Overview

Schema Version

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

Name

Radiographic marker annotations on pelvic radiographs for anonymization (from Mayo Clinic hip arthroplasty registry)

Link

https://dx.doi.org/10.1148/ryai.230085

Indexing

Keywords: radiographic markers, de-identification, anonymization, laterality markers, pelvis, hip arthroplasty, CheXpert, object detection, YOLOv5

Content: MK, CH

RadLex: RID33276, RID49837, RID11256, RID28712

Author(s)

Bardia Khosravi

John P. Mickley

Pouria Rouzrokh

Michael J. Taunton

A. Noelle Larson

Bradley J. Erickson

Cody C. Wyles

Organization(s)

Mayo Clinic, Orthopedic Surgery Artificial Intelligence Laboratory

Mayo Clinic, Radiology Informatics Laboratory

Mayo Clinic, Department of Orthopedic Surgery

Mayo Clinic, Department of Clinical Anatomy

Funding

Supported by the Mayo Foundation Presidential Fund.

Ethical review

Institutional review board approved with waiver of informed consent.

Comments

Dataset consists of 2000 annotated anteroposterior pelvic radiographs used to train and evaluate a radiographic marker localization/removal algorithm; external testing performed on CheXpert chest radiographs.

Date

Published: 2023-09-13

References

[1] Khosravi B, Mickley JP, Rouzrokh P, Taunton MJ, Larson AN, Erickson BJ, Wyles CC. "Anonymizing Radiographs Using an Object Detection Deep Learning Algorithm". Radiology: Artificial Intelligence. 2023-09-01. doi:10.1148/ryai.230085. PMID: 38074777. PMCID: PMC10698585.

[2] Irvin J, Rajpurkar P, Ko M, et al.. "CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison". arXiv:1901.07031. 2019-01-21. Available from: https://arxiv.org/abs/1901.07031

Dataset

Motivation

Enable de-identified sharing and robust AI training by detecting and removing burned-in radiographic markers while optionally retaining laterality markers.

Sampling

2000 AP pelvic radiographs randomly selected from an institutional hip arthroplasty registry (Jan 2000–Jan 2021).

Partitioning scheme

Patient-level split of internal dataset into 60%/20%/20% for training/validation/test.

Missing information

Demographics, site count, and exact patient counts per split not reported; image file format not specified.

Noise

External false positives frequently due to electrocardiography leads not seen during training.

External data

External testing performed on CheXpert chest radiograph validation set (n=234; 3 excluded). Model fine-tuned on 20 images from CheXpert training set.

Confidentiality

Clinical images from an institutional registry; IRB-approved with waiver of informed consent.

Re-identification

Markers may contain PHI (patient identifiers, dates, staff initials, institution); algorithm removes all non-laterality markers to prevent PHI leakage.

Sensitive data

Burned-in identifiers and staff/institutional markers present on internal radiographs.