Evaluating tracking and prediction of tumor motion in a motion-compensating system for adaptive radiotherapy

A mechanical counter-steering prototype which includes a mechanical slider system was developed to evaluate tracking and prediction algorithms used for tumor motion compensation. The compensation aims to maintain a stationary tumor with respect to the radiation beam. Motion tracking was performed with an optical flow algorithm. Three techniques of reference image registration were evaluated. To compensate for system latency, a neural network with a sliding window learning scheme was used to predict the tumor positon. A mean tracking (position) error of -0.05 ± 0.04 mm was obtained using initial image for reference registration. For our prediction module, an averaged mean error and mean absolute error (MAE) of 0.07±0.96 mm and 0.61 ± 0.80 were obtained for seven patients respectively. When incorporating tracking and prediction an averaged mean error of 0.30 ± 1.06 mm and a MAE of 0.8 ± 0.76 mm were observed. Results show that this image-guided prototype is suited for evaluating the tracking and prediction modules of a motion compensating radiotherapy system.