Respiratory motion prediction based on deep artificial neural networks in CyberKnife system: A comparative study.

BACKGROUND: In external beam radiotherapy, a prediction model is required to compensate for the temporal system latency that affects the accuracy of radiation dose delivery. This study focused on a thorough comparison of seven deep artificial neural networks to propose an accurate and reliable prediction model. METHODS: Seven deep predictor models are trained and tested with 800 breathing signals. In this regard, a nonsequential-correlated hyperparameter optimization algorithm is developed to find the best configuration of parameters for all models. The root mean square error (RMSE), mean absolute error, normalized RMSE, and statistical F-test are also used to evaluate network performance. RESULTS: Overall, tuning the hyperparameters results in a 25%-30% improvement for all models compared to previous studies. The comparison between all models also shows that the gated recurrent unit (GRU) with RMSE = 0.108 ± 0.068 mm predicts respiratory signals with higher accuracy and better performance. CONCLUSION: Overall, tuning the hyperparameters in the GRU model demonstrates a better result than the hybrid predictor model used in the CyberKnife VSI system to compensate for the 115 ms system latency. Additionally, it is demonstrated that the tuned parameters have a significant impact on the prediction accuracy of each model.

A retrospective multi-center feasibility study of a new PTV margin estimation approach for moving targets using CyberKnife log files.

PURPOSE: This study investigates a new approach for estimating the planning target volume (PTV) margin for moving tumors treated with robotic stereotactic body radiotherapy (SBRT). METHODS: In this new approach, the covariance of modeling and prediction errors was estimated using error propagation and implemented in the Van Herk formula to form a Modified Van Herk formula (MVHF). To perform a retrospective multi-center analysis, the MVHF was studied using 163 patients treated with different system versions of robotic SBRT (G3 version 6.2.3, VSI version 8.5, and VSI version 9.5) and compared with two established PTV margins estimation methods: The original Van Herk Formula (VHF) and the Uncertainty Estimation Method (UEM). RESULTS: Overall, the PTV margins provided by the three formalisms are similar with 4-5 mm in the lung region and 4 mm in abdomen region to the PTV margins used in clinical. Furthermore, when analyzing individual patients, a difference of up to 1 mm was found between the PTV margin estimations using MVHF and VHF. The corresponding average discrepancies for the superior-inferior (SI) direction ranged between -0.19 mm to 0.38 mm in CK G3 version 6.2.3, -0.36 mm to 0.33 mm in CK VSI version 8.5, and -0.34 mm to 0.40 mm in CK VSI version 9.5. CONCLUSIONS: It was found that for the lower left lung, upper left lung, lower right lung, upper right lung, central liver, and upper liver, the effect of covariance between model and prediction errors in SI direction was around 20%, 30%, 25%, 25%, 25%, and 30%, respectively. Notable covariance effects between model and prediction errors can be considered in PTV margin estimation using a modified VHF, which allowed for more precise target localization in robotic SBRT for moving tumors. Overall, in each of the three directions, the difference between MVHF and utilized clinical margins is 0.65 mm in the lung and abdominal region. Therefore, to improve the clinical PTV margins with the new approach, it is suggested to use the adaptive PTV margins in the next fractions.

Target margin design through analyzing a large cohort of clinical log data in the cyberknife system.

PURPOSE: Calculating the adequate target margin for real-time tumor tracking using the Cyberknife system is a challenging issue since different sources of error exist. In this study, the clinical log data of the Cyberknife system were analyzed to adequately quantify the planned target volume (PTV) margins of tumors located in the lung and abdomen regions. METHODS: In this study, 45 patients treated with the Cyberknife module were examined. In this context, adequate PTV margins were estimated based on the Van Herk formulation and the uncertainty estimation method by considering the impact of errors and uncertainties. To investigate the impact of errors and uncertainties on the estimated PTV margins, a statistical analysis was also performed. RESULTS: Our study demonstrates five different sources of errors, including segmentation, deformation, correlation, prediction, and targeting errors, which were identified as the main sources of error in the Cyberknife system. Furthermore, the clinical evaluation of the current study reveals that the two different formalisms provided almost identical PTV margin estimates. Additionally, 4-5 mm and 5 mm margins on average could provide adequate PTV margins at lung and abdomen tumors in all three directions, respectively. Overall, it was found that concerning the PTV margins, the impact of correlation and prediction errors is very high, while the impact of robotics error is low. CONCLUSIONS: The current study can address two limitations in previous researches, namely insufficient sample sites and a smaller number of patients. A comparison of the present results concerning the lung and abdomen areas with other studies reveals that the proposed strategy could provide a better reference in selection the PTV margins. To our knowledge, this study is one of the first attempts to estimate the PTV margins in the lung and abdomen regions for a large cohort of patients treated using the Cyberknife system.