Effect of inter-planner variability on knowledge-based treatment planning in moderately hypofractionated prostate cancer

Abstract

This study aimed to compare the performance of RapidPlan models developed by different planners and to evaluate the impact of inter-planner variability on treatment plan quality. Three planners with different levels of clinical experience developed RapidPlan models (Models A, B, and C) using the same set of 30 prostate cancer cases. Each planner manually created treatment plans for these cases, referring to PlanIQ for dosimetric guidance. PlanIQ provides estimates of achievable dose sparing for organs-at-risk (OAR) based on computed tomography images and anatomical contours, and was used to guide the planning process. Each model was then used to generate treatment plans (Plans A, B, and C). The models were evaluated using 12 independent cases not included in model training, and plan quality was assessed using dose–volume parameters for the planning target volume excluding the rectum (PTV-Rectum), rectum, and bladder. Plan A yielded better dose uniformity for the PTV-Rectum, as indicated by lower D2% and higher D98%. For the rectum, the mean volume receiving 58 Gy (V58 Gy) was lowest in Plan C, followed by Plans B and A, indicating that Model C provided the most effective dose sparing. Plan A yielded superior dose uniformity for the PTV-Rectum, whereas Plan C more effectively spared OAR. Although no statistically significant differences were observed, the results suggest that planner-related variability can influence model performance. These findings underscore the importance of quality control in model construction and support the feasibility of shared or standardized RapidPlan models, particularly in resource-limited settings.