Real-time tumor tracking with an artificial neural networks-based method: a feasibility study

Abstract

The purpose of this study was to develop and assess the performance of a tumor tracking method designed for application in radiation therapy. This motion compensation strategy is currently applied clinically only in conventional photon radiotherapy but not in particle therapy, as greater accuracy in dose delivery is required. We proposed a tracking method that exploits artificial neural networks to estimate the internal tumor trajectory as a function of external surrogate signals. The developed algorithm was tested by means of a retrospective clinical data analysis in 20 patients, who were treated with state of the art infra-red motion tracking for photon radiotherapy, which is used as a benchmark. Integration into a hardware platform for motion tracking in particle therapy was performed and then tested on a moving phantom, specifically developed for this purpose. Clinical data show that a median tracking error reduction up to 0.7 mm can be achieved with respect to state of the art technologies. The phantom study demonstrates that a real-time tumor position estimation is feasible when the external signals are acquired at 60 Hz. The results of this work show that neural networks can be considered a valuable tool for the implementation of high accuracy real-time tumor tracking methodologies.