Outcome-Driven HDR Gynecological Brachytherapy Plan Selection Using TCP and NTCP Trade-Offs In Multi-Criteria Optimization
Abstract
Purpose
Novel GPU-based multicriteria optimization (gMCO) for high-dose-rate gynecological brachytherapy (GYN-BT) generates large plan pools satisfying EMBRACE-II dosimetric constraints (Pötter et al., 2018). Differentiating plans with comparable dosimetry but potentially distinct predicted outcomes remain challenging. In this work, we integrate six TCP/NTCP models via the EviGUIDE framework (Ecker et al., 2023) to explore patient-specific outcome trade-offs across gMCO plan pools.
Methods
A retrospective study was performed with eight cervical cancer patients undergoing four GYN-BT fractions. Per fraction, gMCO generated 20000 treatment plans across ten catheter configurations. Plans were first filtered to satisfy EMBRACE-II hard dose constraints. Remaining plans were compared to the (non-gMCO) clinical plan using a TCP allowance parameter: only plans with TCP reduction ≤ allowance (relative to clinical) were kept. Among these, the plan maximizing summed NTCP reduction was chosen. , we identified saturation points, where maximum achievable NTCP benefit is reached as remaining plans either violate dosimetric constraints or offer no further improvement. We considered two EQD2 estimations
Methods
(1) adaptive EQD2, incorporating prior fraction doses, and (2) blind EQD2, assuming all fractions deliver identical dose.
Results
Most NTCP benefits were captured within 0-1% TCP allowance. Cumulative NTCP reductions at saturation ranged from 17.9% to 53.2% (mean: 33.7%) relative to clinical plans. Even at 0% allowance, 3/8 patients achieved ≥ 50% of their maximum NTCP benefit; all patients reached 50% of maximum benefit by 0.65% allowance. At 1% allowance, cumulative NTCP reductions ranged from 11.6% to 36.2%. EQD2 summing method influenced both predicted NTCP/TCP values for which adaptive and blind NTCPs differ by -2.96% to +1.99%.
Conclusion
Important toxicity reductions appear captured at minimal tumor control cost. This supports TCP/NTCP trade-offs as potential decision-support for gMCO plan selection. Adaptive EQD2 accumulation propagates inter-fraction dose variability into trade-off estimation.