Airway 3D reconstruction with respiratory motion correction using endoscopic OCT images

Endoscopic optical coherence tomography (OCT) can provide high-resolution, real-time images of luminal tissues. Three-dimensional (3D) airway reconstruction from endoscopic OCT images can visualize the anatomical structure of the airway and assist clinicians in diagnosing and analyzing airway-related diseases. Previous airway reconstruction methods did not differentiate respiratory states during the reconstruction process. By pulling back the endoscopic probe, B-scans at different cross-sections were captured in various respiratory states such as expiration or inspiration. Changes in luminal structure in the images could be attributed to respiratory motion. In this study, an automatic airway 3D reconstruction framework with respiratory motion correction is proposed for endoscopic OCT images. The proposed framework consists of three main steps: automatic airway lumen segmentation by leveraging a convolutional neural network, respiratory state-consistent data extraction according to the contour variation during the respiratory motion, and rotational distortion correction based on the segmented lumen contours. The proposed reconstruction framework was validated on experimental datasets acquired from the rabbit. After acquiring airway B-scans covering the entire respiratory cycle at each spatial position, the airway 3D structure can be reconstructed. The results show that the proposed framework can automatically reconstruct the 3D airway from endoscopic OCT images and keep the respiratory state consistent with accurate performance.