This work introduces a 2D motion prediction algorithm for lung tumors applied to dynamic MR images that combines intrafractional tumor deformations. It is developed in the context of MR-linac treatments and evaluates uniform and patient-specific margins about the gross tumor volume to optimize the tumor coverage. Seven early stage non-small cell lung cancer patients were imaged in treatment position with a 1.5T magnetic resonance using a balanced steady-state free precession sequence for one minute at a rate of four images per second and were instructed to breathe normally. The particle filter, in combination with the autoregressive model, is used to sequentially track and predict the tumor position 250ms in the future from the current image. In addition, the autocontour extracted from a previous study (Bourque et al2016 Med. Phys. 43 5161-9) is projected to the predicted position and various margins are evaluated. Averaged over all patients, the root mean square errors are (1.3 ± 0.5)mm and (2.0 ± 0.8)mm with and without prediction, respectively, and the difference in centroid position is (1.1 ± 0.4)mm with the prediction. The addition of the prediction algorithm leads to inferior errors for all cases. With such predictor, enlarging the propagated contour by a uniform 2mm margin leads to a minimum recall of 97% over the entire patient population. Considering patient-specific margins based on s 2 of the Gaussian distribution around the mean error, the margins vary from 0.8 to 3.0mm in anterior-posterior direction and from 1.2 to 3.2mm in the superior-inferior direction. This study concludes that the advantage of such prediction algorithm highly depends on the tumor motion characteristics. For both uniform and patient-specific margins evaluation, smaller treatment margins could be used when combined to an accurate tracking and motion prediction algorithm. These results could offer guidance for future MR-guided lung tumor treatments.
This work presents a proof of concept of a new autocontouring algorithm for NSCLC patients on dynamic MR images. The contours were generated in good agreement with the expert's contours.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.