SVM based lung cancer classification using texture and fractal features from PET/CT images

Abstract

Early lung cancer detection is extremely challenging as symptoms are not exposed till advanced stage. This study is aimed at developing a computer aided diagnosis (CAD) system with image processing techniques and support vector machine (SVM) in lung cancer classification from positron emission tomography/computed tomography (PET/CT) images. The developed CAD system utilized fuzzy enhancement for contrast improvement. Texture and fractal features were used for training the SVM. This study utilized 82 PET/CT images and 10-fold cross validation to analyze the performance of the classifiers. Experimental study showed that SVM classifier with radial basis function (RBF) kernel of width, σ = 1 outperformed the other SVM models. It produced maximum accuracy of 98.13% using texture and fractal features from PET/CT images. The RBF kernel is effective in handling sparse, non-linear, multi-dimensional data to transform it into linearly separable.