Overview

Schema Version

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

Name

Custom DL-based prostate and OAR MRI segmentation software for LDRPBT (MIM Symphony integration)

Link

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8980936

Indexing

Keywords: MRI, Neural Networks, Radiation Therapy, Prostate, Segmentation, Dosimetry, Organs at risk, Brachytherapy

Content: MR, RO, GU

RadLex: RID163, RID49517, RID38870, RID357

Author(s)

Jeremiah W. Sanders

Rajat J. Kudchadker

Chad Tang

Henry Mok

Aradhana M. Venkatesan

Howard D. Thames

Steven J. Frank

Organization(s)

The University of Texas MD Anderson Cancer Center

Version

1.0

License

Text: © RSNA, 2022

URL: https://pubs.rsna.org/journal/ryai

Contact

Jeremiah W. Sanders, gro.nosrednadm@1srednasj

Ethical review

HIPAA-compliant and institutional review board–approved protocol at the University of Texas MD Anderson Cancer Center.

Date

Updated: 2022-03-01

Published: 2022-01-26

Created: 2021-06-14

References

[1] Sanders JW, Kudchadker RJ, Tang C, Mok H, Venkatesan AM, Thames HD, Frank SJ. "Prospective Evaluation of Prostate and Organs at Risk Segmentation Software for MRI-based Prostate Radiation Therapy". Radiology: Artificial Intelligence. 2022;4(2):e210151.. 2022-01-26. doi:10.1148/ryai.210151. PMID: 35391775. PMCID: PMC8980936.

[2] Sanders JW, Lewis GD, Thames HD, et al.. "Machine segmentation of pelvic anatomy in MRI-assisted radiosurgery (MARS) for prostate cancer brachytherapy". Int J Radiat Oncol Biol Phys. 2020;108(5):1292-1303.. 2020-01-01. PMID: 32634543. Available from: https://pubmed.ncbi.nlm.nih.gov/32634543/

Model

Architecture

Fully convolutional networks for semantic segmentation (deep learning).

Availability

Implemented via software interface into a commercial treatment planning system (MIM Symphony); custom in-house algorithm. No public download URL provided.

Clinical benefit

Generates clinical-quality contours of prostate and surrounding organs at risk on MRI to support LDRPBT planning and postimplant dosimetry; potential workflow efficiency improvement and reduced intraobserver variability.

Clinical workflow phase

Clinical decision support and workflow optimization for radiation therapy planning and postimplant quality assessment.

Decision threshold

User-selectable operating point per organ (displayed confidence thresholds spanning 10%–90%); a fixed operating point mode is also available for fully automatic use.

Degree of automation

Automated segmentation with optional user selection of operating point and physician review/refinement; also evaluated in fully automatic mode with fixed thresholds.

Indications for use

Segmentation of prostate and organs at risk (EUS, seminal vesicles, rectum, bladder) on postimplant prostate MRI for low-dose-rate prostate brachytherapy in patients with prostate cancer within radiation oncology treatment planning/QA environments.

Input

Postimplant prostate MRI (turbo spin-echo or fully balanced SSFP sequences).

Instructions

Run autosegmentation; select per-organ operating point if desired; physician (radiation oncologist) reviews and refines contours as needed; proceed to dose-volume histogram analysis within the treatment planning system.

Limitations

Prospective evaluation on a small single-institution cohort (n=30). Preimplant MRIs not evaluated. Algorithm developed and evaluated at the same institution. Highest variability and need for refinements observed in EUS contours, reflecting variability in human annotations. Biodegradable hydrogel rectal spacers were underrepresented in development data (<1%) though algorithm handled their presence; potential threshold sensitivity at low decision values.

Output

CDEs: RDE101, RDE93, RDE1551, RDE103, RDE1549

Description: Segmentation masks/contours for prostate and organs at risk.

Recommendation

Use as an automatic contouring aid for MRI-based LDRPBT with physician oversight; either user-selected or fixed operating points acceptable given similar dosimetry outcomes.

Reproducibility

Deterministic predictions for a given model and operating point; eliminates intraobserver variability component in contouring.

Sustainability

Execution time approximately 1 minute or less per MRI for autosegmentation (reported qualitatively).

Use

Intended: Image segmentation

User

Intended: Physician, Other, Subspecialist diagnostic radiologist, Researcher