segmentation boundary,which facilitates the subsequent algorithm to fine-tune the landmarks on this basis.Subsequently,a landmark recognition method based on discriminative appearance dictionary(DAD)is introduced to handle lesions and local details.Furthermore,a new vertex search strategy based on gradient vector flow(GVF)filed is also proposed to drive shape deformation to target boundary.The proposed algorithm is tested on 78 3D low-dose CT images with lung tumors.We compare our method with normal-based,shape model,traditional related segmentation algorithm and deep learning algorithm.Quantitative and qualitative experimental results demonstrate that the proposed algorithm has obvious advantages in segmentation of pathological lung.