Related Experiment Video
Updated: Dec 29, 2025
08:48
Neuronavigation-guided Repetitive Transcranial Magnetic Stimulation for Aphasia
Published on: May 6, 2016
12.7K
Reconstruction of undersampled 3D non-Cartesian image-based navigators for coronary MRA using an unrolled deep
Mario O Malavé1, Corey A Baron2, Srivathsan P Koundinyan1
1Magnetic Resonance Systems Research Laboratory, Department of Electrical Engineering, Stanford University, Stanford, CA.
Magnetic Resonance in Medicine
|February 4, 2020
Summary
A novel deep learning model rapidly reconstructs 3D image-based navigators (iNAVs) for coronary magnetic resonance angiography (CMRA). This accelerates motion correction while maintaining accuracy, improving diagnostic imaging.
Area of Science:
- Medical Imaging
- Deep Learning
- Cardiovascular Imaging
Background:
- Coronary Magnetic Resonance Angiography (CMRA) requires accurate motion correction for high-quality imaging.
- Undersampled 3D non-Cartesian image-based navigators (iNAVs) are crucial for capturing cardiac motion but require efficient reconstruction.
Purpose of the Study:
- To develop and validate an unrolled deep learning (DL) model for rapid reconstruction of undersampled 3D non-Cartesian iNAVs.
- To enable accurate nonrigid motion correction in CMRA using DL-based iNAV reconstruction.
Main Methods:
- An end-to-end unrolled DL network was trained to reconstruct 3D iNAVs using a variable-density cones trajectory.
- The model incorporated a nonuniform Fast Fourier Transform (FFT) operator and a CNN-based regularization term.
- Reconstruction accuracy was validated against -ESPIRiT, comparing motion estimates and final CMRA image quality.
Main Results:
- DL-based iNAV reconstruction achieved comparable motion estimates and coronary image quality to -ESPIRiT.
- The unrolled network demonstrated significant speed increases (20x over CPU, 3x over GPU -ESPIRiT).
Conclusions:
- A DL-based approach effectively reconstructs undersampled 3D non-Cartesian iNAVs.
- This method accelerates iNAV reconstruction for CMRA, preserving motion correction accuracy and improving workflow efficiency.
