# ImpressionGPT: An Iterative Optimizing Framework for Radiology Report Summarization with ChatGPT

**Authors:** Chong Ma, Zihao Wu, Jiaqi Wang, Shaochen Xu, Yaonai Wei, Zhengliang Liu, Lei Guo, Xiaoyan Cai, Shu Zhang, Tuo Zhang, Dajiang Zhu, Dinggang Shen, Tianming Liu, Xiang Li  
**arXiv:** [2304.08448](https://arxiv.org/abs/2304.08448)  
**Date:** 2023-04-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.08448v1/

## Key points

- Text summarization is a significant research focus in natural language processing (NLP), especially in the medical domain.
- Automatic impression generation (AIG) has become an area of interest in NLP research for radiology reports.
- Large language models (LLMs) like ChatGPT have shown strong generalization capabilities and performance, making them potential candidates for AIG in radiology reports.
- ImpressionGPT leverages the in-context learning capability of LLMs by constructing dynamic prompts using domain-specific individualized data to learn contextual knowledge from semantically similar examples from existing data.
- An iterative optimization algorithm performs automatic evaluation on generated impressions results and composes corresponding instruction prompts to further optimize the model.
- The proposed ImpressionGPT model achieves state-of-the-art performance on both MIMIC-CXR and OpenI datasets without requiring additional training data or fine-tuning LLMs.
- This work presents a paradigm for localizing LLMs that can be applied in a wide range of similar application scenarios, bridging the gap between general-purpose LLMs and specific language processing needs of various domains.
- The use of pre-trained language models within the biomedical domain has also shown potential for various downstream tasks.

## Summary

Text summarization has become a significant research focus in the field of natural language processing (NLP), particularly in the medical domain. In radiology reports, the impression section serves as a summary of the entire report description and is critical for communication between radiologists and other physicians. Automatic impression generation (AIG) has become an area of interest in NLP research, with earlier studies focusing on using sequence-to-sequence methods to generate impressions. However, these methods have limitations such as factual errors and poor generalization performance. With the emergence of large language models (LLMs) like ChatGPT, which have shown strong generalization capabilities and performance, there is potential for their use in AIG for radiology reports. To address this limitation, researchers propose ImpressionGPT, which leverages the in-context learning capability of LLMs by constructing dynamic contexts using domain-specific individualized data. This dynamic prompt approach enables the model to learn contextual knowledge from semantically similar examples from existing data. Additionally, an iterative optimization algorithm performs automatic evaluation on generated impressions results and composes corresponding instruction prompts to further optimize the model. The proposed ImpressionGPT model achieves state-of-the-art performance on both MIMIC-CXR and OpenI datasets without requiring additional training data or fine-tuning LLMs. This work presents a paradigm for localizing LLMs that can be applied in a wide range of similar application scenarios, bridging the gap between general-purpose LLMs and specific language processing needs of various domains. The use of pre-trained language models within the biomedical domain has also shown potential for various downstream tasks. Overall, text summarization has become a significant research focus in NLP, particularly in the medical domain. The proposed ImpressionGPT model offers a promising solution for AIG in radiology reports without requiring additional training data or fine-tuning LLMs.