Accurate Lung Nodules Segmentation with Detailed Representation Transfer and Soft Mask Supervision

Wang, Changwei; Xu, Rongtao; Xu, Shibiao; Meng, Weiliang; Xiao, Jun; Peng, Qimin; Zhang, Xiaopeng

Accurate lung nodules segmentation from Computed Tomography (CT) images is crucial to the analysis and diagnosis of lung diseases such as COVID-19 and lung cancer. However, due to the smallness and variety of lung nodules and the lack of high-quality labeling, accurate lung nodule segmentation is still a challenging problem. To address these issues, we propose a complete paradigm for accurate lung nodules segmentation. First, we introduce a new segmentation mask named Soft Mask which has richer and more accurate edge details description and better visualization. Correspondingly, we develop a universal semi-automatic Soft Mask annotation pipeline to deal with different datasets. Second, a novel Network with Detailed representation transfer and Soft Mask supervision (DSNet) is proposed to process the input low-resolution images of lung nodules into high-quality segmentation results. In our DSNet, we design a novel Selective Detailed Representation Fusion Module to reconstruct the detailed representation to alleviate the small size of lung nodules images. In addition, the adversarial training framework with Soft Mask is proposed to further improve the accuracy of segmentation. Extensive experiments validate that our DSNet outperforms the state-of-the-art methods for accurate lung nodules segmentation. And our method also demonstrates competitive results in other accurate medical segmentation tasks. Besides, we provide a new challenging lung nodules segmentation dataset for further studies.