COVID Bluepoint Lung Ultrasound (BLUES) dataset

The COVID-BLUES dataset contains bluepoint-specific lung ultrasound videos recorded in a prospective study on COVID-19 suspects between February and May 2021. The study included 63 patients (33 COVID-positive and 30 COVID-negative), with the inclusion criteria being symptoms of a lung infection. The findings of the study and a description of the dataset have been published in Wiedemann et al. (2025), IEEE Journal of Biomedical and Health Informatics. Please cite that paper if you use this data.

The dataset comprises lung ultrasound recordings at six BLUE points for each patient, a database of clinical variables, and severity assessment by medical experts. These parts are described more in detail in the following:

Lung ultrasound videos

Our dataset contains a set of lung ultrasound videos that were recorded in a standardized setting. All recordings were conduced with a Philips Lumify US device with a convex probe. The videos show one position for 60 - 100 frames. In contrast to other ultrasound datasets, the US videos were taken according to the standard BLUE protocol by Lichtenstein (2014), where six points (three for each lung) are scanned. This leads to 6 videos per patient. Since some points could not be scanned for individual patients, there are sometimes less videos per patient, resulting in a total of 362 videos corresponding to 31,746 frames. The videos are labeled according to the schema patient_<PATIENT-ID>_<BLUEPOINT>, e.g. patient_1_L3.mp4, where the blue points are denoted as L1, L2, L3 (left lung), R1, R2, R3 (right lung). In some cases, two videos are available for the same blue point. These cases are labeled patient_<PATIENT-ID>_<BLUEPOINT_2>; see for example patient_15_L2_2.mp4.

All videos are available as .mp4 files in the folder lus_videos. NOTE: If you split videos into frames for your ML model, make sure to not mix frames from the same patient across training and testing data.

Clinical variables

The file clinical_variables.csv contains medical information for each patient. This includes patient characteristics, symptoms, comorbidities, blood test data, vital parameters, and the PCR test result testing for COVID-19. An explanation for the variables and their respective unit is provided in file variable_explanations.csv.

Severity assessment

Last, two medical experts rated the severity of the lung infection visible on the LUS videos. They were asked to follow the scoring scheme by Soldati et al, 2020, with scores from 0-3.

The medical experts further noted whether B-lines and / or A-lines are visible in the video. The severity scores and B-line/A-line informaiton are provided per video in the file severity.csv.

Huggingface

The dataset is also available on Huggingface here. Install requirements:

pip install datasets==3.5.0
pip install av
pip install torchvision

Access dataset with:

from datasets import load_dataset
dataset = load_dataset("jannisborn/COVID-BLUES", split="train")
# Select one video
dataset[0]["video"] # torchvision VideoReader

# Load clinical variables
variables = load_dataset("jannisborn/COVID-BLUES", data_files="clinical_variables.csv")["train"]
# Load severity scores
severity = load_dataset("jannisborn/COVID-BLUES", data_files="severity.csv")["train"]

License

The COVID-BLUES dataset is available via the CC BY-NC-ND 4.0 license.

Related dataset

If you are looking for more, similar data, have a look at the POCUS dataset which contains >200 videos collected from various public sources and divided in three categories (COVID-19, bacterial pneumonia, healthy).

Citation

The dataset was recorded by Maastricht University Medical Center (UMC+) in the Netherlands.

If you use the COVID-BLUES dataset in your research, please cite the following:

@article{wiedemann2025covid,
  author={Wiedemann, Nina and Boer, Dianne de Korte-de and Richter, Matthias and van de Weijer, Sjors and Buhre, Charlotte and Eggert, Franz A. M. and Aarnoudse, Sophie and Grevendonk, Lotte and Röber, Steffen and Remie, Carlijn M.E. and Buhre, Wolfgang and Henry, Ronald and Born, Jannis},
  journal={IEEE Journal of Biomedical and Health Informatics}, 
  title={COVID-BLUeS - A Prospective Study on the Value of AI in Lung Ultrasound Analysis}, 
  year={2025},
  volume={29},
  number={9},
  pages={6301-6310},
  keywords={COVID-19;Artificial intelligence;Videos;Lungs;Ultrasonic imaging;Analytical models;Pathology;Medical diagnostic imaging;Data models;Training;Lung ultrasound;computer vision;COVID-19},
  doi={10.1109/JBHI.2025.3543686}
}