LLMs can generate more comprehensive and factual clinical indications for radiologists than referring clinicians.
Key Details
- 1Study evaluated LLM performance using deidentified records of 28,313 UCSF patients and 77,626 imaging exams.
- 2Both open-source (Qwen 2.5-7B Instruct) and proprietary (Claude 3.5 Sonnet) LLMs were benchmarked.
- 320 radiologists rated LLM vs. clinician-generated indications on comprehensiveness, factuality, and usefulness.
- 4Claude 3.5 Sonnet scored highest for comprehensiveness (37.14%) and factuality (68.05%), outperforming clinicians (6.64% and 50%).
- 5Comprehensiveness was the key driver for radiologists’ rankings; LLMs consistently outranked referring clinicians.
- 6Editorial stresses need for transparency, monitoring, and human oversight in deploying LLMs in radiology.
Why It Matters
Improving the completeness of clinical indications can reduce diagnostic error and improve radiology workflow. The success of LLMs in this domain highlights real-world potential for AI to augment radiologist-clinician communication, though careful implementation is needed.
Source
AuntMinnie
Related News
•AuntMinnie
Open-Source AI Platform Powers Pediatric Bone and Fetal MRI Pipelines
Open-source AI platform successfully integrated and sustained two pediatric imaging pipelines for bone length and fetal MRI at a children's hospital.
•Radiology Business
Meta Funds Ohio State’s Radiology AI Innovation Projects
Meta will fund Ohio State researchers to explore AI-driven advancements in radiology, including chest X-ray reporting and disease prediction.
•AuntMinnie
ChatGPT-5 Performs Poorly at Detecting Distal Radius Fractures on X-rays
ChatGPT-5 missed about half of distal radius fractures on x-rays, showing limited diagnostic capability in a real-world scenario.