Skip to content

Impression Generation in Radiology Reports through a Multi-Agent System

License

Notifications You must be signed in to change notification settings

Fang85/multi-agent-LLM-impression-generation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Impression Generation in Radiology Reports through a Multi-Agent System

Overview

This project generate impression based on the provided procedure name and findings of a radiology report. The framework consists of three specialized agents:

  • Report retriever: a retrieval agent that searches for similar reports from an external database.
  • Radiologist: a generation agent that utilizes LLMs to generate or update the impression.
  • Reviewer: a review agent that assesses the consistency between the generated impression and provided findings.

image

Requirements

  1. LLM: Deployed using Ollama. In the main code, llama3:70b is used.
  2. RAG: pip install langchain langchain-community sentence_transformers faiss-cpu -U langchain-huggingface
  3. Evaluation: pip install nltk sacrebleu rouge-score bert-score

Input file

The input file required for the main code is not provided in this repository. The file includes the following columns:

  • ID: Index
  • ACC: accession number
  • modality: modality used for the imaging study
  • ProcedureNM: procedure name
  • FINDINGS: findings section
  • IMPRESSION_CLEAN: cleaned impression section

Notes: ACC and modality are used as metadata in the vector database. However, they are not essential for generating the impression. If desired, the code can be modified to exclude these columns.

About

Impression Generation in Radiology Reports through a Multi-Agent System

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages