Quantification of airway plugging on computed tomography imaging of patients with cystic fibrosis

Category: Respiratory

Purpose: Cystic fibrosis is characterised in the lungs on computed tomography (CT) imaging by abnormal airway dilatation and mucus plugging of airways. Quantifying airway plugging using visual analysis of CT scans has been shown to associate with disease severity. Yet visual analysis of mucus plugging is associated with interobserver variability. To date no computer algorithms have objectively quantified mucus plugging. Our study aimed to develop a deep-learning approach to quantify mucus plugging in cystic fibrosis.

Methods and materials: CT scans of subjects with cystic fibrosis were obtained from Leeds University Hospitals. Approval for this retrospective study of clinically indicated pulmonary function and CT data was obtained from the local research ethics committees and Leeds East Research Ethics Committee: 20/YH/0120. CT scans (n=10) were manually annotated with two different labels, one for patent airways and one for plugged airways. Annotations were ratified by an expert radiologist (JJ), and corrections made to the labels as required. Case selection for the training CT dataset included subjects with a range of airway dilatation and a range of airway plugging.

A no-new-Unet (nn-Unet) deep-learning model was trained with the labelled CT scans to segment plugged and patent airways. The segmentations produced by the nn-Unet model formed the input into an airway analysis software tool called AirQuant, which skeletonised the segmentation and quantified airway metrics such as airway tapering rate, tortuosity and patent and plugged airway volume.

Results: When the nn-Unet model trained on the manual plugged and patent airway segmentations was tested on unseen data, airway plugging was characterised well by the model. The model was able to accurately discriminate plugged airways from vessels in the lung (Figure 1). Volumes of airway plugging were quantifiable on longitudinal imaging in the same patient.

Conclusion: Our study describes the development of what we believe to be the first computer algorithm to quantify airway plugging in cystic fibrosis. Accurate objective quantification of airway plugging could provide much needed estimates of disease severity and disease progression in cystic fibrosis. This could aid clinical management and therapeutic trials of medications in patients with cystic fibrosis.