Automated Lung Field Segmentation in CT Scan Images

Abstract

Lung cancer is the leading cause of death and most commonly diagnosed worldwide newlineboth in men and women. The presence of the pulmonary nodule is the potential newlinesign of lung cancer very few stages are diagnosed at the primary stage. If these nodules newlineare detected at the localized stage then the chances of survival can be increased. newlinePresently the computed tomography (CT) is used for diagnosis and treatment of lung newlinecancer. However, this CT scanner produces a lot of data hence interpretation and newlinemanual segmentation is challenging and time-consuming. It puts the burden on the newlineradiologist and increases their workload, also the chances of missing some pathological newlinedetails such as abnormalities are more with a visual inspection. The aim of this newlineresearch denotes to contribute towards the development of computer-aided diagnosis newlineand detection of lung cancer, in order to assist radiologists for the efficient diagnosis newlineand detection of lung cancer. In particular, we concentrate on the development of newlinethe algorithmic component of an automatic lung segmentation, juxtapleural nodule newlineinclusion, lung lobe quantification, and pulmonary nodule segmentation. The developed newlinetechnique can be categorized in: lung segmentation, inclusion of juxtapleural newlinenodule and pulmonary nodule segmentation along with and lung lobe quantification. newlineIn this thesis firstly we proposed the novel lung field segmentation method based newlineon unsupervised clustering and non-negative matrix factorization. The proposed considered newlinethe spatial interaction of the neighboring voxel obtain from the novel image newlinefeature and these image features are a model by CNMF based algorithm. The proposed newlinemethod was evaluated on the publicly available LOLA11 dataset and achieves newlinethe average DSC of 0.973, the sensitivity of 0.965 and specificity of 0.941. However newlinethe performance of this method affected by the noise along the lung boundary. newlineTherefore we have proposed the NMF and fuzzy clustering-based method along with newlinethe border smoothing approach. This method gives the average DSC of 0.981