ChatGPT-Assisted Image Interpretation for Inflammatory Bowel Diseases: Ulcerative Colitis and Crohn's Disease

Area of Science:

• Gastroenterology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Differentiating ulcerative colitis (UC) from Crohn's disease (CD) is clinically challenging.
• Artificial intelligence (AI) shows promise in endoscopic diagnosis, but large language models (LLMs) for inflammatory bowel disease (IBD) need validation.
• Current diagnostic methods for UC and CD can be complex and require expert interpretation.

Purpose of the Study:

• To evaluate ChatGPT's ability to distinguish between UC and CD using colonoscopy (CS) images.
• To assess the impact of clinical information on ChatGPT's diagnostic performance for UC and CD.
• To compare ChatGPT's diagnostic accuracy with that of IBD specialists.

Main Methods:

• Retrospective analysis of 386 UC and 161 CD patients with active disease undergoing CS.
• Selection of representative endoscopic images of severe disease activity at initial flare.
• ChatGPT was tasked with classifying UC/CD and estimating UC probability using images alone and with clinical data; IBD specialists performed classification for comparison.

Main Results:

• ChatGPT achieved 75.6% diagnostic accuracy with images alone, increasing to 87.4% with clinical information.
• IBD specialists achieved 84.9% accuracy with images alone, rising to 88.7% with clinical information.
• ChatGPT's area under the receiver operator curve (AUC) improved from 0.750 (images only) to 0.948 (images + clinical data).

Conclusions:

• ChatGPT demonstrates capability in differentiating UC from CD.
• The integration of clinical information significantly enhances ChatGPT's diagnostic accuracy for IBD.
• While promising for clinical application, ChatGPT's current accuracy is slightly lower than that of IBD specialists.