RAVEN: Robust, generalizable, multi-resolution structural MRI upsampling using Autoencoders.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Image Analysis

Background:

• Magnetic Resonance Images (MRIs) offer high inter-tissue contrast, revealing neuroanatomical changes in aging and disease.
• Standard MRI resolution limits detection of subtle, early-stage pathological changes.
• Increasing MRI acquisition resolution presents challenges like noise, time, cost, and patient discomfort.

Purpose of the Study:

• To develop a robust and generalizable single-image super-resolution network for brain MRIs.
• To overcome the limitations of standard MRI resolution for detecting subtle neuroanatomical alterations.
• To introduce Resolution Augmentation with Variational auto-Encoder Networks (RAVEN) using generative adversarial networks (GANs).

Main Methods:

• Developed RAVEN, a single-image super-resolution network integrating Variational Auto-Encoders (VAEs) and Generative Adversarial Networks (GANs).
• Applied RAVEN to upsample in-vivo and ex-vivo MRIs across diverse modalities (T1w, T2w, T2*) and field strengths (3T-7T).
• Targeted achieving isotropic voxel sizes as small as 0.5mm with arbitrary upsampling factors.

Main Results:

• RAVEN demonstrated state-of-the-art performance in upsampling brain MRIs.
• The network effectively preserved true anatomical information compared to existing methods.
• RAVEN achieved high-resolution targets (e.g., 0.5mm isotropic) across various MRI types and field strengths.

Conclusions:

• RAVEN offers a powerful solution for enhancing brain MRI resolution without increasing acquisition time or cost.
• The method shows significant potential for improving the early detection of neurodegenerative diseases and aging-related brain changes.
• RAVEN is open-access, providing valuable tools for the neuroimaging research community.