Research On Automatic Detection Methods Of GGO In 3D Lung CT Images

Lung cancer is the most common disease and leading cause of cancer death in the world.In a lung cancer screening program,Computer-Aided Detection(CAD)systems can help doctors diagnose the disease,effectively improving the diagnostic accuracy and reducing the workload of doctors.In this paper,we focus on the study of the automatic detection of ground glass opacity(GGO)in three-dimensional(3D)lung CT images.The main contributions of our work are summarized as follows:(1)In this paper,we propose a GGO candidate extraction method which is based on supervoxels and multilevel thresholding.Firstly,we segment the lung parenchyma based on traditional thresholding method and morphological methods.Secondly,the voxels which are adjacent in 3D discrete space and sharing similar grayscale are clustered into supervoxels.Thirdly,Hessian matrix is used to emphasize focal GGO candidates.Lastly,for not missing diffuse GGO nodules,an adaptive multilevel thresholding method and morphological methods are applied to extract final GGO candidates.(2)We firstly propose a GGO classification method based on multi-grained cascade forest(gc Forest)in GGO false positive removal stage.After the GGO candidates are extracted,the CT images st ill contain many non-GGO structures,so the false positives need to be further removed.Firstly,the GGO candidates are processed into structured eigenvectors by Multi-Grained Scanning.Then the structured eigenvectors are used for training the Cascade Forest Structure model to classify GGO nodules.This method greatly reduces the number of GGO false positive,with a high accuracy.Finally,the detected GGO nodules are showed with 3D reconstruction and the suspicious lesion areas are marked on the original images.To evaluate the performance of the proposed approach for 3D GGO automatic detection,we conduct experiments on the 3D LISS CT image sets,which contain 19 scans and 166 GGO lesions.The instances of the image sets were collected from the Cancer Institute and Hospital at Chinese Academy of Medical Sciences.The experimental results show that our proposed GGO candidate extraction method is effective,with a sensitivity of 100% and 26.3 of false positives per scan.The sensitivity of gc Forest-based false positive removal experiments is 95.78% with the specificity of 74.61% and the average false positive per scan is 7.3.The repeated experiments and cross-validation prove that our proposed methods significantly improve the accuracy of GGO detection,which are significant theoretically and practically for the development of GGO CAD systems.