Evaluating the Potential of Generative AI in Digestive Endoscopy: ChatGPT-4 for Advanced Image Analysis in Gastroenterology

Aims In recent years, there has been exponential growth in the application of artificial intelligence in healthcare, with development of several tools based on large language models (LLMs). Recently, there is a growing interest in the clinical implementation of this tools, with recent versions incorporating imaging analysis capability. Nevertheless, their potential utility in Gastroenterology remains largely unexplored. This study aimed to assess ChatGPT-4’s performance in interpreting images from different Gastroenterology procedures.

Methods A total of 740 images from five procedures – capsule endoscopy (CE), device-assisted enteroscopy (DAE), endoscopic ultrasound (EUS), digital single-operator cholangioscopy (D-SOC) and high-resolution anoscopy (HRA)–were included and analyzed by ChatGPT-4 using a personalized predefined prompt. ChatGPT-4 predictions were compared to gold standard diagnoses. Statistical analyses included accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the curve (AUC) [1] [2] [3].

Results For CE, ChatGPT-4 demonstrated accuracies ranging from 50.0% to 90.0%, with AUCs between 0.50 and 0.90. For DAE, the model achieved an accuracy of 67.0% (AUC 0.67). For EUS, the system showed AUCs of 0.488 and 0.550 for the detection of pancreatic cystic and solid lesions, respectively. The LLM differentiated benign from malignant biliary strictures with an AUC of 0.550. For HRA, ChatGPT-4 overal accuracy ranged between 47.5% and 67.5%.

Conclusions This multicentric study highlights that while ChatGPT4 demonstrates notable capabilities in text analysis, its performance in image analysis remains suboptimal, underscoring the need for significant improvements before clinical adoption.

Publikationsverlauf

Artikel online veröffentlicht:
27. März 2025

• References

• 1 Koga S., Du W.. From text to image: challenges in integrating vision into ChatGPT for medical image interpretation. Neural Regen Res 2025; 20: 487-488
• 2 Sonoda Y., Kurokawa R., Nakamura Y., Kanzawa J., Kurokawa M., Ohizumi Y., Gonoi W., Abe O.. Diagnostic perfor-mances of GPT-4o, Claude 3 Opus, and Gemini 1.5 Pro in "Diagnosis Please" cases Jpn J Radiol. 2024; doi:10.1007/s11604-024-01619-y.
• 3 Lahat A., Shachar E., Avidan B., Shatz Z., Glicksberg B.S., Klang E.. Evaluating the use of large language model in identi-fying top research questions in gastroenterology. Sci Rep 2023; 13: 4164