Overview

Schema Version

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

Name

Dataset 1 (DS1): Frontal chest radiographs with radiograph-level and lesion-level annotations from Shenzhen People’s Hospital

Link

https://pmc.ncbi.nlm.nih.gov/articles/PMC9530769

Indexing

Keywords: Chest radiograph, Frontal CXR, Bounding boxes, Lesion localization, Cardiomegaly, Pleural effusion, Mass, Nodule, Pneumonia, Pneumothorax, Tuberculosis, Fracture, Aortic calcification

Content: CH

RadLex: RID28788, RID50149, RID34539, RID5335, RID29116, RID39056, RID35057, RID5352

SNOMED: 233604007, 125605004, 36118008, 8186001, 56717001, 309529002, 72704001, 60046008, 786838002

Author(s)

Luyang Luo

Hao Chen

Yongjie Xiao

Yanning Zhou

Xi Wang

Varut Vardhanabhuti

Mingxiang Wu

Chu Han

Zaiyi Liu

Xin Hao Benjamin Fang

Efstratios Tsougenis

Huangjing Lin

Pheng-Ann Heng

Organization(s)

Shenzhen People’s Hospital, Luohu, Shenzhen, China

Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong, China

AI Research Laboratory, Imsight Technology, Shenzhen, China

Department of Radiology, Guangdong Provincial People’s Hospital, Guangzhou, China

Contact

Corresponding author: Hao Chen (email as provided in article: kh.tsu.esc@chj)

Funding

Key-Area Research and Development Program of Guangdong Province, China (2020B010165004, 2018B010109006); Hong Kong Innovation and Technology Fund (ITS/311/18FP); HKUST Bridge Gap Fund (BGF.005.2021); National Natural Science Foundation of China (U1813204); Shenzhen-HK Collaborative Development Zone.

Ethical review

Institutional ethical committee approval no. YB-2021–554; individual consent waived; all institutional data de-identified.

Comments

Retrospective single-center internal dataset used to develop and evaluate CheXNet (classification) and CheXDet (detection) models; includes both radiograph-level labels and fine-grained lesion bounding boxes for nine thoracic findings.

Date

Published: 2022

References

[1] Luo L, Chen H, Xiao Y, Zhou Y, Wang X, Vardhanabhuti V, et al.. "Rethinking Annotation Granularity for Overcoming Shortcuts in Deep Learning–based Radiograph Diagnosis: A Multicenter Study". Radiology: Artificial Intelligence. 2022-07-20. doi:10.1148/ryai.210299. PMID: 36204545. PMCID: PMC9530769.

Dataset

Motivation

Evaluate whether fine-grained lesion-level annotations alleviate shortcut learning and improve generalizability of DL models for chest radiograph diagnosis.

Sampling

Retrospective collection from PACS between Jan 1, 2005 and Sep 31, 2019.

Partitioning scheme

Patient-disjoint splits: training n=28,673; tuning n=2,906; internal testing n=2,922; reader-study subset n=496 sampled from testing. Additional ablations used 20%, 40%, 60%, 80%, and 100% of training set.

Missing information

Exact image pixel dimensions and acquisition parameters not reported; demographic breakdown per partition not reported.

Relationships between instances

Multiple radiographs per patient possible; labels include radiograph-level presence/absence for 9 findings and lesion-level bounding boxes for corresponding abnormalities.

Noise

Not explicitly characterized; external datasets have differing label sources (radiologist-adjudicated vs NLP-derived).

External data

External testing used NIH Google (ChestX-ray14 subset), PadChest, and NIH ChestX-ray14; additional training for a comparison model used CheXpert.

Confidentiality

De-identified retrospective clinical images; internal dataset cannot be released for privacy and safety reasons.

Re-identification

All institutional data de-identified prior to research use.

Sensitive data

Medical imaging data with associated reports; PHI removed prior to analysis.