A Computer-Aided Diagnosis System for Lung Cancer Detection with Automatic Region Growing, Multistage Feature Selection and Neural Network Classifier

Abstract

An effective automatic region growing was developed in this work for the segmentation of suspected lung nodules from the Computed Tomography (CT) lung images. After the segmentation of the suspected lung nodules the eccentricity and area features were calculated to eliminate line like structures and tiny clusters below 3mm. The centroid analysis, contrast, autocorrelation and homogeneity features were extracted for the suspected lung nodules. The extracted features were trained and tested with Artificial Neural Network (ANN) to remove the blood vessels and calcifications (calcium deposition in the lungs). This work was carried out on 106 patients images retrospectively collected from Bharat Scans, Chennai, which had 56 cancerous nodules and 745 non-cancerous nodules (size greater than 3 mm). The proposed work yielded sensitivity, specificity and accuracy of 100%, 93% and 94%, respectively.