Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/dataset.json

Name

UT Southwestern cohort for Kidney Segmentation at Multiphase MRI

Link

https://doi.org/10.1148/ryai.230043

Indexing

Keywords: Kidney segmentation, CycleGAN, Generative adversarial network, Mask R-CNN, Multiphase contrast-enhanced MRI, T2-weighted MRI, Style transfer, Semisupervised learning

Content: GU, MR, IN

RadLex: RID49531

SNOMED: 309088003

Author(s)

Junyu Guo, PhD

Manu Goyal, PhD

Yin Xi, PhD

Lauren Hinojosa, MD

Gaelle Haddad, MD

Emin Albayrak, MD

Ivan Pedrosa, MD, PhD

Organization(s)

University of Texas Southwestern Medical Center

License

Text: Article published under CC BY 4.0; study data available from corresponding author by request.

URL: https://creativecommons.org/licenses/by/4.0/

Contact

Corresponding author: Ivan Pedrosa, University of Texas Southwestern Medical Center

Funding

Partially funded by the National Institutes of Health (R01CA154475). Computational resources provided by the BioHPC supercomputing facility, Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center.

Ethical review

Institutional review board–approved; Health Insurance Portability and Accountability Act–compliant; informed consent waived.

Comments

Retrospective, HIPAA-compliant, IRB-approved study of patients with renal masses imaged with T2-weighted and multiphase contrast-enhanced MRI; style transfer (CycleGAN) used to synthesize multiphase images for training kidney segmentation (Mask R-CNN). Data sharing: available from corresponding author by request.

Date

Published: 2023-09-13

References

[1] Guo J, Goyal M, Xi Y, Hinojosa L, Haddad G, Albayrak E, Pedrosa I. "Style Transfer–assisted Deep Learning Method for Kidney Segmentation at Multiphase MRI". Radiology: Artificial Intelligence. 2023. doi:10.1148/ryai.230043. PMID: 38074795. PMCID: PMC10698598.

Dataset

Motivation

Enable automated kidney segmentation across multiple dynamic MRI phases without manual annotation of each phase, mitigating misregistration challenges and annotation burden.

Sampling

Patients with known renal masses imaged at UT Southwestern (and some outside institutions for cohort 1) during 2011–2015 (cohort 1) and 2016–2019 (cohort 2).

Partitioning scheme

Cohort 1: 102 coronal T2-weighted acquisitions and 27 corticomedullary-phase acquisitions. CycleGAN Cmodel1 trained using 82 T2-weighted (source) and 27 CM (target). Cohort 2: 23 multiphase MCE MRI exams (4 phases each; 92 acquisitions) used for style transfer training for other phases; independent testing on 20 MCE MRI exams (4 phases).

Missing information

Exact file formats, per-image resolutions, and complete train/validation splits are provided only in supplemental tables (S1–S5) and not fully enumerated in the article text.

Relationships between instances

Multiple MRI acquisitions per patient across phases (precontrast, corticomedullary, early nephrographic, nephrographic); T2-weighted and synthetic MCE images are anatomically coregistered within patient.

Noise

Examinations with substantial field inhomogeneity or motion artifacts were excluded from training after manual inspection.

External data

Some MRI examinations performed at outside institutions (cohort 1) met minimum technical requirements.

Confidentiality

Retrospective HIPAA-compliant imaging study; IRB-approved; consent waived.

Re-identification

Not discussed; typical institutional workflows imply de-identified data for research use.

Sensitive data

Clinical imaging of patients with renal masses.