SU‐E‐T‐598: Variability of Computer‐Generated Organ at Risk Contours as Part of An Automated Deformable Registration Workflow for Prostate Cancer

S. Gardner, N. Zaorsky, K. Yamoah, Y. Cui, Y. Xiao, R. Den, Matthew Thomas Studenski

Research output: Contribution to journalArticle

Abstract

Purpose: To compare the contouring variability of organs at risk (OAR) on planning CT (PCT), first day cone‐beam computed tomography (CBCT1), and computer‐generated contours on subsequent fractions (CBCTf) through the use of a deformable registration workflow, as could be used for an adaptive radiotherapy planning scheme for prostate cancer. Methods: Eleven observers (6 physician‐residents, 5 physicists; all with extensive experience with prostate anatomy, treatment planning) contoured the bladder, rectum, and patient skin, using Elekta Focal on PCT (GE HiLite) and CBCT (Elekta XVI). These contours were compared to consensus contours generated with STAPLE method (CERR) which combined the contours of two recognized physician experts. This was done for the PCT and CBCT1. The CBCT1 contours were transformed via deformable registration workflow (MIM software) to 10 subsequent fractions (CBCTf) creating 333 computer‐generated OAR contours in total. The computer‐generated contours were compared to manually segmented reference contours, which were reviewed with the leading physicist and physician of the study. Dice coefficient was used to quantify variability. Statistical analysis utilized two‐sided t‐test; p‐values<0.05 significant. Results: The average Dice coefficient among all users for the bladder: PCT‐91.3%, CBCT1‐89.4%, CBCTf‐87.4% (average Dice coefficient over 10 fractions of computer‐generated contours). Rectum: PCT‐83.3%, CBCT1‐76.2%, CBCTf‐72.9%. Patient skin: PCT‐99.6%, CBCT1‐99.5%, CBCTf‐97.7%. The Dice coefficient difference between PCT and CBCT1 was statistically significant for bladder and rectum (p‐values 0.045 and 0.0013, respectively). CBCTf contours exhibited comparable variability to CBCT1 contour (average Dice coefficient difference 2–4%). Conclusion: The largest variability in computer‐generated contours occurred at superior‐inferior borders of rectum and patient skin. Our results show computer‐generated contours using deformable registration do not contribute significant additional variability to OAR contours. Further investigation which includes dosimetric impact of contouring variability of OAR is needed to assess the viability of computer‐generated contours for adaptive planning.

Original languageEnglish (US)
Number of pages1
JournalMedical Physics
Volume40
Issue number6
DOIs
StatePublished - Jan 1 2013
Externally publishedYes

Fingerprint

Organs at Risk
Workflow
Rectum
Prostatic Neoplasms
Urinary Bladder
Skin
Physicians
Prostate
Anatomy
Radiotherapy
Software
Tomography

ASJC Scopus subject areas

  • Biophysics
  • Radiology Nuclear Medicine and imaging

Cite this

SU‐E‐T‐598 : Variability of Computer‐Generated Organ at Risk Contours as Part of An Automated Deformable Registration Workflow for Prostate Cancer. / Gardner, S.; Zaorsky, N.; Yamoah, K.; Cui, Y.; Xiao, Y.; Den, R.; Studenski, Matthew Thomas.

In: Medical Physics, Vol. 40, No. 6, 01.01.2013.

Research output: Contribution to journalArticle

@article{b3d8e9c697914ba7bc7710164337b3b2,
title = "SU‐E‐T‐598: Variability of Computer‐Generated Organ at Risk Contours as Part of An Automated Deformable Registration Workflow for Prostate Cancer",
abstract = "Purpose: To compare the contouring variability of organs at risk (OAR) on planning CT (PCT), first day cone‐beam computed tomography (CBCT1), and computer‐generated contours on subsequent fractions (CBCTf) through the use of a deformable registration workflow, as could be used for an adaptive radiotherapy planning scheme for prostate cancer. Methods: Eleven observers (6 physician‐residents, 5 physicists; all with extensive experience with prostate anatomy, treatment planning) contoured the bladder, rectum, and patient skin, using Elekta Focal on PCT (GE HiLite) and CBCT (Elekta XVI). These contours were compared to consensus contours generated with STAPLE method (CERR) which combined the contours of two recognized physician experts. This was done for the PCT and CBCT1. The CBCT1 contours were transformed via deformable registration workflow (MIM software) to 10 subsequent fractions (CBCTf) creating 333 computer‐generated OAR contours in total. The computer‐generated contours were compared to manually segmented reference contours, which were reviewed with the leading physicist and physician of the study. Dice coefficient was used to quantify variability. Statistical analysis utilized two‐sided t‐test; p‐values<0.05 significant. Results: The average Dice coefficient among all users for the bladder: PCT‐91.3{\%}, CBCT1‐89.4{\%}, CBCTf‐87.4{\%} (average Dice coefficient over 10 fractions of computer‐generated contours). Rectum: PCT‐83.3{\%}, CBCT1‐76.2{\%}, CBCTf‐72.9{\%}. Patient skin: PCT‐99.6{\%}, CBCT1‐99.5{\%}, CBCTf‐97.7{\%}. The Dice coefficient difference between PCT and CBCT1 was statistically significant for bladder and rectum (p‐values 0.045 and 0.0013, respectively). CBCTf contours exhibited comparable variability to CBCT1 contour (average Dice coefficient difference 2–4{\%}). Conclusion: The largest variability in computer‐generated contours occurred at superior‐inferior borders of rectum and patient skin. Our results show computer‐generated contours using deformable registration do not contribute significant additional variability to OAR contours. Further investigation which includes dosimetric impact of contouring variability of OAR is needed to assess the viability of computer‐generated contours for adaptive planning.",
author = "S. Gardner and N. Zaorsky and K. Yamoah and Y. Cui and Y. Xiao and R. Den and Studenski, {Matthew Thomas}",
year = "2013",
month = "1",
day = "1",
doi = "10.1118/1.4815026",
language = "English (US)",
volume = "40",
journal = "Medical Physics",
issn = "0094-2405",
publisher = "AAPM - American Association of Physicists in Medicine",
number = "6",

}

TY - JOUR

T1 - SU‐E‐T‐598

T2 - Variability of Computer‐Generated Organ at Risk Contours as Part of An Automated Deformable Registration Workflow for Prostate Cancer

AU - Gardner, S.

