Study On The Computer-aided Detection Of Lung Nodules And3D Visualization Of Pulmonary CT Images

Lung cancer is considered as one of the most fatal cancers in the world, and studies haveshown that early diagnosis and treatment is important in improving the cure rate andprognosis of lung cancer patients. Computer Tomography (CT) scans are believed to be thebest imaging method for lung cancer detection. Computer-aided detection (CAD) system oflung cancer, having advantages of reducing the radiologists’ workload and the oversight, aswell as improving lung cancer’s diagnostic accuracy, has become a worldwide researchhotspot. However, there exist three types of nodules that present the challenges for theaccurate detection in CAD system, which are juxta-pleural nodules, juxta-vascular nodulesand ground glass opacity（GGO）nodules. Aiming at these three kinds of difficult-to-segmentnodules, a CAD scheme using techniques of templates matching and shape constraint CVmodel is presented in this study to facilitate the automatic detection and three-dimensional(3D) visualization of pulmonary nodules in computed tomography (CT) images.The general task flow of CAD scheme for lung nodules detection proposed in this paperconsists of four major stages:(1) Segmentation of lung parenchyma from original thoracic CTimages with template matching method, The goal is to include the juxta-pleural nodules intosegmented lung parenchyma.(2) Extraction of nodule candidates based on combination ofintensity and shape enhancements to improve the GGO contract and remove the vessels.(3)Reduction of false-positive (FP) nodules based on combination of shape constraint CV modeland criteria for nodules detection, and segment the juxta-vascular nodule exactly.(4)3Dvisualization of nodules using the hybrid rendering methods. Experimental result shows thatour method is able to improve the precision of lung segmentation effectively, and achieved92%sensitivity and5FPs/case for juxta-pleural nodules detection,92%sensitivity and4FPs/case for juxta-vascular nodules detection and87%sensitivity and3FPs/case for GGOdetection.Detection performance shows that our method is feasible and effective for detection ofthree types of difficult-to-segment nodules, namely juxta-pleural nodules, juxta-vascular nodules and GGO nodules with lower intensity in HU as compared with the surroundingstructures in images. The different types of nodules using the different detection methods caneffectively improve the accuracy of CAD system.