Abstract
We previously described a semi-automated airway volume measurement in IPF (1). We now show a novel fully automated machine learning approach. Five expert radiologists manually delineated airways in CTs from 101 IPF patients. We then created a multistage convolutional neural network to perform end-to-end segmentation from full CT volume. We applied this to 31 new IPF patients, each with 2 CT scans (1a). Patients were divided into two groups: stable (<10% FVC change between scans and specialist MDT review confirming CT stability, n=15) and progressive (>10% FVC change, MDT review progressive, n=16). Our model showed difference in median airway size change between groups, with larger increase in the progressive group (32.4% vs. 8.6% p =0.001, 6 ml/L vs 0.8ml/L corrected lung volume p = 0.0003, 1b). Lung volume declined in both groups, larger in progressive group (-15.4% vs -6.7%). Airway change correlated with FVC change (1c). Application of unsupervised (K-Means) machine learning classified patients into two distinct clusters when adding airway volume to FVC change (1d). Traction bronchiectesis correlates with IPF progression but is time consuming and requires expert radiologists to carefully measure each airway (2). Larger longitudinal studies with our model are needed to see if it can be used as a surrogate marker for clinical progression.
1. McLellen et al ERJO 2020
2. Jacob et al Thorax 2020
Footnotes
Cite this article as: European Respiratory Journal 2021; 58: Suppl. 65, OA3951.
This abstract was presented at the 2021 ERS International Congress, in session “Prediction of exacerbations in patients with COPD”.
This is an ERS International Congress abstract. No full-text version is available. Further material to accompany this abstract may be available at www.ers-education.org (ERS member access only).
- Copyright ©the authors 2021