Cine-MRI Simulation to Evaluate Tumor Tracking.

Conventional evaluations of tumor tracking algorithms require inter-observer segmentations from radiation oncologists on the Cine-MRI (2D sagittal MR video). Instead of performing intensive manual annotations on images, we present a 2D video simulator that uses the pre-treatment images, including a breathing model, that generates Cine-MR images in parallel with the underlined segmentation of the tumor. We include the data of seven patients within a retrospective clinical study that received stereotactic body radiation therapy for liver metastases. Each patient has a pre-treatment 4DCT scan, a pre-treatment 3D MR with tumor and liver delineations, and the treatment Cine-MRI. We augment the data with the simulator by changing breathing motion parameters and adding noise. The simulator generates a total of 84 Cine-MRI sequences, thus having 12 videos per patient. We validate the simulated versus the real Cine-MRI in terms of tumor motion. Finally, we used the simulator to evaluate the performance of real-time tumor tracking algorithms with this dataset.