Robust Simultaneous Optimization of Catheter Positions and Dwell Times for High Dose Rate Prostate Brachytherapy
Abstract
Purpose
Catheter position uncertainty produces dosimetric error in HDR prostate brachytherapy, with reported displacements between planning and delivery of up to 5-7mm. Current inverse planning methods optimize dwell times for fixed catheter positions without accounting for these geometric uncertainties, and existing robustness evaluations typically apply uniform rigid shifts to all catheters simultaneously, failing to capture the independent, stochastic deviations observed clinically. This work develops a novel inverse planning framework that jointly optimizes catheter trajectory selection and dwell times to reduce sensitivity to independent catheter placement errors.
Methods
Thirty-two prostate HDR cases were retrospectively analyzed. Dose was modeled using a TG-43U1 influence matrix. For each clinical catheter, four candidate trajectories were generated at 5mm in anterior-posterior and left-right directions. Optimization minimized a quadratic dose fidelity term with group minimax concave penalty (gMCP) for catheter-level sparsity and a surface-restricted dose gradient penalty (SDGP) on OAR surface voxels. Four strategies were compared: clinical, clinical+SDGP, dose fidelity+gMCP, and dose fidelity+gMCP+SDGP. All plans were normalized to PTV V100=95%. Robustness was evaluated using 1,000 Monte Carlo trials per patient with independent Gaussian displacements applied to each catheter separately (σ=0-3mm). Endpoints included PTV V100, bladder/rectum V75, urethra D0.1cc, and NRG-GU009 acceptability rate. Statistical comparisons used paired Wilcoxon signed-rank tests with Bonferroni correction.
Results
At σ=1.5mm (mean 3D displacement 2.4mm per catheter), dose fidelity+gMCP+SDGP maintained mean PTV V100 of 92.0±1.1% versus 89.5±1.2% for clinical planning (p<0.001). Notably, gMCP alone (89.8±0.9%) performed similarly to clinical, indicating sparsity regularization alone does not confer robustness. Bladder V75 decreased from 0.3±0.3 cc to 0.1±0.2 cc (p<0.01). The clinical acceptability rate based on NRG-GU009 criteria significantly improved from 50.0% to 77.7% (p<0.001).
Conclusion
Coupling gMCP-based catheter selection with OAR surface gradient control significantly improves robustness to independent catheter placement uncertainty, evaluated under a more clinically realistic perturbation model than uniform rigid shifts.