Wave-Encoded Model-Based Deep Learning for Highly Accelerated Imaging with Joint Reconstruction.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Deep Learning in Medical Imaging
• Image Reconstruction

Background:

• Model-based deep learning (MoDL) integrates convolutional neural networks (CNNs) for efficient parallel imaging reconstruction.
• Wave-controlled aliasing in parallel imaging (CAIPI) accelerates acquisition by using sinusoidal gradients and 3D coil profiles.

Purpose of the Study:

• To introduce wave-encoded MoDL (wave-MoDL) for highly accelerated 3D imaging using wave-encoding and unrolled network constraints.
• To extend wave-MoDL for multicontrast reconstruction and rapid quantitative imaging (e.g., 3D-QALAS).

Main Methods:

• Combining wave-encoding strategy with unrolled network constraints for data consistency.
• Extending wave-MoDL for multicontrast reconstruction with CAIPI sampling.
• Applying wave-MoDL to interleaved Look-Locker acquisition sequence with T2 preparation pulse (3D-QALAS).

Main Results:

• Achieved a 40-second MPRAGE acquisition at 1 mm resolution with 16-fold acceleration.
• Enabled 1:50 min acquisition for T1, T2, and proton density mapping at 1 mm resolution with 12-fold acceleration.
• Demonstrated synthesis of contrast-weighted images from quantitative data.

Conclusions:

• Wave-MoDL facilitates rapid MRI acquisition and high-fidelity reconstruction.
• The method integrates unrolled neural networks into wave-CAIPI reconstruction.
• Potential to advance clinical and neuroscientific applications through accelerated and improved MR imaging.