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

Abstract

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.

Prediction of the position of external markers using a recurrent neural network trained with unbiased online recurrent optimization for safe lung cancer radiotherapy

“…However, the task (image prediction) and the network used, based on the combination of a conditional variational autoencoder and LSTMs, differ from those in our work. Miandoab et al claimed that "for a higher system latency, a larger input window is required" [41], but that did not always appear to be accurate, for instance, when considering the UORO cross-validation graphs for h = 0.3s and h = 2.1s in Fig. 10b.…”

“…These signals lasted from 84s to 273s, with an average recording time of 145s and had a mean amplitude in the SI direction of 11mm ± 8mm (standard deviation). Similarly, Miandoab et al also achieved higher performance using a GRU trained with 26Hz Cyberknife VSI data comprising 800 records between 23min and 60min from 30 lung and abdominal cancer patients, leading to an RMSE, MAE, and nRMSE 24 of 0.108mm, 0.086mm, and 0.031, respectively, at h = 115ms [41]. However, the accuracy corresponding to f = 30Hz in our study might be lower because we report 3D errors, and irregular breathing sequences constitute almost half our entire dataset.…”

“…It was previously reported that an extensive search may not be clinically feasible due to high computational costs [19]. To address that challenge, Miandoab et al proposed a nonsequential-correlated hyperparameter optimization algorithm for deep recurrent neural networks (RNN) to reduce the hyperparameter combinations that they explored from 700 million to 30,000 [41]. It was generally found that performance decreased as the horizon increased.…”

“…As a result, these networks can learn patterns and dependencies within sequential and time-series data in an efficient manner. Some recent works demonstrated the potential of deep recurrent architectures based on LSTMs, bi-LSTMs, and bi-gated recurrent unit (bi-GRU) layers for respiratory motion prediction, and it was reported, for instance, that bi-LSTMs had better performance than the adaptiveboosting MLP model [26,55,58,41].…”

“…8c, 8f, and 8i. Indeed, several prior studies about respiratory motion forecasting with a sampling rate close to 30Hz recommend using a learning rate from 0.001 to 0.005 [41,26,58].…”