Poster Session A   |   11:45am Expo - Hall A & C   |   Poster ID #423

Understanding the role of geometric indices in contour peer review for organ-at-risk dose evaluation in radiotherapy

Program:
Prevention
Category:
Tertiary Prevention
FDA Status:
Cleared
CPRIT Grant:
Cancer Site(s):
Head and Neck
Authors:
Barbara Marquez
The University of Texas Health Science Center at Houston
Laurence E Court
The University of Texas Health Science Center at Houston
Kai Huang
The University of Texas Health Science Center at Houston
Constance Owens
The University of Texas Health Science Center at Houston
Alexandra Leone
The University of Texas M.D. Anderson Cancer Center
Mary Gronberg
The University of Texas Health Science Center at Houston
Callistus Nguyen
The University of Texas M.D. Anderson Cancer Center
David T Fuentes
The University of Texas M.D. Anderson Cancer Center
Anuja Jhingran
The University of Texas M.D. Anderson Cancer Center
Tucker J Netherton
The University of Texas M.D. Anderson Cancer Center
Christine B Peterson
The University of Texas M.D. Anderson Cancer Center
Surendra Prajapati
The University of Texas M.D. Anderson Cancer Center
Julianne Pollard-Larkin
The University of Texas M.D. Anderson Cancer Center
Thomas J Whitaker
The University of Texas M.D. Anderson Cancer Center

Introduction

To understand how geometric indices commonly reported in autosegmentation performance are related to dosimetric differences on clinical plans. This work is in an effort to use autosegmentation models as key tools in peer review of organ-at-risk (OAR) contours in radiotherapy to increase OAR sparing and thus reduce toxicity effects.

Methods

An FDA-cleared auto-contouring model is routinely used in our institution’s head & neck (H&N) service. H&N normal structure auto-contours are reviewed and edited as necessary prior to clinical use. In this study, we calculated surface Dice similarity coefficient (sDSC) and distance from each normal structure to the closest PTV to quantify the extent of clinical edits over 292 clinically treated cases. We evaluated the impact of these indices on reported dose from the original clinical plan by comparing them against the mean and maximum doses to the auto-contoured versus edited structures. 

Results

Overall, 90% of all geometric comparisons between auto-contoured and edited structures scored sDSC > 0.86.  Dosimetrically, difference between the mean and max doses to auto- and edited-contours was < 2Gy for 94% of structures. There was a clear monotonically decreasing relationship (R2 = 0.611) between higher sDSC score and percent of contours identified for structure edits that resulted in a dose difference ≥ 2Gy (~3% of dose prescription). Of all structure pairs identified for max dose difference (MDD) ≥ 2Gy (261/4482), 91% were ≤ 2cm from the PTV.

Conclusion

Most cases demonstrate that degree of overlap metrics is indicative of corresponding magnitude in expected dose difference between two contour sets. Although unlikely, high sDSC values are not always indicative of insignificant dose difference; where high sDSC values fail to identify cases with dose differences >2 Gy, distance from the PTV was a key indicator.