AU - Zaorsky, N.

AU - Yamoah, K.

AU - Cui, Y.

AU - Xiao, Y.

AU - Den, R.

AU - Studenski, Matthew Thomas

PY - 2013/1/1

Y1 - 2013/1/1

N2 - Purpose: To compare the contouring variability of organs at risk (OAR) on planning CT (PCT), first day cone‐beam computed tomography (CBCT1), and computer‐generated contours on subsequent fractions (CBCTf) through the use of a deformable registration workflow, as could be used for an adaptive radiotherapy planning scheme for prostate cancer. Methods: Eleven observers (6 physician‐residents, 5 physicists; all with extensive experience with prostate anatomy, treatment planning) contoured the bladder, rectum, and patient skin, using Elekta Focal on PCT (GE HiLite) and CBCT (Elekta XVI). These contours were compared to consensus contours generated with STAPLE method (CERR) which combined the contours of two recognized physician experts. This was done for the PCT and CBCT1. The CBCT1 contours were transformed via deformable registration workflow (MIM software) to 10 subsequent fractions (CBCTf) creating 333 computer‐generated OAR contours in total. The computer‐generated contours were compared to manually segmented reference contours, which were reviewed with the leading physicist and physician of the study. Dice coefficient was used to quantify variability. Statistical analysis utilized two‐sided t‐test; p‐values<0.05 significant. Results: The average Dice coefficient among all users for the bladder: PCT‐91.3%, CBCT1‐89.4%, CBCTf‐87.4% (average Dice coefficient over 10 fractions of computer‐generated contours). Rectum: PCT‐83.3%, CBCT1‐76.2%, CBCTf‐72.9%. Patient skin: PCT‐99.6%, CBCT1‐99.5%, CBCTf‐97.7%. The Dice coefficient difference between PCT and CBCT1 was statistically significant for bladder and rectum (p‐values 0.045 and 0.0013, respectively). CBCTf contours exhibited comparable variability to CBCT1 contour (average Dice coefficient difference 2–4%). Conclusion: The largest variability in computer‐generated contours occurred at superior‐inferior borders of rectum and patient skin. Our results show computer‐generated contours using deformable registration do not contribute significant additional variability to OAR contours. Further investigation which includes dosimetric impact of contouring variability of OAR is needed to assess the viability of computer‐generated contours for adaptive planning.

AB - Purpose: To compare the contouring variability of organs at risk (OAR) on planning CT (PCT), first day cone‐beam computed tomography (CBCT1), and computer‐generated contours on subsequent fractions (CBCTf) through the use of a deformable registration workflow, as could be used for an adaptive radiotherapy planning scheme for prostate cancer. Methods: Eleven observers (6 physician‐residents, 5 physicists; all with extensive experience with prostate anatomy, treatment planning) contoured the bladder, rectum, and patient skin, using Elekta Focal on PCT (GE HiLite) and CBCT (Elekta XVI). These contours were compared to consensus contours generated with STAPLE method (CERR) which combined the contours of two recognized physician experts. This was done for the PCT and CBCT1. The CBCT1 contours were transformed via deformable registration workflow (MIM software) to 10 subsequent fractions (CBCTf) creating 333 computer‐generated OAR contours in total. The computer‐generated contours were compared to manually segmented reference contours, which were reviewed with the leading physicist and physician of the study. Dice coefficient was used to quantify variability. Statistical analysis utilized two‐sided t‐test; p‐values<0.05 significant. Results: The average Dice coefficient among all users for the bladder: PCT‐91.3%, CBCT1‐89.4%, CBCTf‐87.4% (average Dice coefficient over 10 fractions of computer‐generated contours). Rectum: PCT‐83.3%, CBCT1‐76.2%, CBCTf‐72.9%. Patient skin: PCT‐99.6%, CBCT1‐99.5%, CBCTf‐97.7%. The Dice coefficient difference between PCT and CBCT1 was statistically significant for bladder and rectum (p‐values 0.045 and 0.0013, respectively). CBCTf contours exhibited comparable variability to CBCT1 contour (average Dice coefficient difference 2–4%). Conclusion: The largest variability in computer‐generated contours occurred at superior‐inferior borders of rectum and patient skin. Our results show computer‐generated contours using deformable registration do not contribute significant additional variability to OAR contours. Further investigation which includes dosimetric impact of contouring variability of OAR is needed to assess the viability of computer‐generated contours for adaptive planning.

UR - http://www.scopus.com/inward/record.url?scp=85024781368&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85024781368&partnerID=8YFLogxK

U2 - 10.1118/1.4815026

DO - 10.1118/1.4815026

M3 - Article

AN - SCOPUS:85024781368

VL - 40

JO - Medical Physics

JF - Medical Physics

SN - 0094-2405

IS - 6

ER -