AI pinpoints radiology residents’ educational gaps, helping expand personalized training

September 3, 2026 - 11:56

New research shows how artificial intelligence can step in to improve radiology residency programs by spotting weaknesses in clinical experience and reinforcing topics that residents miss most. The study, led by a team at the University of Michigan, used machine learning to analyze hundreds of thousands of radiology reports and imaging requests logged by trainees over several years.

Instead of relying on end-of-rotation exams or subjective feedback, the AI system tracked each resident's actual case mix, including the types of studies they read, the body parts imaged, and the urgency of each exam. It then compared that data against a national benchmark of expected exposure for each training level. The result was a detailed profile showing where a resident might be under-seeing, say, pediatric head CTs or abdominal MRIs, even if they had logged plenty of total scans.

The authors say this approach turns a vague problem into a fixable one. Program directors can now see, at a glance, which residents need more time in certain modalities or which ones should be assigned more complex cases. The AI also flagged when a resident repeatedly misread a specific finding, like a subtle fracture or a lung nodule, allowing faculty to revisit those patterns in a targeted way.

What makes this different from prior efforts is that it does not require extra work from residents or faculty. The system quietly mines data already stored in the hospital's electronic health record. In a field where burnout is high and time is tight, that hands-off design matters.

The researchers caution that AI is not replacing human mentorship. Instead, it acts as a support tool, giving educators a clearer map of each learner's path. They are now testing the model at three other institutions to see if it holds up across different training environments. Early results suggest that even small changes, like shifting a resident's elective schedule based on AI feedback, can close knowledge gaps within a single academic year.