eduzhai > Applied Sciences > Engineering >

Motion Pyramid Networks for Accurate and Efficient Cardiac Motion Estimation

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Cardiac motion estimation plays a key role in MRI cardiac feature trackingand function assessment such as myocardium strain. In this paper, we proposeMotion Pyramid Networks, a novel deep learning-based approach for accurate andefficient cardiac motion estimation. We predict and fuse a pyramid of motionfields from multiple scales of feature representations to generate a morerefined motion field. We then use a novel cyclic teacher-student trainingstrategy to make the inference end-to-end and further improve the trackingperformance. Our teacher model provides more accurate motion estimation assupervision through progressive motion compensations. Our student model learnsfrom the teacher model to estimate motion in a single step while maintainingaccuracy. The teacher-student knowledge distillation is performed in a cyclicway for a further performance boost. Our proposed method outperforms a strongbaseline model on two public available clinical datasets significantly,evaluated by a variety of metrics and the inference time. New evaluationmetrics are also proposed to represent errors in a clinically meaningfulmanner.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×