Model-Based Deep Learning for Reconstruction of Joint k-q Under-sampled High Resolution Diffusion MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Diffusion MRI is crucial for understanding tissue microstructure.
• Current methods face limitations in speed and resolution due to undersampling.
• Accelerated acquisition is essential for clinical utility.

Purpose of the Study:

• To develop a model-based deep learning architecture for highly accelerated diffusion MRI reconstruction.
• To enable high-resolution imaging with significantly reduced scan times.
• To introduce a novel deep learning approach for joint reconstruction of diffusion-weighted images.

Main Methods:

• A model-based deep learning architecture was proposed.
• A pre-trained denoiser, based on a multi-compartmental tissue microstructure model, was used as a regularizer.
• An autoencoder was trained unsupervised to learn the diffusion MRI signal subspace.
• Joint reconstruction from k-q undersampled acquisition in a parallel MRI setting was performed.

Main Results:

• The autoencoder provided a strong denoising prior for recovering q-space signals.
• The method demonstrated high acceleration capabilities on simulated brain data.
• High-resolution diffusion MRI reconstruction was achieved from highly undersampled data.

Conclusions:

• The proposed method enables high-resolution diffusion MRI reconstruction with high acceleration.
• The novel use of a pre-trained denoiser significantly improves the recovery of undersampled diffusion MRI data.
• This approach holds promise for advancing diffusion MRI applications in clinical settings.