Classification of malignant-benign pulmonary nodules in lung CT images using an improved random forest (Use style: Paper title)

To help better radiologists with the early diagnosis of lung cancer on CT scans, we propose a novel classification scheme for malignant-benign classification of pulmonary nodules. We first segment pulmonary nodules by consulting the segmented results drawn by four radiologists. Then, we compute image features (gray feature, shape feature and texture feature) to represent the segmented nodules. Finally, malignant-benign classification of pulmonary nodules is performed using an improved-RF. Experiments were performed on a database of 603 pulmonary nodules, including 288 benign and 315 malignant nodules. And the provided method is compared with some classic classifiers and existing classification schemes of malignant-benign pulmonary nodules, respectively. Experimental results demonstrate that the proposed scheme outperforms the competing techniques.