Dosimetric comparison between laterality-specific and general knowledge-based planning models for nonsmall cell lung cancer

Abstract

To investigate the dosimetric impact of laterality-specific RapidPlan models for nonsmall cell lung cancer. Three RapidPlan models were developed and validated for Right, Left, and General conventional lung radiotherapy. Each model was trained using 50 plans. The right and left models consisted of plans corresponding to their respective laterality. Twenty-five cases were randomly chosen from each laterality-specific model to craft a general model. All models shared identical optimization objectives and the same target and OAR structures. Validation included 13 right-sided and 13 left-sided cases optimized using each RapidPlan model without intervention and normalized such that the prescription dose covered 95% of the target volume. Statistical analysis using a paired sample t-test (p < 0.01) assessed dosimetric endpoints based on RTOG 0617 criteria. For right-sided cases, spinal cord Dmax and D0.03cc were lowest in the left model and highest in the right model (21.08 Gy and 21.22 Gy vs 23.67 Gy and 24.08 Gy). D_max and D_0.03cc esophagus mean dose was also lower in the left model compared to the right model (p < 0.01) for both left and right-sided cases. However, overall plan quality exhibited no substantial difference between general and laterality-specific models. Despite observing small but statistically significant differences, there is no discernible difference in plan quality between laterality-specific and general models, suggesting that a single RapidPlan model is sufficient.