Computer aided detection of minuscule malignant nodules from ct images of the lungs

Abstract

Lung cancer is one of the foremost causes of cancer death in the world. The newlineexpeditious recognition of lung cancer is a tough problem due to the structure of newlinecancer cell, where the majority of the cells are co-occurrence with each other. It is newlinecomplicated to evaluate cancer at its early stage. In the past few years, numerous newlineComputer-aided systems have been intended to identify lung cancer at its early stage. newlineIf lung cancer is effectively rooted out and forecasted in its early stages it will lessen newlinemany treatment options as well as condense the risk of insidious surgery and enhance newlinesurvival rate. As a result, lung cancer detection and prediction systems will provide newlinepromising result for recognition and forecast of lung cancer which would be easy to newlineuse, cost-effective and time saving. This is mostly accomplished on Computer newlineTomography (CT) scan images because of better clarity, low noise, and distortion. newlineThe proposed system comprises of five steps namely image acquisition, newlinepreprocessing, segmentation, feature extraction and classification. Initially CT images newlineare acquired from Lung image database consortium (LIDC). The acquired image are newlinethen passed on to the preprocessing stage where the CT mages are enhanced with the newlinehelp of median filter. In the next stage, segmentation is carried out using modified newlineOTSU segmentation method, from which the GLCM, Statistical, texture and higher newlineorder features are extracted and finally classification is carried out using Support newlineVector Machine (SVM), K Nearest Neighbor (KNN) and Linear Discriminant newlineAnalysis classifiers. newlineThe proposed research aims to develop a CAD system which will reduce the newlinetime required to detect cancer and non-cancer image this will help the radiologists to newlinepredict the malignant nodule and benign nodule by decreasing the number of false newlinepositive rates. The accuracy, sensitivity and specificity of the developed system are newline99.6% , 98% and 97.2% respectively newline