Photo Credit: Streetoncamara

Convolutional neural network (CNN)-based COPD diagnosis and staging using single-phase computed tomography (CT) was comparable with inspiratory-expiratory CT with inclusion of clinical data, according to a study published in Radiology: Cardiothoracic Imaging. The retrospective study used images and measurements acquired between November 2007 and April 2011 for 8,893 COPDGene participants. The researchers observed moderate to good agreement for CNN-predicted and reference standard spirometry measurements (ICC, 0.66-0.79), which improved with clinical data (ICC, 0.70-0.85), apart from FEV1/forced vital capacity in the inspiratory-phase CNN model with clinical data and FEV1 in the expiratory-phase CNN model with clinical data. Single-phase CNN accuracies for Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage, within-one GOLD stage, and diagnosis (GOLD 0 vs 1-4) varied from 59.8% to 84.1%, and agreement was moderate to good (ICC, 0.68-0.70). Using inspiratory and expiratory images, CNN models’ accuracies varied from 60.0% to 86.3%, with moderate to good agreement (ICC, 0.72). For both the single-phase CNNs and inspiratory-expiratory CNNs, the inclusion of clinical data improved agreement (ICCs, 0.72 and 0.77-0.78, respectively) and accuracy (62.5% to 85.8% and 67.6% to 88.0%, respectively).