Early Lung Cancer Detection A New Automated Approach with Improved Diagnostic Performance

Abstract

Lung cancer is the second most common cancer in the World. Death rates due to lung newlinecancer is high because of tobacco smoking and unhealthy lifestyle. There is a need to develop newlineCADx for lung nodule detection as an aid to radiologist in delineating the nodules. This research newlinework aims at developing an automated CADx for demarcating the nodules and classifying them as newlinebenign and malignant. The methodology involves the preprocessing of the LDCT images, newlinefollowed by delineation of the lung nodules, extracting features from the nodules, and finally newlineclassifying as benign and malignant. Images are acquired from LIDC-IDRI public database and newlinepreprocessed by Discrete wavelet transform. The performance evaluation of the existing filters newlinewith the proposed method is carried out using the metrics of mean square error and peak to signal newlinenoise ratio. The contrast of the LDCT images is improved by the Adaptive histogram equalization newlinemethod. The threshold limit is computed by genetic algorithm. The LDCT images undergo lung newlinesegmentation by fuzzy c-means clustering enhanced by genetic algorithm. The lung boundaries are newlinereconstructed by morphological operations and the nodules are delineated by area filtering newlineoperator. The radiomics feature are extracted from the nodules for feature reduction and newlineclassification. The radiomic features of Histogram features, Gray level co-occurrence matrix newline(GLCM), Gray level run length matrix (GLRLM), neighborhood gray-tone difference matrix newline(NGTDM), and Region of interest features are extracted for the LDCT images. The feature newlinereduction is performed by Least absolute selection and shrinkage operator. LASSO shrinks the newlinecoefficients of non-deterministic feature variables to zero. The variables which highly influence newlinethe nodule classification is selected for the classification phase. The classification of the nodules newlineas benign and malignant was carried out by ensemble method of bagging as well as AdaBoost. newlineThis method was compared with the SVM and showed improved performance with accuracy of newline96%, specifi