Stress-testing pelvic autosegmentation algorithms using anatomical edge cases

Aasheesh Kanwar; Brandon Merz; Cheryl Claunch; Shushan Rana; Arthur Hung; Reid F. Thompson

doi:10.1016/j.phro.2023.100413

Stress-testing pelvic autosegmentation algorithms using anatomical edge cases

Aasheesh Kanwar, Brandon Merz, Cheryl Claunch, Shushan Rana, Arthur Hung, Reid F. Thompson

Research output: Contribution to journal › Article › peer-review

1 Scopus citations

Abstract

Commercial autosegmentation has entered clinical use, however real-world performance may suffer in certain cases. We aimed to assess the influence of anatomic variants on performance. We identified 112 prostate cancer patients with anatomic variations (edge cases). Pelvic anatomy was autosegmented using three commercial tools. To evaluate performance, Dice similarity coefficients, and mean surface and 95% Hausdorff distances were calculated versus clinician-delineated references. Deep learning autosegmentation outperformed atlas-based and model-based methods. However, edge case performance was lower versus the normal cohort (0.12 mean DSC reduction). Anatomic variation presents challenges to commercial autosegmentation.

Original language	English (US)
Article number	100413
Journal	Physics and Imaging in Radiation Oncology
Volume	25
DOIs	https://doi.org/10.1016/j.phro.2023.100413
State	Published - Jan 2023

Keywords

Anatomical variability
Autosegmentation
Deep learning
Edge case
Prostate cancer

ASJC Scopus subject areas

Radiation
Oncology
Radiology Nuclear Medicine and imaging

Access to Document

10.1016/j.phro.2023.100413

Cite this

@article{55fd6e76d1534811bbd2775ce8197225,

title = "Stress-testing pelvic autosegmentation algorithms using anatomical edge cases",

abstract = "Commercial autosegmentation has entered clinical use, however real-world performance may suffer in certain cases. We aimed to assess the influence of anatomic variants on performance. We identified 112 prostate cancer patients with anatomic variations (edge cases). Pelvic anatomy was autosegmented using three commercial tools. To evaluate performance, Dice similarity coefficients, and mean surface and 95% Hausdorff distances were calculated versus clinician-delineated references. Deep learning autosegmentation outperformed atlas-based and model-based methods. However, edge case performance was lower versus the normal cohort (0.12 mean DSC reduction). Anatomic variation presents challenges to commercial autosegmentation.",

keywords = "Anatomical variability, Autosegmentation, Deep learning, Edge case, Prostate cancer",

author = "Aasheesh Kanwar and Brandon Merz and Cheryl Claunch and Shushan Rana and Arthur Hung and Thompson, {Reid F.}",

note = "Publisher Copyright: {\textcopyright} 2023",

year = "2023",

month = jan,

doi = "10.1016/j.phro.2023.100413",

language = "English (US)",

volume = "25",

journal = "Physics and Imaging in Radiation Oncology",

issn = "2405-6316",

publisher = "Elsevier Ireland Ltd",

}

TY - JOUR

T1 - Stress-testing pelvic autosegmentation algorithms using anatomical edge cases

AU - Kanwar, Aasheesh

AU - Merz, Brandon

AU - Claunch, Cheryl

AU - Rana, Shushan

AU - Hung, Arthur

AU - Thompson, Reid F.

PY - 2023/1

Y1 - 2023/1

N2 - Commercial autosegmentation has entered clinical use, however real-world performance may suffer in certain cases. We aimed to assess the influence of anatomic variants on performance. We identified 112 prostate cancer patients with anatomic variations (edge cases). Pelvic anatomy was autosegmented using three commercial tools. To evaluate performance, Dice similarity coefficients, and mean surface and 95% Hausdorff distances were calculated versus clinician-delineated references. Deep learning autosegmentation outperformed atlas-based and model-based methods. However, edge case performance was lower versus the normal cohort (0.12 mean DSC reduction). Anatomic variation presents challenges to commercial autosegmentation.

AB - Commercial autosegmentation has entered clinical use, however real-world performance may suffer in certain cases. We aimed to assess the influence of anatomic variants on performance. We identified 112 prostate cancer patients with anatomic variations (edge cases). Pelvic anatomy was autosegmented using three commercial tools. To evaluate performance, Dice similarity coefficients, and mean surface and 95% Hausdorff distances were calculated versus clinician-delineated references. Deep learning autosegmentation outperformed atlas-based and model-based methods. However, edge case performance was lower versus the normal cohort (0.12 mean DSC reduction). Anatomic variation presents challenges to commercial autosegmentation.

KW - Anatomical variability

KW - Autosegmentation

KW - Deep learning

KW - Edge case

KW - Prostate cancer

UR - http://www.scopus.com/inward/record.url?scp=85147289823&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85147289823&partnerID=8YFLogxK

U2 - 10.1016/j.phro.2023.100413

DO - 10.1016/j.phro.2023.100413

M3 - Article

AN - SCOPUS:85147289823

SN - 2405-6316

VL - 25

JO - Physics and Imaging in Radiation Oncology

JF - Physics and Imaging in Radiation Oncology

M1 - 100413

ER -

Stress-testing pelvic autosegmentation algorithms using anatomical edge cases

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this