Evaluation of An Auto-Segmentation Tool On Full Field-Of-View and Limited Field-Of-View Cone Beam Computed Tomography
J Marasco*, S Hendley, J Wong, A Granatowicz, A Besemer, S Zhou, S Wang, University of Nebraska Medical Center, Omaha, NE
Presentations
PO-GePV-M-156 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: Auto-segmentation tools are rapidly changing the practice of treatment planning in radiation oncology. They have immense potential to become an essential component in accelerating target/OAR delineation for adaptive radiotherapy. The purpose of this study is to evaluate the performance of a commercial auto-segmentation tool on Cone Beam Computed Tomography (CBCT) with different scanning protocols.
Methods: Forty-two prostate cancer patients who had undergone radiotherapy treatments have been retrospectively selected for this study. Twenty-two patients received SpaceOAR hydrogel (Boston Scientific) implant prior to radiotherapy. All patients were treated on Varian TrueBeam or Varian Edge Linear Accelerators (Varian Medical Systems). The earliest Full Field-of-View (FOV) CBCTs and the limited FOV CBCT on the nearest dates were selected as the test images to evaluate AutoContour, an auto-segmentation tool developed by Radformation (New York). The rectum of each patient was delineated on the CBCTs by a trained researcher and validated by a board-certified medical physicist before the automatic contours were generated to avoid the bias for manual delineations. Both the manual and the automatic contours of the rectum were converted into a binary mask for DICE coefficient analysis. The volumes of manual and automatic contours were compared.
Results: The auto-segmentation tool successfully generated rectum contours on all the full FOV CBCTs but failed on 15 out of 42 (35.7%) limited FOV CBCTs. On the full FOV CBCTs, the average DICE coefficient between the manual and automated contours of the rectum was 0.80 ± 0.12. The absolute volume difference between the manual and automatic contours in the patients with SpaceOAR is slightly greater than that in the patients without SpaceOAR.
Conclusion: AutoContour automatically generated acceptable rectum contours on full FOV CBCTs with or without SpaceOAR. Improvements are needed to apply this tool on limited FOV CBCTs.
Funding Support, Disclosures, and Conflict of Interest: The project is partially supported by the Otis Glebe Medical Foundation
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email