Technical - Emerging CMR Methods - Motion Correction & Rapid Imaging-Deep Learning Reconstruction of Real-time Cine MRI Acquired During Exercise Stress

Deep Learning Reconstruction of Real-time Cine MRI Acquired During Exercise Stress

Video Transcription

Video Summary

This presentation described using deep learning to reconstruct real-time cardiac MRI during exercise stress. The goal was to enable free-breathing, ungated imaging with enough speed and quality for cardiac function assessment. A 3D U-Net was trained on simulated radial undersampled data generated from segmented, EKG-gated rest MRI from 503 patients. It was then tested on a healthy volunteer scanned at rest and during supine bicycle exercise. The model reconstructed images quickly, about one second per slice, with good image quality and visible physiologic changes. Further studies are needed to confirm diagnostic accuracy for detecting wall motion abnormalities.

Keywords

deep learning

cardiac MRI

exercise stress imaging

3D U-Net

real-time reconstruction