A novel quantitative method for CT assessment of bronchiectasis using AI-based airway segmentation and upper limits of normal airway caliber

Bronchiectasis represents a growing global health burden. While computed tomography (CT) is essential for diagnosis, current criteria rely on broncho-arterial ratios and visual assessments, which are subject to inter-observer variability and can be confounded by pathological changes in arterial caliber. This study aimed to develop a novel quantitative method for assessing bronchiectasis by incorporating artificial intelligence-based airway segmentation and establishing location-specific upper limits of normal (ULN) for airway dimensions derived from healthy controls. We analyzed chest CT scans from 459 healthy non-smokers who participated in a lung cancer screening program. Of these, 445 were used to calculate sex- and height-specific ULN values, while 14 served as independent controls for comparison with 14 patients clinically diagnosed with bronchiectasis. Following automatic segmentation of airway trees using previously-established artificial intelligence-based quantitative CT image analysis software, we extracted centerlines. At each point along the airway centerline, we measured the cross-sectional radius and the distance from the lobar bronchus origin. The ULN for airway radius was defined as the 95th percentile of measurements from healthy controls, stratified by sex and height. In patients with bronchiectasis, airways exceeding the ULN were classified as bronchiectatic, and bronchiectatic airway volume percentage (BEV%) was calculated. Male sex and height, but not age, were significantly associated with lobar bronchus radius. Location-specific ULN values were established for all five lobes as a function of distance from lobar origins, stratified by sex and height. BEV% was significantly higher in patients with bronchiectasis compared to matched controls (p < 0.001) and strongly correlated with the Reiff score (rho = 0.722, p = 0.004). This study established a novel reproducible approach for quantifying bronchiectasis using AI-based airway segmentation and location-specific normative airway dimensions without relying on broncho-arterial ratios. This method potentially enhances diagnostic accuracy and disease monitoring in the management of bronchiectasis.