Automated Segmentation of Lung Lobes and Recognition of Pulmonary Nodules from Multi-slice Chest X-ray CT Images

摘要

The aim of our CAD (computer-aided diagnostic) system is to discover indistinct nodules in Chest X-ray CT. We firstly get the accurate segmentation of lung lobes from multi-slice images based on an improved rolling ball algorithm using morphology. Secondly a multiple-template matching technique was used to search for the location of nodules in the lung areas, where image fusion of true nodules were used as templates. We finally focus on the elimination of blood vessels by using the 3D reconstruction of detected nodules position. Experiments show that 78 detection rate with approximately 85 false positives elimination rate.