Abstract
Artificial intelligence (AI)-based auto-contouring techniques in radiation treatment planning require thorough clinical validation. In this study, we evaluated the geometric similarity between AI-generated auto-contours and manually drawn contours in prostate cancer treatment plans for 15 patients using OncoStudio (Oncosoft, Seoul, South Korea). The mean Dice similarity coefficients for the bladder, rectum, and right and left femoral heads were 0.90, 0.95, 0.97, and 0.97, respectively. The mean 95th percentile Hausdorff distances were 7.3 mm, 7.1 mm, 5.6 mm, and 5.0 mm for the same structures. These results demonstrate that OncoStudio's auto-segmentation provides high-quality contours suitable for prostate cancer treatment planning.