Segmentation Of Pulmonary Nodules Based On Chest X-ray CT Images

The mortality of lung cancer is higher than that of other kinds of cancers in the world, the main reason of which is that symptoms of lung cancer are too inconspicuous to diagnose easily. One of the most important methods for reducing the mortality of lung cancer is "to discover, diagnose, and treat it at early stage". The pulmonary nodule is a main representation of lung cancer on chest CT image slices, however, other tissues with similar gray scale to pulmonary nodules, such as blood vessels and bronchi, often overlap with the nodules, so even experienced radiologists might make mistakes in diagnosis of lung cancer. In recent years, the Computer-Aided Diagnosis (CAD) system has been developed and used clinically in discriminative diagnosis of lung cancer at very early stages, however, many technical problems haven't been solved yet.The work presented in this thesis is supported by the National Natural Science Foundation of China under grant number 60671050 (the Research of key algorithm in computer-aided diagnosis of lung cancer). We propose a novel and effective approach for segmentation of pulmonary nodules, which works in four main steps:pretreatment of CT image, segmentation of pulmonary parenchyma, localization of pulmonary nodules and segmentation of pulmonary nodules. Firstly, a CT image is transformed into a BMP-format gray scale image, and is de-noised then. Secondly, pulmonary parenchyma is segmented based on the dynamic threshold method and the Rolling-ball algorithm. Thirdly, location of pulmonary nodules is ascertained based on the improved self-adaptive multi-template matching method. Lastly, for a small size nodule, it is segmented based on an improved dual fast marching method. For a big size nodule, transition region, which is defined as the ambiguous region between nodule and background, is ascertained depending on statistic features of wavelet coefficients, and then precise boundary of nodule is segmented based on the dual fast marching method. The validity of the proposed approach is demonstrated in abundant experiments based on real chest CT images, and the experimental results confirm high accuracy of the approach proposed in this thesis.