MoDL-MUSSELS: Model-Based Deep Learning for Multishot Sensitivity-Encoded Diffusion MRI.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging
• Computational Imaging

Background:

• Phase errors in multishot diffusion-weighted echo-planar images degrade image quality.
• Existing correction methods like MUSSELS can be computationally intensive.

Purpose of the Study:

• Introduce MoDL-MUSSELS, a novel model-based deep learning architecture.
• Generalize the MUSSELS algorithm for improved efficiency.
• Reduce computational complexity while maintaining performance.

Main Methods:

• Developed a hybrid convolutional neural network (CNN) architecture.
• Integrated k-space and image-space CNNs for enhanced data processing.
• Replaced the computationally expensive filter bank in MUSSELS with a learned CNN.

Main Results:

• MoDL-MUSSELS achieves reconstructions comparable to state-of-the-art methods.
• Demonstrated several orders of magnitude reduction in run-time compared to existing algorithms.
• Validated the effectiveness of the hybrid CNN in exploiting annihilation relations and projecting data to an image manifold.

Conclusions:

• MoDL-MUSSELS provides an efficient and effective solution for phase error correction in diffusion-weighted MRI.
• The deep learning approach significantly accelerates image reconstruction.
• This method holds promise for improving clinical MRI workflows and data analysis.