AI-Based HRCT Quantification in Connective Tissue Disease-Associated Interstitial Lung Disease

<b>Background:</b> Interstitial lung disease (ILD) is a frequent and potentially progressive manifestation in patients with connective tissue diseases (CTDs). Accurate and reproducible quantification of parenchymal abnormalities on high-resolution computed tomography (HRCT) is essential for evaluating treatment response and monitoring disease progression, particularly in complex cases undergoing antifibrotic therapy. Artificial intelligence (AI)-based tools may improve consistency in visual assessment and assist less experienced radiologists in longitudinal follow-up. <b>Methods:</b> In this retrospective study, 48 patients with CTD-related ILD receiving antifibrotic treatment were included. Each patient underwent four HRCT scans, which were evaluated independently by two radiologists (one expert, one non-expert) using a semi-quantitative scoring system. Percentage estimates of lung involvement were assigned for four parenchymal patterns: hyperlucency, ground-glass opacity (GGO), reticulation, and honeycombing. AI-based analysis was performed using the Imbio Lung Texture Analysis platform, which generated continuous volumetric percentages for each pattern. Concordance between AI and human interpretation was assessed, along with mean absolute error (MAE) and inter-reader differences. <b>Results:</b> The AI-based system demonstrated high concordance with the expert radiologist, with an overall agreement of 81% across patterns. The MAE between AI and the expert ranged from 1.8% to 2.6%. In contrast, concordance between AI and the non-expert radiologist was significantly lower (60-70%), with higher MAE values (3.9% to 5.2%). McNemar's and Wilcoxon tests confirmed that AI aligned more closely with the expert than the non-expert reader (<i>p</i> < 0.01). AI proved particularly effective in detecting subtle changes in parenchymal burden during follow-up, especially when visual interpretation was inconsistent. <b>Conclusions:</b> AI-driven quantitative imaging offers performance comparable to expert radiologists in assessing ILD patterns on HRCT and significantly outperforms less experienced readers. Its reproducibility and sensitivity to change support its role in standardizing follow-up evaluations and enhancing multidisciplinary decision-making in patients with CTD-related ILD, particularly in progressive fibrosing cases receiving antifibrotic therapy.