SU-F-J-164: A Simulation Study On Anovel 4D MRI Reconstruction Method Based On Probability-Driven Sorting.

PURPOSE: To evaluate a novel probability-driven sorting based 4D MRI reconstruction method by comparing with the conventional 4D MRI reconstruction methods.

METHODS: We have proposed a novel probability-driven sorting based 4D MRI reconstruction method capable of retaining breathing variation information in breathing signals by generating multiple breathing cycles instead of a single cycle as the case with conventional 4D MRI reconstruction methods. It is expected that our proposed method can reduce motion artifacts caused by breathing variations, and can generate more accurate Average Intensity Projections (AIP). To evaluate the performance of the novel method, the probability-driven sorting based method, along with two conventional phase-based methods, for cine acquisition mode, and sequential acquisition mode, was applied on six patient breathing curves to simulate the 4D MRI reconstruction process using eXtended CArdiac Torso (XCAT) phantom. In the 4D images, a spherical target of 3 cm in diameter was generated in the liver and set to move with the liver. AIPs were generated for all three sets of 4D images. Ground-truth AIP was generated by averaging all the 3D volumes of the phantom. Motion artifacts were evaluated by the tumor volume consistency in 4D images and volume difference from the tumor volumes measured in 3D images. AIPs from the three sorting methods were evaluated and compared to the reference AIP in terms of intensity difference.

RESULTS: Probability-driven sorting has higher target volume consistency between phases for all the six patients, indicating the less motion artifacts. AIP from probability-driven sorting has the best agreement with the ground-truth AIP among the three reconstruction methods.

CONCLUSION: Based on the cases in this study, it has been shown that probability-driven sorting method is capable of generating multi-cycle 4D MRI images with reduced motion artifacts and improved AIP accuracy. 1R21CA165384.