MR SIGNATURE MATCHING (MRSIGMA) WITH ADAPTIVE MOTION LEARNING FOR ROBUST REAL-TIME 4D MRI ON A 1.5T MR-LINAC

Motivation: Real-time 4D MRI on the MR-Linac is still sensitive to respiratory motion baseline drifts. Goal(s): To further develop MRSIGMA for real-time adaptation to out-of-range anatomical changes during signature matching. Approach: The motion dictionary was continuously updated using a sliding window of stack-of-stars data and fast motion-resolved Movienet reconstruction. MRSIGMA with adaptive motion learning was implemented in real-time using an external computer connected to the MR-Linac and tested on a programmable 4D phantom and a patient with pancreatic cancer. Results: The adaptive motion learning approach was able to update the dictionary after 2 sliding window periods, which improved robustness of real-time volumetric motion tracking. Impact: MRSIGMA with adaptive learning would enable real-time volumetric motion tracking robust to respiratory motion baseline drifts and other anatomical changes on the MR-Linac for improved monitoring and adaptation of radiotherapy of tumors affected by respiratory motion.