B79-11 From Fibrosis Burden to Outcome Prediction in Progressive Interstitial Lung Disease: A Systematic Review of CT and Radiograph-based Artificial Intelligence/Deep Learning Models

Abstract Rationale Fibrosing interstitial lung diseases (fILDs), including idiopathic pulmonary fibrosis and other progressive fibrotic phenotypes, exhibit a heterogeneous clinical course characterized by irreversible parenchymal scarring. High-resolution computed tomography (HRCT) is integral for diagnosis and disease monitoring, yet visual interpretation alone is subject to interobserver variability and lacks precise prognostic value. Advances in artificial intelligence (AI), machine learning (ML), and deep learning (DL) have enabled automated quantification of fibrotic burden, offering potential for standardized and predictive imaging biomarkers. We aimed to systematically review and synthesize evidence on the prognostic utility of AI/ML/DL-based imaging models that quantify fibrosis extent on chest CT in patients with fILD, specifically evaluating associations with clinical outcomes such as functional decline, transplant-free survival, or mortality. Methods A systematic search of PubMed, Embase, and Scopus was conducted through October 2025. Studies were included if they applied AI/ML/DL techniques to quantify fibrosis extent on CT and assessed its association with outcomes in fILD. Four eligible studies (n = 1,874 patients) met inclusion criteria. Results All four studies reported significant correlations between AI-derived fibrosis quantification and adverse clinical outcomes. Koh et al. (2024) and Oh et al. (2024) demonstrated that increases in DL-quantified fibrosis and total ILD extent independently predicted transplant-free survival, even when adjusted for forced vital capacity (FVC) decline and CT pattern. Guerra et al. (2024) applied a U-Net convolutional neural network (CNN) to serial CTs, identifying a fibrosis progression threshold (≥4% per year) that strongly predicted mortality or lung transplantation. Swaminathan et al. (2025) showed that higher quantitative lung fibrosis scores were associated with disease progression risk, though predictive power diminished after adjusting for clinical covariates. Conclusion AI/ML imaging models that quantify fibrotic burden on CT provide reproducible, clinically relevant prognostic information in fibrosing ILD. These tools outperform visual assessment in certain contexts and offer promise for earlier risk stratification and treatment optimization. Prospective studies are needed to validate and standardize their integration into routine care. This abstract is funded by: NA