Anisotropic neural deblurring for MRI acceleration.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Magnetic Resonance Imaging (MRI) is essential for brain imaging, offering superior soft tissue contrast and detailed anatomical, functional, and neurochemical information.
• Current MRI scan durations, often exceeding 30 minutes, negatively impact patient experience, appointment availability, and operational costs.
• Reducing MRI scan times is critical for improving healthcare efficiency and patient satisfaction.

Purpose of the Study:

• To investigate an enhancement method for low-resolution (LR) brain MRI scans.
• To enable significantly shorter MRI acquisition times without compromising image diagnostic value.
• To develop a technique for accelerating MRI scans through image quality enhancement.

Main Methods:

• A novel fully convolutional neural network was developed for image enhancement.
• The network is optimized for accelerated LR MRI acquisitions by reducing the matrix size in the phase encoding direction.
• The model is trained end-to-end to convert LR images into high-resolution (HR) counterparts, using real LR acquisitions for training.

Main Results:

• The proposed method achieved a 4x acceleration factor, demonstrating favorable comparisons against state-of-the-art algorithms like DeblurGAN and DAGAN.
• Quantitative validation using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) scores confirmed the method's effectiveness.
• Qualitative assessment by four senior neuroradiologists corroborated the enhanced image quality and diagnostic utility.

Conclusions:

• The developed method shows potential as a valuable tool for reducing brain MRI scan times.
• Further validation is recommended with larger datasets, diverse imaging protocols, and multi-vendor MRI machines.
• This approach could significantly improve the efficiency and accessibility of brain MRI diagnostics.