3dcnn for pulmonary nodule segmentation and classification

Abstract

Lung cancer is the leading cause of cancer-related death. Early stage lung cancer detection using computed tomography (CT) could prevent patient death effectively. However, millions of CT scans will have to be analyzed thoroughly worldwide, which represents an enormous burden for radiologists. Therefore, there is significant interest in the development of computer algorithms to optimize this clinical process. In the paper, we developed an algorithm for segmentation and classification of Pulmonary Nodule. Firstly, we established a CT pulmonary nodule annotation database using three major public databases. Secondly, we adopt state-of-the-art algorithms of deep learning method in medical imaging processing. The proposed algorithm shows the superior performance on the LNDb dataset. We got an outstanding accuracy in the segmentation and classification tasks which reached 77.8% (Dice Coefficient), 78.7% respectively.