Technical - Emerging CMR Methods - Motion Correction & Rapid Imaging-Deep-Learning-Based Motion Correction For Myocardial Perfusion MRI

Deep-Learning-Based Motion Correction For Myocardial Perfusion MRI

Video Transcription

Video Summary

The speaker presents a deep learning method for motion correction in perfusion cardiac MRI (CMR). The approach uses a segmentation network to classify pixels into heart structures, then feeds the images and segmentation likelihoods into a registration network that predicts a deformation field. Training minimizes differences between segmentation likelihoods and adds smoothness regularization. Tested on short-axis images, the method achieved strong segmentation performance and outperformed the ENTS registration toolbox in accuracy and speed, even on contrast-mismatched cases. The method focuses on cardiac regions while leaving the rest of the body largely unchanged, supporting real-time automated perfusion CMR processing.

Keywords

deep learning

motion correction

perfusion cardiac MRI

segmentation network

image registration