Year: 2025
Type: article
Abstract: Artificial intelligence algorithms can now match or exceed radiologist-level accuracy on narrowly definedimage-classification tasks, yet their integration into real-time clinical imaging workflows remains inconsistent acrossmodalities and diagnostic applications. We evaluated 228 real-time AI medical imaging programmes deployed or pilotedacross radiology departments in France, Italy, and Austria between 2016 and 2024, spanning five modality categories:chest radiograph triage, mammographic screening augmentation, CT pulmonary embolism detection, retinal fundusscreening, and ultrasound point-of-care assistance. A Real-Time AI Imaging Deployment Index (RTAIDI) wasconstructed from five sub-scores -- diagnostic accuracy in clinical conditions, inference latency, radiologist trust andadoption rate, PACS/workflow integration depth, and measurable patient outcome improvement -- with weights fromregression against sustained operational deployment beyond 12 months. RTAIDI correlated with deployment retention atr = +0.84 and discriminated retained from discontinued programmes with an AUC of 0.884. Chest radiograph triageprogrammes scored highest (mean RTAIDI 0.824), while ultrasound point-of-care trailed at 0.596. Only 35.5 percent ofprogrammes exceeded the 0.75 threshold. Diagnostic accuracy carried the largest regression weight (beta = +0.280),followed by radiologist trust (beta = +0.226).
Authors: Hugo Lindberg, Laura Hansen, Ivan Silva
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7203588560