Overview

• The study published in the Journal of the American College of Surgeons compared ChatGPT-4’s output against manual radiologist reviews of 3,198 MRI and CT scans from 991 patients monitored between 2010 and 2024.
• ChatGPT-4 matched gold-standard manual review across nine high-risk factors, achieving 97% accuracy on solid components and up to 99% on calcific lesions.
• Accuracy rates for other critical measures included 92% for cyst size and 97% for main pancreatic duct diameter.
• Researchers highlighted AI’s efficiency and cost savings over labor-intensive chart reviews, freeing clinicians to focus on analysis and quality assurance.
• Authors plan to extend this proof-of-concept into predictive models for malignant progression but caution that reliance on a single AI model and dataset may limit broader applicability.