Large-scale Instance-diverse Synthetic COVID-19 CT Dataset
<p><strong>Dataset Description</strong></p> <p>This dataset consists of 367000 synthetic COVID-19 CT images generated from a new GAN algorithm known as the stacked residual dropout GAN (sRD-GAN) [1]. The 367 input images are acquired from the iCTCF dataset [2] and are stored in 512 × 512 × 3 in .jpg format. The input images [2] and their corresponding synthetic images are included in this dataset. </p> <p> </p> <p><strong>Stacked Residual Dropout GAN (sRD-GAN)</strong></p> <p>The sRD-GAN utilizes a regularization-based strategy in an Image-to-Image (I2I) translation setting to facilitate instance-level diversity. In this study, we show that the stacked dropout regularization in the generator model can induce significant latent-space stochasticity which generates perceptually significant structural dissimilarity in the output space. </p> <p>Paper: <a href="https://www.mdpi.com/2306-5354/9/11/698">Diverse COVID-19 CT Image-to-Image Translation with Stacked Residual Dropout</a></p> <p>DOI: <a href="https://doi.org/10.3390/bioengineering9110698">10.3390/bioengineering9110698</a></p> <p> </p> <p><strong>Advantages</strong></p> <p>1) High-resolution, diverse patterns of synthetic ground-glass-opacities presented in chest CT images with different anatomy. </p> <p>2) Generalize across GAN-based models since the stacked residual dropout mechanism is not task- or dataset-specific. </p> <p>3) Does not require any auxiliary condition to generate diverse outputs. </p> <p>4) Does not require any non-trivial modification on the model's architectures. </p> <p> </p> <p><strong>Disadvantages</strong></p> <p>1) Diversity is not presented in large perceptual differences, and it focuses only on fine-grained details. </p> <p>2) Since sRD-GAN is trained in an unsupervised image-to-image setting, and the synthesis process does not require any auxiliary condition, thus, the magnitude of the style attributes (GGO features) cannot be manipulated. </p> <p> </p> <p><strong>Acknowledgments: </strong></p> <p>If you use this dataset in your research, please credit the author: </p> <p>[1] Lee, K.W.; Chin, R.K.Y. Diverse COVID-19 CT Image-to-Image Translation with Stacked Residual Dropout. <em>Bioengineering</em> <strong>2022</strong>, <em>9</em>, 698. https://doi.org/10.3390/bioengineering9110698</p> <p> </p> <p><strong>References: </strong></p> <p>[2] Ning, W.; Lei, S.; Yang, J.; Cao, Y.; Jiang, P.; Yang, Q.; Zhang, J.; Wang, X.; Chen, F.; Geng, Z.; et al. Open resource of clinical data from patients with pneumonia for the prediction of COVID-19 outcomes via Deep Learning. Nat. Biomed. Eng. <strong>2020</strong>, 4, 1197–1207.</p>
ShareScore
32/100
Overall dataset sharing score
Score breakdown
These five areas show where the dataset supports — or may limit — practical reuse.
- Stewardship
- 4
- Harmonization
- 4
- Access
- 16
- Reuse readiness
- 8
- Engagement
- 0
Dataset details
- Repository
- zenodo
- Published
- November 20, 2022
- License
- cc-by-4.0
- DOI
- 10.5281/zenodo.7340325
Scored with release 0.9.0
ScienceDex for faculty
Your academic CV and NIH biosketch
Make your CV in your school's format and your NIH biosketch from it, then download both as Word files.