Swin transformer based benign and malignant pulmonary nodule classification

Lung cancer is the leading cause of cancer-related death worldwide. Early detection and diagnosis of pulmonary nodules are crucial to improve the patients’ relative survival rate. Although existing deep learning-based methods have achieved good results in distinguishing benign and malignant pulmonary nodules, most works mainly focus on designing new models to obtain deeper features, rather than on effective feature representations of the discriminative pulmonary nodules. In the present work, a benign and malignant pulmonary nodule classification method based on Swin Transformer model is developed. The multi-head self-attention hierarchical architecture of Swin Transformer capable of flexible modeling and linear computational complexity, enables the deep models simultaneously extract the local and global features, paying more attention on the key areas with a large amount of information in CT images, while suppressing other irrelevant information. The experimental results on LIDC-IDRI dataset demonstrate that the presented model is an effective and competitive method for the classification task of benign and malignant pulmonary nodules, compared with recent state-of-the-art approaches.