Author: Theodore Higgins Arsenault, Kenneth W. Gregg, Beatriz Guevara, Lauren E Henke, Angela Jia, Rojano Kashani, Kyle O'Carroll, Alex T. Price, Adithya Reddy, Atefeh Rezaei, Daniel E Spratt, Runyon C. Woods 👨🔬
Affiliation: University Hospitals Seidman Cancer Center 🌍
Purpose: To evaluate the effect of unedited AI-generated contours used for online adaptive radiotherapy (FLOW-ART) on the plan quality of prostate treatments as compared to non-adaptive (non-ART) processes.
Methods: A retrospective analysis was conducted on ten prostate SBRT patients, each with daily cone-beams, alignment fiducials, and rectal spacers. All patients were prescribed 37.5Gy in 5 fractions to the prostate and proximal seminal vesicles. Organ-at-risk contours on the daily CBCTs were contoured by physicians as well as a commercial AI software. Dice similarity coefficients (DSC) and average symmetric surface distance (ASSD) were calculated for both contour sets to test similarity. Additionally, adaptive plans were generated using the daily CBCTs with un-edited AI generated contours on Ethos 2.0 and the dosimetric impact of the non-ART and un-edited FLOW-ART plan were evaluated using DVH metrics based on the daily ground truth (physician generated) anatomy.
Results: Average DSC for the rectum, bladder, and sigmoid colon were 93.2, 93.2, and 75.4 respectively, and for ASSD were 0.1, 0.1, and 0.3, respectively. An erroneous AI contour was present in a single fraction of one patient which contoured the rectal spacer gel as sigmoid colon. For the 49/50fx evaluable, FLOW-ART plans generated from unedited AI autocontours were dosimetrically superior to non-ART plans for all high-dose rectal constraints (P <0.05), without statistical difference in bladder sparing or PTV V37.5Gy coverage (p >0.05), and without clinically meaningful difference in CTV V37.5Gy coverage (<1% difference; p <0.01).
Conclusion: AI autocontours generated in the FLOW-ART workflow were comparable to physician-delineated ground-truth contours across daily CBCTs. Additionally, OAR sparing for most parameters was improved using the FLOW-ART workflow as compared to a traditional non-ART workflow with an added benefit of reduced planning time. This work suggests the feasibility of introducing the FLOW-RT workflow with the caveat that diligence in review is essential.