Prospective Evaluation of Automated Contouring for CT-Based Brachytherapy Treatment Planning of Gynecologic Malignancies

Abstract

<h3>Purpose/Objective(s)</h3> The use of artificial intelligence to automatically-contour OARs in gynecologic radiation treatment has been well-established in the literature. Yet, there is limited data on the prospective use of AC in clinical practice. The objective of this study was to assess the accuracy and efficiency of AC for CT-based brachytherapy treatment planning of gynecologic malignancies. <h3>Materials/Methods</h3> An in-house AC tool was used to automatically-delineate five organs-at-risk (OARs) used for gynecologic radiation treatment planning: the bladder, small bowel, sigmoid, rectum and urethra. The accuracy of each auto-contour was evaluated by the treating physician using a 5-point Likert scale: a score of five indicated the contour could be used without edits, while a score of one indicated the contour was unusable (Table 1). To assess the efficiency of AC in clinical practice, the planning time of a prospective AC cohort was compared to the planning time of a retrospective MC cohort. Planning time of each case was quantified as the time between import of CT image to TPS and final plan approval. Mean, standard deviation and standard error of each cohort's time to approval was analyzed via unpaired t-test with Welch's correction. <h3>Results</h3> Eight prospective AC cases from January 2022 to February 2022 and thirty retrospective MC cases from July 2021 to January 2022 were included in the study. The average time to plan approval in the AC cohort was 30% less than MC cohort (AC vs MC, 109.0 ± 6.4 minutes vs. 155.1 ± 15.4 minutes, p=0.0092). The AC group had less time variance between cases with a standard deviation of 17.0 minutes, compared to a 31.5-minute variance of the MC cohort. The mean accuracy score in the AC cohort was 4.8 (SD=0.7) for the bladder, 4.0 (SD=0.9) for the small bowel, 4.2 (SD=0.7) for the sigmoid, 4.6 (SD=0.7) for the rectum, and 4.8 (SD=0.4) for the urethra. No auto-contoured OARs required major edits from the attending physician. Physician slice-by-slice review of the AC contours took on average 3.6 minutes (SD=1.6) revising OARs (including edits if needed) and overall contouring time (included adding targets) took an average of 14.1 minutes (SD=6.0). <h3>Conclusion</h3> Automated contouring appears to be safe and accurate for use in clinical practice. Clinical implementation of AC shows promise to streamline radiation treatment workflows and decrease time required to design and approve gynecologic brachytherapy plans.