Exploring Medical Caption Generation through OpenAI’s ChatGPT-4 Model: A PRISMA Review

Abstract

This study explores the importance of the ChatGPT-4 model in medical caption generation, its advantages, applications, and limitations, using a PRISMA strategy on Medline and PubMed medical datasets to extract relevant studies from over a year ago concerning “ChatGPT” and “Medical Report Generation.” The search employed keywords such as (“ChatGPT” OR “GPT model”) AND (“medical caption generation” OR “medical image captioning” OR “radiology captioning”). The PRISMA search strategy led to the selection of seven promising papers. We conducted a brief comparison among the selected papers, taking into account their key focus, the datasets used, the models evaluated, the research results, and the challenges highlighted. Additionally, ChatGPT4’s performance was evaluated by uploading sample medical images from different dataset modalities such as PathVQA, VQA-Med 2020, RadioGraphy Captions (RGC), and Radiology Objects in Context (ROCO) to establish whether it could generate coherent and contextually correct medical captions as true outputs and correctly answer medical questions with output performance BLEU = 0.5012 and ROUGE-L = 0.8000 scores. This study provides state-of-the-art evidence that ChatGPT demonstrates remarkable performance in report generation and answering medical questions under supervision.