ChatGPT-4o and ChatGPT-5 matched or surpassed nuclear medicine experts in diagnosing neurodegenerative diseases using textual FDG-PET/CT scan descriptions.
Key Details
- 1University of Cologne team tested ChatGPT-4o and ChatGPT-5 on 100 F-18 FDG-PET/CT brain scan reports.
- 2Models achieved median diagnostic agreement scores of 1.00 against expert interpretations.
- 3ChatGPT-4o identified the correct main diagnosis in 86% of cases, ChatGPT-5 in 89%.
- 4Performance was highest in typical cases (e.g., Alzheimer's disease), lower in complex or atypical presentations.
- 5No imaging data or specific fine-tuning was used; models relied on general training.
- 6Reproducibility from run to run was 75% for ChatGPT-4o and 55% for ChatGPT-5 in a subset.
Why It Matters
This study suggests that large language models can align with expert nuclear medicine interpretation based solely on textual imaging descriptions, potentially augmenting workflow efficiency and consistency in neuroimaging. It highlights opportunities and limitations for LLMs in aiding diagnostic processes, especially when combined with automated image analysis.
Source
AuntMinnie
Related News
•AuntMinnie
Study: Computer Vision Models Best LLMs in Chest CT Breast Abnormality Detection
Computer vision models (CVMs) surpass large language models (LLMs) in accurately labeling incidental breast abnormalities on chest CT scans.
•Cardiovascular Business
Radiology Maintains Lead in FDA-Cleared AI Algorithms, Cardiology Follows
Radiology remains the top specialty for FDA-cleared AI, with cardiology as a strong second, particularly in cardiovascular imaging.
•AuntMinnie
Deep Learning Models Rival Radiologists for Pancreatic Cancer Detection on CT
Deep-learning models achieved comparable or superior accuracy to experienced radiologists in detecting pancreatic cancer on CT scans, especially for small tumors.