Novel Classification Technique for Diagnosis of Lung Cancer on Medical Images

Abstract

Among cancer patients the most general cause for death in the world is lung cancer. Survivalfrom lung cancer is specifically identified with its development at its discovery time. Earlydiagnosis of lung malignancy is the most encouraging approach to reduce the risk for newlinesurvival. Staging of cancer and its investigation is the major predictor of survival, whichdetermines the treatment. Various imaging modalities are available for detecting the lungcancer. Among which ,Computed Tomography (CT) is the most commonly used imaging technique.Many Computer Aided Diagnosis (CAD) systems were developed to detect lung cancerin its early stage using Computer Tomography (CT) images. The CAD systems mainlyconcentrate on identifying and detecting the lung nodules. Staging the lung cancer and its detectionneed to be focused mainly for better treatment. The major drawbacksof existing CAD systems are the inaccuracy in segmenting the nodule and staging the lungcancer. As a solution to this, this research concentrates on developing a new CAD methodsfor segmenting the nodules and detecting the various stages of the lung cancer, thereby aiding the radiologists for analysing the disease. newline newlineTo facilitate the segmentation accuracy of nodules in the CT images, a hybrid method isdeveloped in this research. The accuracy of segmentation is improved by combining random walk and watershed algorithms. This hybrid approach gives healthier results principally fornodules with irregular boundaries. Using 23 images the proposed approach is validated by comparing random walk, with modified weights, and hybrid approaches. The maximum dicesimilarity coefficient values obtained are 0.92 for Random walk, 0.94 for random walk withmodified weight and 0.97 for hybrid method. newline newlineThe most initial step in detecting and classifying lung cancer is, feature analysis ofCT images. Among different types of image features, statistical measures of Asymmetry,Border irregularity, Colour variegation and Diameter (ABCD) texture has been used for image texture description. U