Exhibit Hall | Forum 3
Purpose: To quantify the impact of automated contouring (AC) for CT-based HDR GYN brachytherapy treatment on clinical workflow and efficiency.
Methods: An in-house AC tool capable of delineating five OARs (bladder, rectum, sigmoid, small bowel, and urethra) from a HDR GYN CT image series was developed and clinically commissioned. Furthermore, this tool was used to prospectively produce contours for nine clinical cases (4 tandem and ovoid, 2 tandem and ring, and 3 interstitial needle cases). The accuracy of each auto-contour was qualitatively evaluated by a physician using a 5-point Likert scale (5 = contour used without edits, 1 = contour unusable). The contours were also quantitatively evaluated using dose constraint metrics (D2cc or D0.1cc) and dice coefficients for the auto-contours and the final clinical contours. Finally, timestamps for image import, contouring, and plan approval were analyzed and compared between the nine AC GYN cases and thirty previously treated GYN cases that were manually contoured.
Results: The average time from CT simulation to plan approval was significantly (p < 0.015) reduced from 155±84.5 minutes to 112±19.6 minutes when comparing the manually and automatically contoured clinical cases, respectively, using a Welch’s t-test. All of the contours produced by the AC tool had a physician score of 3 or higher, meaning all contours were useful as bases for contouring. The average physician scores are as follows: Bladder - 4.78±0.67, Small Bowel - 4.11±0.93, Sigmoid - 4.25±0.71, Rectum - 4.63±0.74, and Urethra - 4.67±0.5. For the quantitative comparison, 68%, 88% and 100% of contours had dice coefficients greater than 0.95, 0.9, and 0.84, respectively. All but 1 AC structure agreed with clinical goals (D2cc/D0.1cc) compared to the final clinical contour.
Conclusion: In this study we prospectively demonstrate improvement of planning efficiency by utilizing an automated contouring tool.