Author: Nan Li, Yaoying Liu, Shouping Xu, Gaolong Zhang 👨🔬
Affiliation: Department of Radiation Oncology, School of Physics, Beihang University, School of physics, Beihang University, National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College 🌍
Purpose: In intensity-modulated proton therapy (IMPT) for head and neck cancer, CBCT registration ensures accurate setup, minimizing dose errors. Unlike IMRT, IMPT plans directly define tumor volumes using CTV and GTV. Robust optimization must account for patient position errors and proton range uncertainties. For head and neck patients, position errors are 0.3 cm isotropic, and range uncertainties are 3.5%. Evaluating 21 scenarios is necessary but time-consuming. More significant errors would require additional scenarios. A state space model (SSM) is recommended to improve planning efficiency, reduce wait times, and optimize workflow.
Methods: We developed VSSA-GAN, which integrates a visual state space module into the network generator. We incorporated beam weights, angles, energy layer information, and isocenter data during training. Additionally, we designed a novel "NT Gamma" loss function that separately accounts for dose deposition location errors and numerical discrepancies. The experimental dataset comprised 78 brain tumor patients who had undergone proton therapy, with 58 patients allocated for training and the remaining 20 for validation.
Results: For single-beam doses, the 3D overall gamma passing rates (GPRs) were as follows: 99.81% ± 0.10% under the 3mm/3% threshold, 99.32% ± 0.13% under the 2mm/3% threshold, and 98.70% ± 0.11% under the 1mm/3% threshold. For entire-plan doses, the 3D overall gamma passing rates were: 99.72% ± 0.10% under the 3mm/3% threshold, 99.11% ± 0.12% under the 2mm/3% threshold, and 98.61% ± 0.14% under the 1mm/3% threshold. The average differences in target coverage (TC), dose selectivity (DS), gradient index (GI), and homogeneity index (HI) were 0.32% ± 0.16%, 0.05 ± 0.01, 0.09 ± 0.02, and 0.08 ± 0.03, respectively.
Conclusion: This innovative approach significantly improves the speed of robust-proton-dose optimization, substantially enhances the efficiency of physicists, and comprehensively optimizes the overall workflow of proton therapy.