Time series prediction of respiratory motion for lung tumor tracking radiation therapy

A time series prediction problem is considered in this paper. In radiotherapy, the target motion often affects the conformability of the therapeutic dose distribution delivered to thoracic and abdominal tumors, and thus tumor motion monitoring systems have been developed. Even we can observe tumor motion accurately, however, radiotherapy systems may inherently have mechanical and computational delays to be compensated for synchronizing dose delivery with the motion. For solving the delay problem, we develop a novel system to predict complex time series of the lung tumor motion. An essential core of the system is an adaptive prediction modeling by which time-varying cyclic dynamics is transferred into time invariant one by a phase locking technique. After the transformation, some linear and nonlinear models including neural networks can be used for accurate time series prediction. Simulation studies demonstrate that the proposed system can achieve a clinically useful high accuracy and long-term prediction of the average error 1.59 ± 1.61 [mm] at 1 [sec] ahead prediction.