Journal article
Unsupervised segmentation and quantification of COVID-19 lesions on computed Tomography scans using CycleGAN
Methods (San Diego, Calif.), Vol.205, pp.200-209
09/2022
DOI: 10.1016/j.ymeth.2022.07.007
PMCID: PMC9288584
PMID: 35817338
Abstract
[Display omitted]
•We trained a modified CycleGAN to segment pulmonary lesions on COVID-19 CT scans.•Our unsupervised model performed similarly to weakly supervised models. Our model preserved normal physiology in generated images.
Lesion segmentation is a critical step in medical image analysis, and methods to identify pathology without time-intensive manual labeling of data are of utmost importance during a pandemic and in resource-constrained healthcare settings. Here, we describe a method for fully automated segmentation and quantification of pathological COVID-19 lung tissue on chest Computed Tomography (CT) scans without the need for manually segmented training data.
We trained a cycle-consistent generative adversarial network (CycleGAN) to convert images of COVID-19 scans into their generated healthy equivalents. Subtraction of the generated healthy images from their corresponding original CT scans yielded maps of pathological tissue, without background lung parenchyma, fissures, airways, or vessels. We then used these maps to construct three-dimensional lesion segmentations. Using a validation dataset, Dice scores were computed for our lesion segmentations and other published segmentation networks using ground truth segmentations reviewed by radiologists.
The COVID-to-Healthy generator eliminated high Hounsfield unit (HU) voxels within pulmonary lesions and replaced them with lower HU voxels. The generator did not distort normal anatomy such as vessels, airways, or fissures. The generated healthy images had higher gas content (2.45 ± 0.93 vs 3.01 ± 0.84 L, P < 0.001) and lower tissue density (1.27 ± 0.40 vs 0.73 ± 0.29 Kg, P < 0.001) than their corresponding original COVID-19 images, and they were not significantly different from those of the healthy images (P < 0.001). Using the validation dataset, lesion segmentations scored an average Dice score of 55.9, comparable to other weakly supervised networks that do require manual segmentations.
Our CycleGAN model successfully segmented pulmonary lesions in mild and severe COVID-19 cases. Our model’s performance was comparable to other published models; however, our model is unique in its ability to segment lesions without the need for manual segmentations.
Details
- Title: Subtitle
- Unsupervised segmentation and quantification of COVID-19 lesions on computed Tomography scans using CycleGAN
- Creators
- Marc Connell - University of PennsylvaniaYi Xin - University of PennsylvaniaSarah E. Gerard - Harvard UniversityJacob Herrmann - Boston UniversityParth K. Shah - University of PennsylvaniaKevin T. Martin - University of PennsylvaniaEmanuele Rezoagli - University of Milano-BicoccaDavide Ippolito - Azienda Ospedaliera San GerardoJennia Rajaei - Stanford UniversityRyan Baron - University of PennsylvaniaPaolo Delvecchio - University of PennsylvaniaShiraz Humayun - University of PennsylvaniaRahim R. Rizi - University of PennsylvaniaGiacomo Bellani - University of Milano-BicoccaMaurizio Cereda - University of Pennsylvania
- Resource Type
- Journal article
- Publication Details
- Methods (San Diego, Calif.), Vol.205, pp.200-209
- Publisher
- Elsevier Inc
- DOI
- 10.1016/j.ymeth.2022.07.007
- PMID
- 35817338
- PMCID
- PMC9288584
- ISSN
- 1046-2023
- eISSN
- 1095-9130
- Grant note
- DOI: 10.13039/100000002, name: National Institutes of Health, award: 1R01HL137389-01A1
- Language
- English
- Date published
- 09/2022
- Academic Unit
- Roy J. Carver Department of Biomedical Engineering
- Record Identifier
- 9984306747702771
Metrics
25 Record Views