Automatic Detection Of Lung Nodules Based On CT Images

Malignant tumor is a very serious health problem for human beings and lung canceris the most harmful one at present. Lung nodule detection in early stage is very importantfor lung cancer treatment. And X-ray computed tomography (CT) is the most sensitivemodality for lung nodule detection in clinical. However, with the continuous improvementof CT equipment, it produces more and more images. Large amount of CT images notonly burden the image interpretation physician, but also increase the rate of overlook andmisdiagnosis. This dissertation focuses on computer-aided detection of lung nodules inchest CT images which is the most important part of lung CAD system.Segmentation of lungs from CT images is the precondition of lung nodules detection.At first, thorax region is segmented from the binarized CT images through border tracingalgorithm and ?ood fill algorithm. Then initial lung border is found in the binarized thoraxregion. In order to ensure the integrity of lung region, a new smoothing algorithm whichis based on the curve expression that take arc-length as a parameter is presented and it isused to smooth the initial lung border. Another lung border correction algorithm which isused to correct the smoothed lung border has also been proposed. At the end, lung regionis segmented completely.In this dissertation quoit filter which is based on Mathematical Morphology dilationis used to detect lung nodules. At the beginning, the principle of two-dimensional quoitfilter is introduced and then quoit filter is extended to three-dimensional in order to getrid of those false positive results caused by the vessels. For the inadequacies of quoitfilter, three improvements are proposed and ultimately obtained more adaptive variablequoit filter. The problem of distance transform of an image which is encountered in theimprovement process has also been studied in this paper. An e?cient Euclidean distancetransform algorithm has been given and a new path length is defined to achieve the gray-level weighted distance transform. Results of the experiment showed that this detectionsystem can drastically reduce the burden of the doctor.