A ranking-based lung nodule image classification method using unlabeled image knowledge

In this paper, we propose a novel semi-supervised classification method for four types of lung nodules, i.e., well-circumscribed, vascularized, juxta-pleural and pleural-tail, in low dose computed tomography (LDCT) scans. The proposed method focuses on classifier design by incorporating the knowledge extracted from both training and testing datasets, and contains two stages: (1) bipartite graph construction, which presents the direct similar relationship between labeled and unlabeled images, (2) ranking score calculation, which computes the possibility of unlabeled images for each of the given four types. Our proposed method is evaluated on a publicly available dataset and clearly demonstrates its promising classification performance.