Overview

Schema Version

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

Name

FASTR-SCANN (Fully Automated Segmental Relaxometry using Synthetic Contrast Augmentation with Neural Networks) / U-NetSCANN

Link

https://pubmed.ncbi.nlm.nih.gov/36523641/

Indexing

Keywords: cardiac MRI, T1 mapping, extracellular volume, MOLLI, STONE sequence, myocardial segmentation, U-Net, deep learning, synthetic contrast augmentation

Content: CA, MR

SNOMED: 50920009, 85898001, 399020009, 233873004, 22298006, 414545008, 16573007

Author(s)

Nitish Bhatt

Venkat Ramanan

Ady Orbach

Labonny Biswas

Matthew Ng

Fumin Guo

Xiuling Qi

Lancia Guo

Laura Jimenez-Juan

Idan Roifman

Graham A. Wright

Nilesh R. Ghugre

Organization(s)

University of Toronto

Sunnybrook Research Institute

Sunnybrook Health Sciences Centre

St Michael’s Hospital

Version

1.0

License

Contact

ac.otnorotu.liam@ttahb.hsitin

Funding

Ontario Research Fund (Canada) (ORF-RE7–21); Natural Sciences and Engineering Research Council (NSERC) Discovery Program (RGPIN-2019–06367); N.R.G. supported by the Heart and Stroke Foundation of Canada National New Investigator award.

Ethical review

The Research Ethics Board of Sunnybrook Health Sciences Center approved the study; all patients provided written informed consent.

Date

Updated: 2022-10-19

Published: 2022-11-02

Created: 2021-11-24

References

[1] Bhatt N, Ramanan V, Orbach A, Biswas L, Ng M, Guo F, Qi X, Guo L, Jimenez-Juan L, Roifman I, Wright GA, Ghugre NR. "A Deep Learning Segmentation Pipeline for Cardiac T1 Mapping Using MRI Relaxation–based Synthetic Contrast Augmentation". Radiology: Artificial Intelligence. 2022 Nov;4(6):e210294.. 2022-11-02. doi:10.1148/ryai.210294. PMID: 36523641. PMCID: PMC9745444.

Model

Architecture

Convolutional Neural Network (U-Net) with synthetic contrast augmentation at multiple inversion times and label voting aggregation.

Clinical benefit

Automated segmentation of cardiac T1 maps and computation of segmental/global native T1, postcontrast T1, and extracellular volume to reduce manual effort and variability.

Degree of automation

Fully automated segmentation and quantification pipeline.

Indications for use

Automated analysis of cardiac MRI T1 mapping studies (native and postcontrast) for patients across various cardiac abnormalities to obtain LV/RV segmentation, RV insertion point, and segmental/global T1 and ECV values in clinical research or clinical imaging settings.

Input

Cardiac MRI T1-weighted image series (multiple inversion times) used to reconstruct T1 maps; native and postcontrast series (e.g., MOLLI) and native STONE for external testing; hematocrit for ECV when available.

Limitations

Evaluated only at 1.5T and on MOLLI and STONE sequences; findings may not generalize to other field strengths or sequences. Possible registration errors during T1 reconstruction and absence of additional registration between native and postcontrast maps for ECV; however, agreement with experts was strong. Failure cases included thin myocardial wall, artifacts/partial volume, and dilated myocardium. RV insertion point localization showed larger inter-method distances than interobserver variability.

Output

CDEs: RDE218, RDE2204, RDE215, RDE226, RDE2205, RDE223, RDE217, RDE216

Description: Segmentation masks for LV myocardium, LV blood pool, and RV blood pool; automatic RV insertion point; AHA 17-segment assignments; per-segment and global T1 (native and postcontrast) and extracellular volume values.

Reproducibility

Performance validated with fivefold cross-validation, internal test set with two expert readers, and an external multi-center/multi-vendor/sequence dataset (generalization demonstrated).

Use

Intended: Image segmentation

Out-of-scope: Other

Excluded: Decision support, Artifact reduction

User

Intended: Physician, Subspecialist diagnostic radiologist, Researcher