This Repository is a quasi-fork of the original repository published by Yu et al. (2022). The original authors did not publish their code on GitHub, but as a Zip-file via Google Drive. See the last section of this README for the full reference.
Repository referenced in the paper "Measuring Progress in Automatic Chest X-Ray Radiology Report Generation". This repository provides code for computing metric scores for radiology report evaluation. The metrics are:
- BLEU
- BERTscore
- CheXbert labeler vector similarity
- RadGraph entity and relation F1
- Composite metric RadCliQ
To install the dependencies, run the following command with Python 3.7:
pip install -r requirements.txt
Ground Truth and Predicted reports must be arranged in the same order in a column named "report" in two CSV files. The CSVs should also contain a corresponding "study_id" column that contains unique identifies for the reports.
In config.py, set GT_REPORTS and PREDICTED_REPORTS to paths to the CSVs.
Set OUT_FILE to the desired path for the output metric scores.
To compute the CheXbert vector similarity (s_emb) metric score, download the
CheXbert model checkpoint here.
Set CHEXBERT_PATH in config.py to the path to the downloaded checkpoint.
The code for computing the CheXbert metric score is adapted from stanfordmlgroup/CheXbert.
Paper (Accepted to EMNLP 2020): https://arxiv.org/abs/2004.09167.
To compute the RadGraph metric score, download the RadGraph model checkpoint
from PhysioNet here.
The checkpoint file can be found under the "Files" section at path
models/model_checkpoint/.
Set RADGRAPH_PATH in config.py to the path to the downloaded checkpoint.
The code for computing the RadGraph metric score is adapted from dwadden/dygiepp. Note: You need to apply for credentialed access to RadGraph on PhysioNet.
Paper (Accepted to NeurIPS 2021): https://arxiv.org/abs/2106.14463.
from CXRMetric.run_eval import calc_metric
calc_metric(gt_reports, predicted_reports, out_file)
This repository is made publicly available under the MIT License.
If you are using this repo, please cite this paper:
@techreport{yuEvaluatingProgressAutomatic2022,
type = {preprint},
title = {Evaluating {Progress} in {Automatic} {Chest} {X}-{Ray} {Radiology} {Report} {Generation}},
url = {http://medrxiv.org/lookup/doi/10.1101/2022.08.30.22279318},
language = {en},
urldate = {2023-06-02},
institution = {Radiology and Imaging},
author = {Yu, Feiyang and Endo, Mark and Krishnan, Rayan and Pan, Ian and Tsai, Andy and Reis, Eduardo Pontes and Fonseca, Eduardo Kaiser Ururahy Nunes and Ho Lee, Henrique Min and Abad, Zahra Shakeri Hossein and Ng, Andrew Y. and Langlotz, Curtis P. and Venugopal, Vasantha Kumar and Rajpurkar, Pranav},
month = aug,
year = {2022},
doi = {10.1101/2022.08.30.22279318},
}