Abstract
Lung cancer is the most common form of cancer in the world affecting millions yearly. Early detection and treatment is critical in driving down mortality rates for this disease. A traditional form of early detection involves radiologists manually screening low dose computed tomography (CT) images which can be tedious and time consuming. We propose an automatic system of deep learning methods for the detection, segmentation, and texture characterization of pulmonary nodules in chest CT images to produce patient follow-up recommendations based on the 2017 Fleischner society pulmonary nodule guidelines. The system was developed as part of the LNDb challenge and participated in the main challenge as well as all sub-challenges. It is composed of 3D convolutional neural networks based on VGG and U-net architectures. Each network produces an attribute required to calculate a follow-up recommendation. Recommendations are based on the number of nodules in a CT image, their volume, and their texture. The networks were trained using a dataset of 294 CT images along with radiologist annotations. The proposed method produces promising results with an objective agreement of 0.5092 on the test dataset according to Fleiss-Cohen weighted Cohen’s kappa. While the method struggled with false positives for the detection task and a class imbalance for the texture characterization task, it presents a baseline for further work. Our method was submitted to and accepted for publication at the 17th International Conference on Image Analysis and Recognition (ICIAR 2020).