Late Breaking Abstract - AI-powered quantification of ILD patterns on CT for progression evaluation in SSc-ILD patients

Interstitial lung disease (ILD) is responsible for ~35% of systemic sclerosis (SSc)-related deaths. Accurate evaluation of ILD extent on CT is crucial for identifying progressive SSc-ILD, guiding treatment decisions. However, current visual assessment is prone to inter- and intra-observer variability, limiting its reliability and reproducibility. We aimed to validate a novel, AI-based method for automated and reproducible quantification of ILD pattern volumes (reticulations, ground-glass opacities, honeycombing and consolidation) on CT, and to assess its clinical relevance for the management of SSc-ILD. An nnU-Net-based ILD pattern segmentation model developed with cross-validation on 188 patients (8076 axial CT-slices) was tested on 26 patients (1200 axial CT-slices) and used to segment ILD patterns of 74 SSc-ILD patients. We investigated the association between ILD pattern volumes and clinical diagnosis of progressive SSc-ILD, as well as their correlation with pulmonary function tests routinely used in standard clinical practice. Model-derived ILD volumes on the test set strongly correlated with manually segmented volumes (ρRET=0.67, ρGGO=0.83, ρHC=0.96, ρCONS=0.92; p<0.001). In the SSc-ILD cohort, ILD volumes were significantly larger in progressive (21) versus non-progressive (53) patients (pRET<0.001, pGGO=0.01, pHC<0.001, pCONS=0.003). Moreover, the total ILD volume correlated with FVC%pred (ρ=-0.48, p<0.001) and DLCO%pred (ρ=-0.65, p<0.001). Our AI-based method reliably quantifies ILD pattern volumes on CT, overcoming visual assessment limitations and enabling objective monitoring of progressive SSc-ILD and associated lung function decline to support clinical decision-making.