Overview

Schema Version

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

Name

Denoising Multiphase Functional Cardiac CT Angiography Using Deep Learning and Synthetic Data (study dataset)

Link

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10982910/

Indexing

Keywords: Cardiac CT Angiography, Deep Learning, Image Denoising, Left ventricular function, Retrospective gating

Content: CA, CT, RS

RadLex: RID11224

Author(s)

Veit Sandfort

Martin J. Willemink

Marina Codari

Domenico Mastrodicasa

Dominik Fleischmann

Organization(s)

Department of Radiology, Stanford University School of Medicine

Funding

Authors declared no funding. Acknowledgments note support from the Stanford Radiology Department Radcombinator program and computer support from the Stanford 3D and Quantitative Imaging Laboratory.

Ethical review

HIPAA-compliant retrospective study approved by the institutional review board; requirement for patient consent was waived due to use of anonymized data.

Date

Published: 2024-02-28

References

[1] Sandfort V, Willemink MJ, Codari M, Mastrodicasa D, Fleischmann D. "Denoising Multiphase Functional Cardiac CT Angiography Using Deep Learning and Synthetic Data". Radiology: Artificial Intelligence. 2024-02-01. doi:10.1148/ryai.230153. PMID: 38416035. PMCID: PMC10982910.

Dataset

Motivation

Enable use of high-noise functional phases from dose-modulated retrospectively gated coronary CTA by denoising to facilitate valid LV functional measurements.

Sampling

566 consecutive adult patients undergoing coronary CTA with dose-modulated retrospective gating.

Partitioning scheme

Split per patient 80:10:10 into training, validation, and test. Resulted time series counts: 29,699 (train), 3,087 (validation), 3,325 (test).

Missing information

Due to anonymization, specific patient characteristics (e.g., demographics) were not available for correlation.

Relationships between instances

Each patient has multiphase time series (10 frames); synthetic low-noise time series created by deforming the lowest-noise frame to match higher-noise frames using nonrigid registration.

Noise

High time-variable noise in dose-modulated phases; synthetic low-noise labels generated via nonrigid coregistration for training.

External data

No external datasets reported; all data acquired at a single center with a single vendor scanner.

Confidentiality

Anonymized clinical imaging data; IRB approved; HIPAA compliant.

Re-identification

Data were anonymized; no patient-level identifiers available.

Sensitive data

Clinical cardiac CT images of adult patients; anonymized.