Multimodal LLMs achieved up to 94% accuracy for scoliosis detection on spine x-rays, but struggled with lumbar stenosis on MRI.
Key Details
- 1Five multimodal LLMs tested: Grok 2, 3, 4, ChatGPT 4o, Gemini 1.5 Flash.
- 2171 spine x-rays (100 scoliosis, 71 normal) and 200 lumbar spine MRIs (100 severe stenosis, 100 normal) used in the study.
- 3Best x-ray result: Grok 4 with 94.2% accuracy for scoliosis detection; best MRI result: Gemini at 60% for stenosis.
- 4ChatGPT 4o showed better confidence calibration when incorrect, considered a 'superior metacognitive capability.'
- 5Authors emphasize LLMs not ready for clinical diagnosis; highlight potential for patient education in obvious cases.
- 6Study published in World Neurosurgery on May 2, 2024.
Why It Matters
As patients increasingly use commercial LLMs for medical advice, understanding their capabilities and risks in radiology is crucial. These results highlight both the promise and current limitations of generalist AI in medical image interpretation, especially for more subtle pathologies.
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.