Artificial intelligence for lung disease quantification in systemic sclerosis-associated interstitial lung disease and other connective tissue disease-associated interstitial lung disease.
Authors
Affiliations (2)
Affiliations (2)
- Department of Diagnostic, Interventional, and Pediatric Radiology, University Hospital of Bern, Inselspital, University of Bern.
- University of Bern, ARTORG Center, Bern, Switzerland.
Abstract
Systemic sclerosis-associated interstitial lung disease (SSc-ILD) is the leading cause of disease-related mortality in systemic sclerosis and the connective tissue disease-associated ILD (CTD-ILD) in which artificial intelligence imaging has advanced most rapidly. Visual high-resolution CT (HRCT) scoring is reader-dependent and limits clinical decision-making. This review summarizes clinically relevant artificial intelligence and radiomics publications from approximately the last 18 months, focusing on quantification, diagnosis and classification, response assessment, and prognosis. The field has moved from visual-score emulation toward outcome-oriented quantitative imaging biomarkers. In SSc-ILD, deep-learning usual interstitial pneumonia (UIP) probability stratifies FVC decline and long-term survival, while whole-chest quantitative imaging biomarkers extend risk prediction beyond lung involvement alone. Automated SSc-specific segmentation, explainable Goh-equivalent scoring, slice-reduced radiomics, and open datasets have strengthened the methodological base. Parallel work in CTD-ILD, RA-ILD, and IIM-ILD shows that artificial intelligence-derived HRCT parameters correlate with DLCO/TLC, predict mortality, and support disease-pattern classification. In broader fibrosing ILD, qCT definitions of progressive pulmonary fibrosis and clinically meaningful CT thresholds provide the most important conceptual advance for future SSc-ILD trials. Artificial intelligence-based CT quantification is becoming a credible adjunct for rheumatology and radiology practice, but the most defensible deployment model is human-in-the-loop decision support. Prospective multicenter validation, protocol harmonization, calibration, version control, and integration into multidisciplinary discussion remain essential before artificial intelligence outputs can be used as treatment-triggering biomarkers.