Pulmonary nodule classification in thoracic CT images using random forest algorithm

Abstract

In this paper, an automatic classification of thoracic pulmonary nodules with Computed Tomography Image as input is performed. We can crisply classify the nodules into two categories: Benign and Malignant. Benign nodules are the ones which do not cause any harm and even if they do, the impact is negligible. Malignant Nodules are the ones which, if not detected on time can cause severe damage to a person, even resulting in death. Henceforth, detection at early stage of lung cancer is critical. We plan to perform our analysis in 4 steps. Firstly, a noise free CT image is obtained after preprocessing. Then, we apply the improved Random Walker algorithm to perform region-based segmentation, resulting in generation of foreground and background seeds. The next step is to bring out important features of the segments. The features can be intensity, texture and geometry based. Finally we used an improved Random Forest method to generate classification trees, comprising of different class labels. Using RF Algorithm, we predict the accurate class label which corresponds to a particular type of nodule and the stage of cancer that it has developed.