A comparison of neural network approaches for on‐line prediction in IGRT

Abstract

Image-guided radiation therapy aims to improve the accuracy of treatment delivery by tracking tumor position and compensating for observed movement. Due to system latency it is sometimes necessary to predict tumor trajectory evolution in order to facilitate changes in beam delivery. Neural networks (NNs) have previously been investigated for predicting future tumor position because of their ability to model non-linear systems. However, no attempt has been made to optimize the NN training algorithms, and no mention has been made of potential errors which can be caused by using NNs for extrapolation purposes. In this work, after giving a brief explanation of NN theory, a comparison is made between 4 different adaptive algorithms for training time-series prediction NNs. New error criteria are introduced which highlight error maxima. Results are obtained by training the NNs using previously published data. A hybrid algorithm combining Bayesian regularization with conjugate-gradient backpropagation is demonstrated to give the best average prediction accuracy, whilst a generalized regression NN is shown to reduce the possibility of isolated large prediction errors.

Feasibility of predicting tumor motion using online data acquired during treatment and a generalized neural network optimized with offline patient tumor trajectories

“…This value is in between the range of latencies (50 to 1400 ms) observed in some image‐guided adaptive radiotherapy systems . It is well‐known that prediction accuracy deteriorates as the prediction horizon increases …”

“…Various prediction models have been investigated for use during tracking of tumor motion in radiotherapy . These include: linear prediction, auto‐regressive moving average, stationary linear regression, sinusoidal modeling, adaptive filters, Kalman filters, and neural networks . In three of the comparative studies, a neural network has been shown to outperform the other prediction models .…”

Investigation of the optimum location of external markers for patient setup accuracy enhancement at external beam radiotherapy

“…The CCA is optimal way to respect correlations and at the same time to find a corresponding correlation, in which the correlation matrix between the variables is diagonal and the correlations on the diagonal are maximized. In fact, the dimensionality of these new bases is equal to or less than the smallest dimensionality of the two variables (35) . By applying CCA algorithm, a small amount of data will be lost when more than 90% of canonical correlation of all markers is covered.…”

Real-time Tumor Tracking with Respiratory Motion Based on Short-term Prediction

“…Boosting methods have drawn lots of attention since it was proposed in 1990 [20,21,22], and keeps active in wide range of research fields and applications considering its excellent performance [23]. There is a popular believe that the ability of anti-overfitting is the key to interpreting its great performance and broad adaptability [24]. Gradient boosting estimates iteratively.…”