Prediction of Lung Tumor Motion Based on Recurrent Neural Network

Abstract

This paper handles the prediction of the future lung tumor position which exhibits respiratory-induced motion inside the body of the patient. The prediction of the future tumor position is a necessary step for dynamic tumor tracking radiotherapy to compensate the positioning lag of the gantry of the clinical linear accelerator. This lag is known to be almost one second and the objective of the current study is accordingly to generate the one second future position estimate of the tumor with the tolerance of 1mm prediction error in the three-dimensional space. Several researches have been done to calculate predicted position of the tumor by establishing mathematical models. The current study focuses on the development of a mathematical model which provides one-second future position of the tumor using the recurrent neural network (RNN). The result of the numerical validation has been reported to justify the choice of the model structure and to show that the model can be trained to have the sufficient prediction accuracy necessary for use in real-time dynamic tumor tracking radiotherapy.