Two-stage contextual transformer-based convolutional neural network for airway extraction from CT images

Accurate airway segmentation from computed tomography (CT) images is critical for planning navigation bronchoscopy and realizing a quantitative assessment of airway-related chronic obstructive pulmonary disease (COPD). Existing methods face difficulty in segmentation, particularly the small branches airway. These difficulties arise due to constraints limited labeling failure meet clinical use requirements COPD. We propose two-stage framework with novel 3D contextual transformer segmenting overall using CT images. The method consists two training stages sharing same modified U-Net network. block integrated into both encoder decoder path network effectively capture long-range information. In first stage, proposed segments mask. To improve performance result, we generate intrapulmonary branch label, train focus on producing second stage. Extensive experiments were performed in-house multiple public datasets. Quantitative qualitative analyses demonstrate that our extracts significantly more longer lengths tree while accomplishing state-of-the-art performance. code available at https://github.com/zhaozsq/airway_segmentation.