Fig. 3

Example cases for nodule segmentation. Four cases from the LIDC dataset and four In-house datasets were randomly selected and displayed here. The LIDC data with radiologist hand-drawn annotations for training and testing and the In-house data were used for inference only. For the LIDC dataset, columns (a), (b), and (c) are the original CT images, ground-truth segmentation provided by radiologists, and the generated masks by segmentation model, respectively. For the In-house dataset, columns (d), (e), and (f) are the original CT images, background removed (nodule only) images via auto-segmentation, and the background only (soft lung tissue) images, respectively