AI LLMs notably improve diagnostic accuracy for less experienced brain MRI readers, with diminishing benefits for experts.
Key Details
- 1Study involved 12 readers (4 neuroradiologists, 4 radiology residents, 4 neurology/neurosurgery residents) analyzing 40 confirmed brain MRI cases.
- 2Readers provided initial diagnoses, then revised after receiving AI suggestions from GPT-4.1, Gemini 2.5 Pro, and DeepSeek-R1, based on their written inputs.
- 3AI assistance increased top-3 diagnostic accuracy by 19.4 percentage points for neurology/neurosurgery residents and 14.7 points for radiology residents (both statistically significant, p < 0.001).
- 4Board-certified neuroradiologists improved by only 4.4 percentage points (not statistically significant, p = 0.086).
- 5LLM model accuracy paralleled the detail and correctness of reader inputs; more experienced readers provided higher quality inputs.
Why It Matters
These results highlight that LLM-based AI tools may be especially valuable for augmenting less experienced clinicians, potentially narrowing accuracy gaps in diagnostic radiology. Targeted implementation could improve clinical support, though value for expert radiologists in routine cases remains limited.
Source
AuntMinnie
Related News
•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
AI Triage Tool Speeds Up ICH Head CT Turnaround in ED
AI-based triage for intracranial hemorrhage on head CT reduced emergency department report turnaround by 13 minutes.
•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.