Overview

Schema Version

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

Name

RadImageNet: An Open Radiologic Deep Learning Research Dataset for Effective Transfer Learning

Link

https://github.com/BMEII-AI/RadImageNet

Indexing

Keywords: CT, MR Imaging, US, Head/Neck, Thorax, Brain/Brain Stem, Evidence-based Medicine, Computer Applications–General (Informatics), Thyroid Nodules, Breast Masses, Anterior Cruciate Ligament Injuries, Meniscal Tears, Pneumonia, COVID-19, SARS-CoV-2, Intracranial Hemorrhage, Transfer Learning, Convolutional Neural Network

Content: BR, CH, CT, GI, GU, HN, MK, MR, NR, OI, US

Author(s)

Xueyan Mei

Zelong Liu

Philip M. Robson

Brett Marinelli

Mingqian Huang

Amish Doshi

Adam Jacobi

Chendi Cao

Katherine E. Link

Thomas Yang

Ying Wang

Hayit Greenspan

Timothy Deyer

Zahi A. Fayad

Yang Yang

Organization(s)

Icahn School of Medicine at Mount Sinai

University of Oklahoma

Cornell Medicine

East River Medical Imaging

License

Text: CC BY 4.0

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

Contact

yy5cc@virginia.edu

Funding

Authors declared no funding for this work.

Ethical review

The institutional review boards waived the requirement for written informed consent for this retrospective, Health Insurance Portability and Accountability Act–compliant study, which evaluated de-identified data and involved no potential risk to patients.

Comments

The RadImageNet database is a large-scale dataset consisting of 1.35 million radiologic images covering CT, MRI, and US modalities and 11 anatomic regions, which were annotated by fellowship-trained and board-certified radiologists. It includes 165 pathologic labels.

Date

Created: 2005-01-01

Dataset

Motivation

To create and evaluate a large-scale, diverse medical imaging dataset, RadImageNet, to generate pretrained convolutional neural networks (CNNs) trained solely from medical imaging to be used as the basis of transfer learning for medical imaging applications.

Sampling

The RadImageNet dataset was collected between January 2005 and January 2020 from 131,872 patients at an outpatient radiology facility. Key images and associated labels were retrospectively extracted from original study interpretations by 20 board-certified, fellowship-trained radiologists. Pathologic finding labels were assigned to 'key images,' and regions of interest were created. 8528 normal studies with 263,039 images were also included, identified via SQL query and confirmed by a board-certified radiologist.

Confidentiality

To avert any potential breach of confidentiality, no link between the patients, data provider, and data receiver was made available. A third party issued a certification of de-identified data transfer from the data provider to the data receiver.