Abstract
Lung nodules are incipient manifestations of lung cancer, and lung nodules detection and benign/malignant classification based on CT images are of great clinical significance. Aiming at overcoming the difficulty of benign/malignant classification of lung nodules in clinical practice, we put forward a benign or malignant classification algorithm on the strength of multi-level feature fusion. Our method can not only fuse the context features at different levels, but also performs benign or malignant classification at different levels, and the final classification result is obtained by voting. Experiments on LUNA16, a public dataset of chest CT images, show that our method is very effective.