Radiology in the age of AI

Meghann Maiellano, 41, endured weeks of undiagnosed abdominal pain after emergency CT scans failed to clarify her condition. Her experience at Strong Memorial Hospital in Rochester, New York, underscored delays in radiology follow-ups, leaving her in pain and uncertainty. The University of Rochester Medical Center (URMC) serves over 1 million patients annually and has integrated AI into its radiology workflows to address inefficiencies. Tools like Microsoft’s PowerScribe One assist radiologists by drafting concise ‘impression’ summaries from dictated findings, acting as a TLDR for lengthy reports. Another feature, Dragon Copilot, condenses years of prior imaging reports into clickable summaries, linking back to original sources for verification. Sean Cleary, a cardiothoracic radiologist and vice chair of informatics at URMC, clarified that AI does not replace human interpretation. PowerScribe One generates draft impressions based on radiologists’ dictated findings, while Dragon Copilot organizes historical data without independent analysis. Radiologists review, edit, and approve all AI-assisted content before finalizing reports. The system aims to reduce workloads in a strained healthcare environment, but risks persist if errors go unchecked. Cleary emphasized that AI tools are limited to summarizing or retrieving information, not diagnosing. For patients like Maiellano, faster access to imaging and AI-enhanced workflows could mitigate delays—but human oversight remains critical to patient safety. URMC’s approach reflects a cautious adoption of AI, balancing efficiency gains with the need for accuracy. While tools like PowerScribe One and Dragon Copilot streamline radiology processes, they function as assistants rather than replacements for trained professionals. The goal is to improve workflows without compromising diagnostic precision.