Parallel imaging and convolutional neural network combined fast MR image reconstruction: Applications in low-latency...

Area of Science:

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

Background:

• Accelerated real-time MRI is crucial for dynamic imaging but often compromises image quality.
• Existing reconstruction methods like GRAPPA and L1-ESPIRiT have limitations in speed and artifact reduction.

Purpose of the Study:

• To develop and evaluate a novel framework integrating parallel imaging (PI) with convolutional neural networks (CNNs) for accelerated real-time MRI.
• To achieve both low-latency and high-quality image reconstruction.

Main Methods:

• A CNN was designed where PI reconstruction steps were integrated with convolutional layers.
• The network parameters were learned offline and applied to unseen data.
• Evaluated on real-time cardiac (1.5T) and abdominal (0.35T) imaging with retrospective and prospective undersampling.
• Compared against GRAPPA and L1-ESPIRiT methods.

Main Results:

• The PI-CNN framework successfully reconstructed images with reduced noise and aliasing artifacts compared to conventional methods.
• Achieved frame reconstruction times under 100 ms, demonstrating clinical feasibility.
• Outperformed single-coil, GRAPPA, and L1-ESPIRiT reconstructions in both cardiac and abdominal imaging.

Conclusions:

• The combined PI-CNN framework offers a promising solution for low-latency, high-quality real-time MRI.
• This deep learning approach enhances accelerated MRI acquisition and reconstruction.
• Potential for improved clinical workflow and diagnostic capabilities in real-time MRI applications.