Overview

Schema Version

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

Name

EOS AP images and external radiographs

Link

https://pubs.rsna.org/doi/10.1148/ryai.220158

Indexing

Keywords: scoliosis, Cobb angle, EOS imaging, anterior-posterior radiograph, spine, radiography, hardware-invariant measurement

Content: MK

RadLex: RID10345, RID28733, RID4756, RID10409

SNOMED: 298382003

Author(s)

Abhinav Suri

Sisi Tang

Daniel Kargilis

Elena Taratuta

Bruce J. Kneeland

Grace Choi

Alisha Agarwal

Nancy Anabaraonye

Winnie Xu

James B. Parente

Ashley Terry

Anita Kalluri

Katie Song

Chamith S. Rajapakse

Organization(s)

University of Pennsylvania Perelman School of Medicine

License

Text: Data generated or analyzed during the study are available from the corresponding author by request.

Funding

Supported by NIH NIAMS R01AR068382.

Ethical review

Institutional review board approval obtained; imaging studies downloaded and analyzed in a HIPAA-compliant manner.

Comments

Retrospective dataset of EOS anterior–posterior images and external radiographs used to train, validate, and test a deep learning system (SpineTK) for automated Cobb angle measurement in patients with suspected scoliosis.

Date

Published: 2023-06-21

References

[1] Suri A, Tang S, Kargilis D, Taratuta E, Kneeland BJ, Choi G, Agarwal A, Anabaraonye N, Xu W, Parente JB, Terry A, Kalluri A, Song K, Rajapakse CS. "Conquering the Cobb Angle: A Deep Learning Algorithm for Automated, Hardware-Invariant Measurement of Cobb Angle on Radiographs in Patients with Scoliosis". Radiology: Artificial Intelligence. 2023-06-21. doi:10.1148/ryai.220158. PMID: 37529207. PMCID: PMC10388214.

[2] Fraiwan M, Audat Z, Manasreh T. "A dataset of scoliosis, spondylolisthesis, and normal vertebrae X-ray images". . 2022-01-17. PMID: 35500000. PMCID: PMC9060368. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9060368/

Dataset

Motivation

To develop and evaluate an automated, hardware-invariant method to measure Cobb angles on radiographs for scoliosis diagnosis and monitoring.

Sampling

Convenience sample of all available EOS images from 2005–2020 at six centers; external public radiographs added to include radiographs and broader age representation.

Partitioning scheme

Training (n=509) and validation (n=180) EOS images with fivefold cross-validation to select best fold; holdout internal test set of EOS images (n=460) and external test set of radiographs (n=161).

Missing information

Imaging acquisition parameters (EOS and external radiographs) and complete racial demographics were not available.

Noise

Some images contained surgical hardware (pins, screws, rods, pacemakers) that could occlude parts of the spine.

External data

161 external radiographs from a public dataset (Jordan University of Science and Technology) were added for external testing after network training.

Confidentiality

HIPAA-compliant handling of imaging studies; retrospective study with IRB approval.