Accurate dose calculation is essential for CyberKnife, a non-invasive robotic radiation therapy that delivers highly focused radiation beams to tumors while aiming to protect surrounding healthy tissues.
Monte Carlo dose calculation is widely regarded as a highly accurate method, especially for complex treatment sites such as the lung and liver. However, its relatively long computation time can limit routine use in clinical treatment planning and verification.
To improve both accuracy and efficiency, a physics- and spatially informed diffusion model with a Vision Transformer, named PSIDMViT, was developed for Monte Carlo-consistent dose prediction. Their findings are published in the KeAi journal Intelligent Oncology .
“The model uses planning CT images, finite-size pencil beam dose distributions, and signed distance maps derived from anatomical structures,” shares senior and co-corresponding author Lian Zhang. “By combining anatomical information, physical dose priors, and spatial structure constraints, PSIDMViT learns to predict dose distributions that are consistent with Monte Carlo calculations.” (Figure 1)
The study retrospectively included 251 patients who received CyberKnife radiotherapy, including 117 patients with head-and-neck cancer, 76 with lung cancer, and 58 with liver cancer. Monte Carlo dose distributions were used as the reference standard.
“Compared with conventional finite-size pencil beam calculations and baseline deep-learning models, PSIDMViT showed better agreement with Monte Carlo reference doses,” says Zhang. “The average 3D Gamma passing rates were 98.000% ± 2.500% for head-and-neck cancer, 94.000% ± 1.500% for lung cancer, and 95.000% ± 1.800% for liver cancer using the 1%/1 mm/10% criterion.”
The model also improved computational efficiency. “It reduced the time required to obtain Monte Carlo–consistent dose estimates from approximately one hour to about 18 minutes,” adds Zhang.
These findings suggest that physics-informed artificial intelligence may help make high-accuracy dose estimation more practical for CyberKnife treatment planning and dose verification.
“Further external and multi-institutional validation will be needed before routine clinical implementation. However, this study indicates that PSIDMViT may provide a promising strategy for improving the accuracy and efficiency of radiotherapy dose prediction,” Zhang notes.
###
Contact the author: Lian Zhang. The First Hospital of Hebei Medical University, lianzhang@hebmu.edu.cn.
The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 200 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).
Intelligent Oncology
Monte Carlo–consistent dose prediction for clinical CyberKnife radiotherapy using a physics- and spatially-informed diffusion model.
The authors declare no competing interests.