Abstract
BACKGROUND: Dyspnea is one of the most common symptoms in the post-acute phase of COVID-19 pneumonia. Conventional pulmonary function tests and computed tomography (CT) scores often fail to show correlation with symptom severity, highlighting the need for more sensitive imaging biomarkers. Machine-learning–based quantitative CT analysis and parametric response mapping (PRM) can capture subtle structural and functional abnormalities that may be associated with persistent dyspnea. METHODS: We analyzed inspiratory and paired inspiratory–expiratory CT scans of early (3–6 months) post-COVID-19 pneumonia patients. Inspiratory CT images were segmented using a random forest algorithm to quantify lung parenchymal patterns. Paired inspiratory/expiratory scans were co-registered to derive ventilation metrics and PRM-defined functional small airway disease (fSAD), emphysema, emptying emphysema, and normal lung. Associations between imaging metrics and patient-reported dyspnea assessed by a visual analogue scale (VAS) were evaluated using univariable and multivariable linear regression, with adjustment for age, sex, BMI, and smoking history. RESULTS: One hundred twenty-three patients had usable inspiratory CT scans, and 116 patients had paired inspiratory/expiratory scans of sufficient quality for analysis. In the adjusted multivariable models, greater PRM-defined functional small airway disease (fSAD) was positively associated with dyspnea (standardized β = 1.21, p = 0.002). Moreover, a lower standard deviation of dense ground-glass attenuation in the left lung (standardized β = −0.82, p = 0.033) and greater total volume of dense ground-glass opacities (standardized β = 0.71, p = 0.033) were independently associated with dyspnea. CONCLUSIONS: In early post-COVID-19 pneumonia, machine-learning–based CT pattern recognition and PRM revealed that functional small airway disease, and the total volume and heterogeneity of lung dense ground-glass opacities are significantly associated with persistent dyspnea. These findings highlight the potential of quantitative CT to identify pulmonary imaging biomarkers relevant to long COVID symptom burden. TRIAL REGISTRATION: ClinicalTrials.gov (NCT04406324).