# From Text to Image: Exploring GPT-4Vision's Potential in Advanced Radiological Analysis across Subspecialties

**Authors:** Felix Busch, Tianyu Han, Marcus Makowski, Daniel Truhn, Keno Bressem, Lisa Adams  
**arXiv:** [2311.14777](https://arxiv.org/abs/2311.14777)  
**Date:** 2023-11-24  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2311.14777v1/

## Key points

- Study by Felix Busch, Tianyu Han, Marcus Makowski, Daniel Truhn, Keno Bressem, and Lisa Adams focuses on GPT-4Vision in advanced radiological analysis.
- GPT-4Vision can recognize radiological features from images, enhancing diagnostic potential over text-based descriptions.
- Introduction emphasizes the success of GPT-4 in the medical field and its application in radiology for text-based tasks.
- GPT-4V's visual capabilities introduce a new dimension to imaging analysis in radiology.
- Research aims to explore how incorporating both textual descriptions and visual data with GPT-4Vision can revolutionize radiological practices for enhanced diagnostic accuracy and efficiency.

## Summary

The study by Felix Busch, Tianyu Han, Marcus Makowski, Daniel Truhn, Keno Bressem, and Lisa Adams delves into the potential of GPT-4Vision in advanced radiological analysis across subspecialties. The research evaluates and compares GPT-4 and GPT-4Vision for radiological tasks, indicating that GPT-4Vision has the ability to recognize radiological features from images. This capability enhances its diagnostic potential over traditional text-based descriptions. The introduction highlights the significance of GPT-4 in the medical field and showcases its success in achieving high accuracy on exams like the USMLE. In radiology specifically, GPT has been utilized for various text-based applications such as question evaluation and report structuring. However, with the introduction of GPT-4V's visual capabilities, a new dimension is added to imaging analysis. The study aims to explore how these advancements can impact and potentially revolutionize radiological practices by incorporating both textual descriptions and visual data into the analysis process. By utilizing this multimodal approach, researchers seek to determine the role of GPT-4Vision in enhancing diagnostic accuracy and efficiency within radiology.