Accelerating High b-Value Diffusion-Weighted MRI Using a Convolutional Recurrent Neural Network (CRNN-DWI).

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Diffusion-weighted imaging (DWI) is crucial for non-invasive tissue characterization.
• High undersampling rates in DWI reduce scan time but compromise image quality.
• Reconstruction of undersampled DWI data remains a significant challenge in MRI.

Purpose of the Study:

• To develop and validate a novel convolutional recurrent neural network (CRNN-DWI).
• To apply CRNN-DWI for reconstructing highly undersampled multi-b-value, multi-direction DWI datasets.
• To assess the performance of CRNN-DWI against conventional reconstruction methods.

Main Methods:

• A deep neural network combining CNN and RNN architectures was developed.
• The CRNN-DWI model processed diffusion images at undersampling rates of R=4 and R=6.
• Reconstructed images and diffusion parameter maps were quantitatively assessed using SSIM and PSNR metrics.

Main Results:

• CRNN-DWI significantly outperformed zero-padding and 3D-CNN in reconstructing DWI images and diffusion parameter maps.
• Higher average SSIM and PSNR values were achieved by CRNN-DWI at both R=4 and R=6.
• Diffusion parameter maps from CRNN-DWI demonstrated superior SSIM values compared to other methods.

Conclusions:

• CRNN-DWI presents a viable and effective method for reconstructing highly undersampled DWI data.
• This deep learning approach has the potential to reduce data acquisition burden in MRI.
• CRNN-DWI offers improved image quality and parameter estimation for advanced diffusion MRI applications.