Advanced AI-based image analysis for quantification of interstitial lung disease in rheumatoid arthritis patients

<bold>Background:</bold> Rheumatoid arthritis (RA) is known to be associated with interstitial lung disease (RA-ILD) leading to a significant increase in morbi-mortality. Tools which could assess lung involvement on chest computed tomography (CT) scans are needed to help ILD quantification and, ultimately, improve patients’ management. <bold>Methods:</bold> An artificial intelligence (AI)-based automated system of ILD quantification (Icolung ®) was applied to a retrospective cohort of 182 patients with RA. Progressive pulmonary fibrosis (PPF) was defined according to ERS recommendations. <bold>Aims and objectives:</bold> To look for correlation between image-based quantification and pulmonary function tests. <bold>Results:</bold> The percentage of interstitial lung abnormalities with regard to the total lung volume (TLV) (in % of TLV) was significantly correlated with forced vital capacity (FVC) and diffusing capacity of the lungs for carbon monoxide (DLCO) values (r=-0.24, p<0.01 and r=-0.27, p<0.01). Out of 49 patients with RA-ILD, 34 had PPF. PPFs, compared with non-progressive RA-ILDs, had a significantly higher percentage of interstitial lung abnormalities out of the total lung volume (2.45 (0.58-5.47) versus 0.34 (0.08-1.21) (p<0.01) and a higher CT-severity score 3 (1-5) versus 1 (0-2) (p=0.01). <bold>Conclusions:</bold> Automatic AI-based quantification of pulmonary interstitial lesions in RA patients correlates with physiological parameters such as DLCO and FVC. This system identified more extensive interstitial lesions in patients with PPF and a higher CT-severity score compared to patients with stable RA-ILDs. Further investigations are needed to validate its usefulness in patients’ follow-up, management and outcome.