LLMs show early promise for automated feedback in radiology education, but systematic review finds limited, preliminary evidence

Large language models (LLMs) for automated feedback in medical imaging education show preliminary gains in report quality, diagnostic accuracy, and learner satisfaction, but a systematic review of 7 studies finds early-stage, limited evidence; human oversight remains essential.

• Systematic review of 7 studies, all involving radiology residents (one also included fellows); all were single-site and small-scale.
• Fine-tuned LLMs outperformed general-purpose models, but agreement with expert consensus was variable; learners preferred a hybrid human-AI feedback model.
• Evidence is preliminary and insufficient to determine educational effectiveness; larger, multicenter studies with standardized methods are needed before adoption.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.