Segmentation and Classification of Lung Nodules using Split Bregman and SVM Classifier

Abstract

Lung nodules are a commonly occurring problem in the society. This problem is more prevalent in populations that expose themselves to risk factors such as smoking, pollution, etc. The current techniques used for segmentation of lung nodules from CT images have the following drawbacks: Most segmentation algorithms use Local Fitting Models that need re-initialization for the sign distance function. The inhomogeneity in the CT images make the algorithms to settle at local minima and lead to wrong segmentation. Re-initialization increases the time required and makes the algorithm slow. Region-based active contour models are powerful and flexible methods which can able to segment real and synthetic images. In the proposed method Global Convex Segmentation (GCS) and Split Bregman technique is incorporated into a region based active contour model such as Chan-Vese (CV) with Region-Scalable Fitting (RSF) scheme to segment the Lung nodules region. Local Binary Pattern descriptor (LBP) is used to extract the tumor features. The extracted features are used to classify the nodules as tumor or non-tumor with the help of Support Vector Machine (SVM). The classification accuracy obtained is enhanced compared to other existing methods. Experimental results are demonstrated by using Lung CT images.