Home

Awesome

MIRQI - Evaluation metrics for radiological report generation

MIRQI-r,p,f provides a quanitative assessment of the quality of generated radiology reports.

Prerequisites

  1. Make the virtual environment (python 3):

    conda env create -f environment.yml

  2. Install NLTK data:

    python -m nltk.downloader universal_tagset punkt wordnet

  3. Download the GENIA+PubMed parsing model:

>>> from bllipparser import RerankingParser
>>> RerankingParser.fetch_and_load('GENIA+PubMed')

Usage

Place reports in a headerless, single column csv {reports_file_path}. Each report must be contained in quotes if (1) it contains a comma or (2) it spans multiple lines. See the sample csv files in the folder for an example.

python evaluate.py --reports_path_gt {gt_reports_file_path} -- report_path_cand {generated_reports_file_path}

Run python evaluate.py --help for descriptions of all of the command-line arguments.

Contributions

This repository builds upon the work of NegBio and CheXpert.

This tool was developed by Xiaosong Wang.

Citing

If you're using MIRQI evaluation metrics, please cite #this paper:

@article{Zhang_Wang_Xu_Yu_Yuille_Xu_2020, 
title={When Radiology Report Generation Meets Knowledge Graph}, 
journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
author={Zhang, Yixiao and Wang, Xiaosong and Xu, Ziyue and Yu, Qihang and Yuille, Alan and Xu, Daguang},
url={https://ojs.aaai.org/index.php/AAAI/article/view/6989}, DOI={10.1609/aaai.v34i07.6989}, 
volume={34}, number={07}, year={2020}, month={Apr.}, pages={12910-12917} }