Overview

Schema Version

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

Name

CT-based Osteoporosis Assessment study dataset

Link

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9530763/

Indexing

Keywords: opportunistic screening, bone mineral density, L1 vertebra, trabecular attenuation, Hounsfield units, deep learning, abdominal CT, osteoporosis assessment

Content: CT, GI, MK, BQ

RadLex: RID29193, RID7480, RID5389, RID28662, RID10363, RID10321, RID39161, RID12719

SNOMED: 64859006

Author(s)

Perry J. Pickhardt

Thang Nguyen

Alberto A. Perez

Peter M. Graffy

Samuel Jang

Ronald M. Summers

John W. Garrett

Organization(s)

Department of Radiology, University of Wisconsin School of Medicine & Public Health

Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, Department of Radiology and Imaging Sciences, National Institutes of Health Clinical Center

Funding

Authors declared no funding for this work (study-specific).

Ethical review

IRB approved; HIPAA-compliant; informed consent waived.

Comments

Single-center, retrospective study using abdominal CT scans with manual L1 trabecular attenuation as reference standard; development, testing, and validation of a DL tool against a prior feature-based algorithm.

Date

Published: 2022-08-31

References

[1] Pickhardt PJ, Nguyen T, Perez AA, Graffy PM, Jang S, Summers RM, Garrett JW. "Improved CT-based Osteoporosis Assessment with a Fully Automated Deep Learning Tool". Radiology: Artificial Intelligence. 2022-09-01. doi:10.1148/ryai.220042. PMID: 36204542. PMCID: PMC9530763.

Dataset

Motivation

Develop, test, and validate a DL tool for automated CT-based BMD assessment and compare with manual reference and a prior feature-based algorithm.

Sampling

All abdominal CT studies from PACS over long period without focused selection to mirror opportunistic approach; noncontrast preferentially selected for manual reference, but contrast-enhanced exams included.

Partitioning scheme

DL model training/validation on image pairs from multiple vertebral levels; independent test on separate single-scan-per-patient cohort.

Missing information

No public release details or de-identification protocol reported.

Relationships between instances

Multiple axial slices per study; training used multiple vertebral levels (approximately 1.7 ROIs per case).

Noise

Heterogeneous clinical CT protocols (kVp mostly 120; varying mA, IV/oral contrast phases, slice thickness, kernels).

Confidentiality

Retrospective single-center PACS data; HIPAA-compliant.

Sensitive data

Clinical CT images from human subjects.