Overview

Schema Version

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

Name

High-resolution cadaveric proximal femur micro-CT dataset for super-resolution of trabecular bone

Link

https://doi.org/10.1148/ryai.220251

Indexing

Keywords: CT, Image postprocessing, Super-resolution, Denoising diffusion probabilistic model, Trabecular bone microstructure, Proximal femur, Hip, Quantification, Radiation effects, Semisupervised learning

Content: MK, BQ

RadLex: RID35976, RID49215, RID12751, RID10553, RID12734, RID12314, RID6106, RID2637, RID10323, RID36016

SNOMED: 64859006

Author(s)

Trevor J. Chan

Chamith S. Rajapakse

Organization(s)

University of Pennsylvania

Department of Bioengineering, University of Pennsylvania

Department of Radiology, University of Pennsylvania

Department of Orthopedic Surgery, University of Pennsylvania

Contact

Corresponding author: Chamith S. Rajapakse, PhD, Departments of Radiology and Orthopedic Surgery, University of Pennsylvania, 3400 Spruce St, Philadelphia, PA 19104-6243.

Funding

National Science Foundation (grant no. 2026906) and National Institutes of Health (grant nos. R01 AR068282 and R01 AR076392).

Ethical review

Retrospective, HIPAA-compliant study approved by the institutional review board.

Comments

Dataset consists of high-resolution axial micro-CT scans of 26 human cadaveric femurs used for training/validation/testing of a diffusion-based super-resolution model. Images were downsampled to simulate clinical CT resolution.

Date

Published: 2023-09-20

References

[1] Chan TJ, Rajapakse CS. "A Super-Resolution Diffusion Model for Recovering Bone Microstructure from CT Images". Radiology: Artificial Intelligence. 2023. doi:10.1148/ryai.220251. PMID: 38074790. PMCID: PMC10698592.

Dataset

Motivation

Enable recovery of trabecular bone microstructure from low-resolution CT to allow lower radiation dose while preserving physiologic and mechanical fidelity.

Sampling

26 human cadaveric femurs (14 male, 12 female), ages 36–99 years (mean 73).

Partitioning scheme

26 scans split into training/validation/testing sets of 20/2/4 scans (77%/8%/15%).

Missing information

File formats, scanner manufacturer/model, acquisition parameters beyond resolution, and exact per-partition image counts are not reported.

Relationships between instances

Approximately 91,000 cross-sectional bone images derived from 26 scans; multiple slices per femur.

Noise

Low-resolution images generated via bicubic downsampling (factor of three) from high-resolution ground truth.

External data

No external datasets reported; all data obtained from cadaveric femur micro-CT scans within the study.

Confidentiality

Cadaveric samples; HIPAA-compliant; no patient-identifying information reported.

Re-identification

Low risk; cadaveric data and no PHI reported.

Sensitive data

No PHI; cadaveric specimens.