Overview

Schema Version

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

Name

Deep Learning Segmentation of Ascites on Abdominal CT Scans for Automatic Volume Quantification

Link

https://github.com/rsummers11/Ascites

Indexing

Keywords: ascites, segmentation, volume quantification, abdominal CT, nnU-Net, ovarian cancer, liver cirrhosis

Content: GI, CT

RadLex: RID45687, RID35976, RID49570, RID3822, RID10321, RID1541, RID10626

SNOMED: 389026000, 363443007

Author(s)

Benjamin Hou

Sungwon Lee

Jung-Min Lee

Christopher Koh

Jing Xiao

Perry J. Pickhardt

Ronald M. Summers

Organization(s)

Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, Department of Radiology and Imaging Sciences, Clinical Center, National Institutes of Health

Department of Radiology, The Catholic University of Korea, Seoul St. Mary’s Hospital

Women’s Malignancies Branch, National Cancer Institute, NIH

Liver Diseases Branch, NIDDK, NIH

Ping An Technology, Shenzhen, China

Department of Radiology, University of Wisconsin School of Medicine & Public Health

Contact

Ronald M. Summers (corresponding author)

Funding

Intramural Research Program (1Z01 CL040004) of the National Institutes of Health, Clinical Center; utilized NIH Biowulf HPC resources.

Ethical review

Nonpublic NIH and UofW datasets were HIPAA compliant; IRB approval obtained with waiver of informed consent.

Comments

Study trained on TCGA-OV CT scans and evaluated on three nonpublic test sets (NIH-LC, NIH-OV, UofW-LC). The paper states that curated training annotations and the trained segmentation model will be published at the linked repository.

Date

Published: 2024-06-20

References

[1] Hou B, Lee S, Lee JM, Koh C, Xiao J, Pickhardt PJ, Summers RM. "Deep Learning Segmentation of Ascites on Abdominal CT Scans for Automatic Volume Quantification". Radiology: Artificial Intelligence. 2024-06-20. doi:10.1148/ryai.230601. PMID: 38900043. PMCID: PMC11449171.

[2] Holback C, Jarosz R, Prior F, et al.. "The Cancer Genome Atlas Ovarian Cancer Collection (TCGA-OV) (Version 4) [Data set]". The Cancer Imaging Archive. 2016-01-01. doi:10.7937/K9/TCIA.2016.NDO1MDFQ.

Dataset

Motivation

Automated identification and volumetric quantification of ascites to aid clinical assessment in liver cirrhosis and ovarian cancer.

Sampling

For TCGA-OV, 285 abdominopelvic scans selected from 143 enrolled participants (140 used); mix of contrast-enhanced (n=212) and noncontrast (n=73).

Partitioning scheme

Fivefold cross-validation on TCGA-OV during training; external testing performed on NIH-LC, NIH-OV, UofW-LC.

Missing information

No explicit license stated for the released annotations/model in the paper; image-level file formats and exact series/image counts per set not provided.

Relationships between instances

Multiple scans per patient in TCGA-OV (285 scans from 140 patients).

Noise

Model performance influenced by CT section thickness (e.g., 5 mm in NIH-LC associated with oversegmentation/penalty).

External data

Images sourced from TCGA-OV (TCIA).

Confidentiality

Training images are from a public de-identified dataset (TCGA-OV). Test sets from NIH and UofW are nonpublic, HIPAA compliant.

Re-identification

Source public dataset (TCGA-OV) is de-identified; nonpublic datasets HIPAA compliant with IRB approval and consent waiver.

Sensitive data

Medical images; no PHI reported for public TCGA-OV.