Duke Liver MRI Dataset with Segmentation Masks and Series Labels
The Duke Liver Dataset contains 2146 abdominal MRI series from 105 patients, including a majority with cirrhotic features, and 310 image series with corresponding manually segmented liver masks. The dataset supports development and evaluation of automated series classification and liver segmentation models.
2025-11-21https://doi.org/10.1148/atlas.1763595090013
172
Overview
Schema Version
https://atlas.rsna.org/schemas/2025-11/dataset.json
Name
Duke Liver MRI Dataset with Segmentation Masks and Series Labels
Link
https://zenodo.org/record/7774566
Indexing
Keywords: MR Imaging, Abdominal Imaging, Liver, Image Segmentation, Feature Detection, Series Classification, Liver Disease, Cirrhosis
Content: GI, MR
RadLex: RID58
Author(s)
Jacob A. Macdonald
Zhe Zhu
Brandon Konkel
Maciej A. Mazurowski
Walter F. Wiggins
Mustafa R. Bashir
Organization(s)
Duke University
Version
2.0
Contact
jacob.macdonald@duke.edu
Ethical review
Approved by data licensing committee and local institutional review board. Requirement for written informed consent was waived due to retrospective nature and full de-identification.
Comments
The Duke Liver Dataset contains 2146 abdominal MRI series from 105 patients, including a majority with cirrhotic features, and 310 image series with corresponding manually segmented liver masks. The dataset supports development and evaluation of automated series classification and liver segmentation models.
Date
Created: 2020-10-28
Dataset
Motivation
To provide a resource for developing liver segmentation tools and automated series classification models, especially given heterogeneous labeling of image sequence types and the time-consuming nature of manual segmentation.
Sampling
Derived from a subset of patients who underwent clinical contrast-enhanced MRI examinations of the abdomen for liver cirrhosis indication at one of three centers. 87 of 105 patients had imaging findings suggestive of cirrhosis.
Missing information
Not all possible contrast types are represented, and MRI is the only modality represented.
Relationships between instances
Includes 2146 image series from 105 patients with corresponding human-derived labels for MRI sequence type and liver segmentation.
Noise
Mild to moderate artifacts, such as motion and susceptibility, are not uncommon. Cirrhosis causes characteristic changes in liver shape that could confound models trained on healthy livers.
External data
Can be supplemented with Combined (CT-MR) Healthy Abdominal Organ Segmentation (CHAOS) or Liver Tumor Segmentation (LiTS) datasets.
Confidentiality
Full de-identification of the data was performed.
Re-identification
Full de-identification of the data was performed.
Sensitive data
Data was fully de-identified.