Post by Jan Beger

Adding AI to prostate MRI reading in real time does not significantly improve cancer detection, regardless of how experienced the radiologist is. 1️⃣ Six radiologists at three experience levels read 100 prostate MRI scans, with and without a commercial AI decision-support tool. 2️⃣ Readers interpreted images first, then reviewed AI output before finalising, designed to simulate real-world clinical practice. 3️⃣ Overall cancer detection accuracy did not differ significantly between AI-assisted and unassisted readings. 4️⃣ Residents changed their Prostate Imaging Reporting and Data System scores most after AI input: 19 changes versus 7 for expert readers. 5️⃣ Of 8 cases upgraded to a higher Prostate Imaging category by AI, only 1 was actually clinically significant cancer. 6️⃣ Residents increased their overall suspicion estimates by roughly 9% with AI, but final category assignments barely shifted. 7️⃣ Expert readers showed almost no change with AI, suggesting limited benefit for more experienced radiologists. 8️⃣ AI offered a modest benefit in targeting high-grade cancers among experts and avoiding unnecessary biopsies among residents. 9️⃣ The same AI system performed better as a standalone tool in other studies, pointing to human-AI interaction as a limiting factor. 🔟 User trust, cognitive bias, and workflow design may matter as much as algorithm accuracy in real-world AI integration. ✍🏻 Andrea Ponsiglione, Giuseppe Di Costanzo, Alfonso Maria Ponsiglione, Ciro Riccio, Andrea Rinaldo, Anna Giacoma Tucci, Lorenzo Pinto, Luigi Palumbo, Francesca Angelone, Francesco Amato, Arnaldo Stanzione, Renato Cuocolo, Rossano Girometti, Anwar Padhani, Massimo Imbriaco. Concurrent AI-human interaction in prostate cancer MRI interpretation: More hype than help? European Radiology Experimental. 2026. DOI: 10.1186/s41747-026-00695-1 | Open Access