Deep learning-derived quantitative interstitial abnormalities in early rheumatoid arthritis and healthy controls: A multicenter, prospective cross-sectional study
Authors
Affiliations (1)
Affiliations (1)
- Mass General Brigham / Brigham and Women's Hospital and Harvard Medical School
Abstract
ObjectiveQuantitative computed tomography (QCT) can automatically quantify parenchymal abnormalities on chest CT imaging using deep learning. We leveraged QCT to detect pulmonary abnormalities in patients with early rheumatoid arthritis (RA) compared to healthy controls. MethodsWe analyzed high-resolution CT chest imaging from participants with early RA in the prospective, multicenter, SAIL-RA study and healthy non-smoking controls from the COPDGene study. A deep learning classifier quantified the percentage of normal lung, interstitial abnormalities, and emphysema for each participant. We compared the percentage of QCT features between early RA participants and healthy comparators and examined associations using multivariable linear regression. ResultsWe analyzed 200 participants with early RA (median RA duration 8.3 months, mean age 55.7 years, 74.5% female) and 104 healthy controls (mean age 62.0 years, 68.3% female). The median percentage of interstitial abnormalities on QCT was 3.7% (IQR 2.1, 6.1%) for early RA and 1.6% (IQR 0.8, 2.4%) for healthy controls (p<0.0001). Early RA was associated with 9.3% less normal lung on QCT than healthy controls, adjusted for age and sex (p<0.0001). Among RA participants, QCT interstitial abnormalities were associated with older age (multivariable {beta}=0.1 per year, 95%CI 0.07-0.2, p<0.0001) and higher DAS28-ESR (multivariable {beta}=0.6 per unit, 95%CI 0.01-1.3, p=0.046). ConclusionParticipants with early RA had less normal lung and more interstitial abnormalities on a deep learning-derived QCT measure than healthy controls. These results suggest that loss of normal lung is already present in early RA and emphasizes the urgent need for strategies to preserve lung health in RA.