During the COVID-19 pandemic, rapid and reliable image interpretation became critical, particularly when molecular testing was limited or delayed. This study developed an AI-powered clinical decision support system for chest CT that goes beyond detecting COVID-19 by simultaneously distinguishing it from other pulmonary infections and quantifying disease involvement.

The model was developed through the Imaging COVID-19 AI initiative, a large-scale European collaboration involving 20 institutions across seven countries. The dataset comprised more than 2,800 CT scans, with an institution-level split providing external evaluation across heterogeneous scanners and acquisition protocols.

The deep learning classification model achieved micro- and macro-average AUCs of 0.93 and 0.91, respectively. For COVID-19 versus other cases, it reached 87% sensitivity, 94% specificity and 90% accuracy.

Importantly, the system also segments infectious lung opacities, quantifies affected lung volume and generates a CT severity score, integrating these outputs with diagnostic probabilities into an automated visual report.

Unlike many early COVID-19 AI studies based on smaller or geographically restricted datasets, this work emphasizes multicenter generalizability and real-world deployment. The resulting application was used for more than 7,000 analyses across four continents and deployed with PACS integration in over 10 institutions, demonstrating how collaborative medical imaging AI can move from model development toward scalable clinical implementation.