Bronchio: A Zero-shot Learning-based Tool for Pulmonary Airway Analyses

摘要: Rationale Morphological changes to airway lumen are cardinal features of airway disease associated with COPD. Chest computed tomography (CT) image processing has led to numerous airway metrics to quantify lumen diameter and wall thickness. However, these measurements require multiple tools, utilization of several labeled airway trees to train segmentation models, and advanced image post-processing steps. We propose BronchIO, an open-source tool to segment and analyze airways with zeroshot learning using Segment Anything Model (SAM). This tool will help clinicians and researchers compute metrics of airway lumen, walls, and provide 3D visualization of specific branches for subsequent analyses. Methods Data from 12 subjects (2 nonsmokers; 2 subjects from each of GOLD stages 0 through 4) enrolled in the COPDGene study were included. BronchIO segments airways using weights from SAM …