Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Mar 7;12(1):402.
doi: 10.1038/s41597-025-04709-2.

Manual segmentation of opacities and consolidations on CT of long COVID patients from multiple annotators

Diedre S Carmo  1 , Alejandro A Pezzulo  2 , Raul A Villacreses  2 , McKenna L Eisenbeisz  2 , Rachel L Anderson  2 , Sarah E Van Dorin  2 , Letícia Rittner  3 , Roberto A Lotufo  3 , Sarah E Gerard  4 , Joseph M Reinhardt  4 , Alejandro P Comellas  2
Affiliations

Manual segmentation of opacities and consolidations on CT of long COVID patients from multiple annotators

Diedre S Carmo et al. Sci Data. .

Abstract

The field of supervised automated medical imaging segmentation suffers from relatively small datasets with ground truth labels. This is especially true for challenging segmentation problems that target structures with low contrast and ambiguous boundaries, such as ground glass opacities and consolidation in chest computed tomography images. In this work, we make available the first public dataset of ground glass opacity and consolidation in the lungs of Long COVID patients. The Long COVID Iowa-UNICAMP dataset (LongCIU) was built by three independent expert annotators, blindly segmenting the same 90 selected axial slices manually, without using any automated initialization. The public dataset includes the final consensus segmentation in addition to the individual segmentation from each annotator (360 slices total). This dataset is a valuable resource for training and validating new automated segmentation methods and for studying interrater uncertainty in the segmentation of lung opacities in computed tomography.

PubMed Disclaimer

Conflict of interest statement

Competing interests: Diedre Carmo reports an employment relationship with NeuralMind. Roberto A. Lotufo reports an employment and equity relationship with NeuralMind. Joseph M. Reinhardt reports an equity and consulting relationship with VIDA Diagnostics Inc. and a consulting relationship with Auris Health, Inc. The remaining authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Figures

Fig. 1
Fig. 1
Overview of the process of creating this dataset, from selecting the slices to be segmented, through blind multiple rater segmentation and consensus, technical validation and making the data and involved code public.
Fig. 2
Fig. 2
A sample slice and its accompanying segmentations, with GGO in red and consolidation in green. Notice how STAPLE arrives at a final consensus with a opacity sensitivity higher than annotator 3 but lower than annotator 1. The difficulty of differentiating vessels from consolidation is also illustrated in the anterior region of the right lung.
Fig. 3
Fig. 3
File tree for the dataset, including images, annotations per annotator and consensus, MEDPSeg output, optional proposed training splits, and STAPLE statistics.
Fig. 4
Fig. 4
On the left, number of pixels classified as GGO and consolidation for each annotator. On the right, agreement Dice overlap between annotators 2 and 3, 1 and 3, 1 and 2 for overall infection in blue, (Inf = GGO + consolidation), consolidation only in green (Con.) and GGO in red. The horizontal blue line corresponds to the inter-human annotator agreement for overall infection reported by Sotudeh-Paima et al..
Fig. 5
Fig. 5
Cohen’s Kappa interrater agreement between the three human raters and the MEDPSeg automated method for (a) The multilabel agreement, (b) GGO only, (c) consolidation only, and (d) overall infection, (GGO + Consolidation).
See this image and copyright information in PMC

References

    1. Lee, S. M. et al. Deep learning applications in chest radiography and computed tomography: current state of the art. Journal of thoracic imaging34, 75–85 (2019). - PubMed
    1. Gerard, S. E. et al. CT image segmentation for inflamed and fibrotic lungs using a multi-resolution convolutional neural network. Scientific Reports11, 1455, 10.1038/s41598-020-80936-4 (2021). - PMC - PubMed
    1. Gerard, S. E., Patton, T. J., Christensen, G. E., Bayouth, J. E. & Reinhardt, J. M. FissureNet: A deep learning approach for pulmonary fissure detection in CT images. IEEE transactions on medical imaging38, 156–166, 10.1109/TMI.2018.2858202 (2019). - PMC - PubMed
    1. Althof, Z. W. et al. Attention U-net for automated pulmonary fissure integrity analysis in lung computed tomography images. Scientific Reports13, 14135, 10.1038/s41598-023-41322-y (2023). - PMC - PubMed
    1. Wang, Y., Liu, L. & Wang, C. Trends in using deep learning algorithms in biomedical prediction systems. Frontiers in Neuroscience17, 1256351 (2023). - PMC - PubMed