A 3D Fusion U-Net with Dual CNN and Transformer Encoders for Lung Airway Segmentation

The health of the lungs is significant importance to human well-being. As a vital respiratory organ, lung diseases have a high incidence and mortality rate worldwide. Accurate segmentation of the lung airways, including the trachea and bronchi, is crucial for the diagnosis of lung diseases. The recent coronavirus has further shown the importance of accurately distinguishing the lung airways from the lung parenchyma to assist in diagnosis. Therefore, developing an automated segmentation method that can accurately segment the lung airways from lung CT images is essential. Despite extensive research in this field, challenges remain. The unique tree-like structure of the lung airways and the fine-grained details at the terminal bronchioles pose difficulties for achieving precise segmentation using existing methods. In this study, we propose a novel dual-encoder network for accurate lung airways segmentation. Our method can efficient capture the local and global features using the combination of Convolutional Neural Networks (CNNs) and Transformer network. We evaluate our method on a lung CT dataset provided by Shandong University and demonstrate its superiority over existing methods.