Overview

Schema Version

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

Name

Curation of the CANDID-PTX Dataset with Free-Text Reports

Link

https://doi.org/10.1148/ryai.2021210136

Indexing

Keywords: Conventional Radiography, Thorax, Trauma, Ribs, Catheters, Segmentation, Diagnosis, Classification, Supervised Learning, Machine Learning

Content: CH

RadLex: RID5352, RID5573

Author(s)

Sijing Feng

Damian Azzollini

Ji Soo Kim

Cheng-Kai Jin

Simon P. Gordon

Jason Yeoh

Eve Kim

Mina Han

Andrew Lee

Aakash Patel

Joy Wu

Martin Urschler

Amy Fong

Cameron Simmers

Gregory P. Tarr

Stuart Barnard

Ben Wilson

Organization(s)

Department of Radiology, Dunedin Hospital

Eastern Health

Auckland District Health Board

Waitemata District Health Board

Waikato District Health Board

The University of Auckland Faculty of Medical and Health Sciences

University of Otago Medical School

IBM Almaden Research Center

School of Computer Science, University of Auckland

Department of Radiology, Auckland City Hospital

Department of Radiology, Middlemore Hospital

Contact

sijingfeng@gmail.com

Funding

Royal Australia and New Zealand College of Radiologists (RANZCR) research grant in 2020.

Ethical review

Approved by the University of Otago Human Ethics Committee, which waived individual patient informed consent due to low-risk nature and use of anonymized data.

Comments

This large chest radiograph dataset has segmented annotations for pneumothoraces, acute rib fractures, and intercostal chest tubes. The dataset, which can be used to train and test machine learning algorithms in the identification of these features on chest radiographs, includes corresponding anonymized, free-text radiology reports. The temporal relationship between images of the same patient has been preserved.

Dataset

Motivation

To curate the first large pneumothorax segmentation dataset in a New Zealand population to address the bottleneck in developing large, well-annotated datasets for training and testing AI algorithms on adult chest radiographs.

Sampling

A total of 295,613 chest radiographs and their radiology reports were retrospectively acquired from Dunedin Hospital PACS between January 2010 and April 2020. Inclusion criteria: frontal chest radiographs (including bedside images) from patients over 16 years of age.

Missing information

30 images had no information on patient sex when extraction from the original DICOM metadata was attempted.

Relationships between instances

The temporal relationship between images of the same patient has been preserved.

Noise

80 images were excluded due to lacking clear 'pneumothorax' or 'no pneumothorax' labels, likely due to human error. The dataset's single-institution origin reduces heterogeneity, potentially limiting AI algorithm generalizability.

External data

MD.ai, a Health Insurance Portability and Accountability Act–compliant, commercial, cloud-based image annotation platform, was used for image upload and annotation.

Confidentiality

Image metadata and reports were de-identified using algorithm-based tools to remove protected health information (PHI) and satisfy New Zealand Ministry of Health standards. Manual review ensured quality of anonymization.

Re-identification

File names containing acquisition dates were anonymized consistently and irreversibly by translating them into future dates, while preserving temporal relationships between images.

Sensitive data

Protected health information (PHI) was removed from images and reports during de-identification.