Oropharyngeal enlargement in obstructive lung disease: quantification and machine learning.

Background: While lower airway remodelling of obstructive lung diseases (OLDs), such as asthma and COPD, is comprehensively studied, the understanding of upper airway remodelling in OLD remains limited. This study aimed to investigate upper airway dimensions in patients with OLD using quantitative computed tomography (QCT) imaging and to identify relevant parameters for predicting OLD using machine learning techniques. Methods: A prospective cohort of 26 healthy controls, 73 COPD patients and 86 asthma patients underwent upper airway computed tomography (CT) scans from the oral cavity to the subglottal region. Multiscale lung structure and function were assessed using ITK-SNAP and in-house QCT software. Feature-importance estimation methods from STREAMLINE were utilised to select potentially relevant upper airway metrics. The Wilcoxon rank-sum test and Pearson's correlation were employed for pairwise comparisons and correlation analysis, respectively. The Youden index was used to determine optimal cut-off values of relevant upper airway features. Results: After standardising QCT results, patients with OLD exhibited greater mouth-to-supraglottal metrics, notably greater oral space air fraction and pharyngeal length. Both metrics showed a negative correlation with forced expiratory volume in 1 s/forced vital capacity (R=-0.24; p=0.001). Feature-importance analysis identified oral space air fraction and normalised pharyngeal length as key features discriminating patients with OLD from healthy controls. An oral space air fraction value of ≥0.8 predicted OLD with approximately 100% sensitivity and 69% specificity. Conclusions: Quantitative upper airway CT measurement combined with machine learning analysis revealed oropharyngeal enlargement in patients with OLD.