Construction of a model for the prediction of lung tumor motion for radiotherapy

Abstract

This paper discloses the development of a recursive mathematical model for radiotherapy which predicts future position of the lung tumor based on the current and past tumor position measurements. Four adaptation algorithms have been applied to train the model coefficients so as to achieve 1[mm] prediction error in 1[s] ahead prediction. Numerical evaluation has been performed for 9 different lung tumor trajectories and the models are successfully trained to achieve the required precision for all dataset. Analyzing the results indicate that SHARF algorithm gives the smallest prediction error for relatively wide-bandwidth lung motion which might be partly affected by the heartbeat, whereas SHSER algorithm fits the best for tumor motions mainly dominated by the respiration of the patients.