High performance Automated lung segmentation and tumor detection in Pet ct images

Abstract

Cancer has a greater impact on human lives over all parts of the world. It is expected newlineto have 22 and 1.7 million new cancer cases worldwide and in India respectively newlineby 2030. Lung cancer is the second most common cancer in both genders. Every newlineyear around 1.8 million people are diagnosed with lung cancer. The overall 5-year newlinesurvival rate of lung cancer is very deprived. However, diagnosing lung cancer at newlinean early stage helps to increase the survival rate by 73%. Currently several imaging newlinemodalities are available for medical diagnostics. Among various imaging newlinemodalities, Positron Emission Tomography combined with Computed newlineTomography (PET/CT) is predominantly used in cancer detection, accurate newlinestaging, treatment planning and monitoring. PET/CT provides better imaging data newlinewith the use of anatomic information by Computed Tomography (CT) scan and newlinefunctional information by Positron Emission Tomography (PET) scan. newlineComputer Aided Diagnosis (CAD) systems are developed to assist doctors in newlineanalyzing the medical images captured by arious imaging modalities in detecting the lung cancer. This study aims at eveloping an intelligent system for automatic lung cancer detection from PET/CT images using texture and fractal descriptors.Adaptive pre-processing methods have been carried out to improve the quality of newlinethe image which will have greater impact on the efficiency of tumor detection. newlineStatistical features based on texture and fractal have been extracted and analyzed newlinefor selection of salient image features, which facilitate accurate lung cancer newlinedetection. Dimensionality reduction has been performed based on the correlation newlineco-efficient, which helps to attain less time complexity and computational newlinecomplexity. newlineClassical Machine Learning (ML) techniques have been investigated in lung cancer newlineclassification. Performance of the developed intelligent system with various newlineclassifiers has been analyzed and observed that Support Vector Machine (SVM) newlinewith Radial Basis Function (RBF) kernel with width, and#963; = 1 outperformed other newlineclassifiers