Automatic Detection of COVID-19 from Ultrasound Data

Summary

This repo contains the code for the paper Accelerating Detection of Lung Pathologies with Explainable Ultrasound Image Analysis.

News

• April '24: We released COVID-BLUES, a new dataset of 371 videos from 63 patients collected in a prospective clinical study. Please check out the data and consider using it instead of this one since it is of much higher quality. The paper for this data will appear soon.

Dataset

Feel free to use (and cite) our dataset. We currently have >200 LUS videos labelled with a diagnostic outcome. Moreover, lung severity scores for 136 videos are made available in the dataset_metadata.csv under the column "Lung Severity Score" from Gare et al., 2022. Further clinical information (symptoms, visible LUS patterns etc) are provided for some videos. For details see data/README.md.

If you are looking for more data, please consider using the 40,000 carefully simulated LUS images and paired labels from the paper by Zhao et al. (2024, Communications Medicine). In addition, segmentation labels for a subset of the in vivo data in this repo are also available. For details see data/pulselab/README.md.

NOTE: Please make sure to create a meaningful train/test data split. Do not split the data on a frame-level, but on a video/patient-level. The task becomes trivial otherwise. See the instructions here.

Please note: The founders/authors of the repository take no responsibility or liability for the data contributed to this archive. The contributing sites have to ensure that the collection and use of the data fulfills all applicable legal and ethical requirements.

Contribution

Overview figure about current efforts. Public dataset consists of >200 LUS videos.

Motivation:

From the ML community, ultrasound has gained much less attention than CT and X-Ray in the context of COVID-19. But many voices from the medical community have advocated for a more prominent role of ultrasound in the current pandemic.

Summary

We developed methods for the automatic detection of COVID-19 from Lung Ultrasound (LUS) recordings. Our results show that one can accurately distinguish LUS samples from COVID-19 patients from healthy controls and bacterial pneumonia. Our model is validated on an external dataset (ICLUS) where we achieve promising performance. The CAMs of the model were validated in a blinded study by US experts and found to highlight relevant pulmonary biomarkers. Using model uncertainty techniques, we can further boost model performance and find samples which are likely to be incorrectly classified. Our dataset complements the current data collection initiaves that only focus on CT or X-Ray data.

Evidence for ultrasound

Ultrasound is non-invasive, cheap, portable (bedside execution), repeatable and available in almost all medical facilities. But even for trained doctors detecting patterns of COVID-19 from ultrasound data is challenging and time-consuming. Since their time is scarce, there is an urgent need to simplify, fasten & automatize the detection of COVID-19.

• LUS is more sensitive than X-Ray in diagnosing COVID-19
• COVID-19 outbreak: less stethoscope, more ultrasound
• Ultrasound can evidence the same symptoms as CT: (Point-by-point correspondance of CT and ultrasound findings through COVID-19 disease process)

Learn more about the project

• Read our manuscript
• Read our blogpost

Installation

Ultrasound data

Find all details on the current state of the database in the data folder.

Deep learning model (pocovidnet)

Find all details on how to reproduce our experiments and train your models on ultrasound data in the pocovidnet folder.

Web interface (pocovidscreen)

Find all details on how to get started in the pocovidscreen folder.

Contact

• If you experience problems with the code, please open an issue.
• If you have questions about the project, please reach out: jannis.born@gmx.de.

Citation

An abstract of our work was published in Thorax as part of the BTS Winter Meeting 2021. The full paper is available via the COVID-19 special issue of Applied Sciences. Please cite these in favor of our deprecated POCOVID-Net preprint.

Please use the following bibtex entry to cite this dataset:

@article{born2021accelerating,
  title={Accelerating Detection of Lung Pathologies with Explainable Ultrasound Image Analysis}, 
  author={Born, Jannis and Wiedemann, Nina and Cossio, Manuel and Buhre, Charlotte and Brändle, Gabriel and Leidermann, Konstantin and      Aujayeb, Avinash and Moor, Michael and Rieck, Bastian and Borgwardt, Karsten}, 
  volume={11}, ISSN={2076-3417}, 
  url={http://dx.doi.org/10.3390/app11020672}, 
  DOI={10.3390/app11020672}, 
  number={2}, 
  journal={Applied Sciences}, 
  publisher={MDPI AG}, 
  year={2021}, 
  month={Jan}, 
  pages={672}
}