BTS-Net: Barlow twins-based superresolution for 7T human brain MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Image Analysis

Background:

• Standard 3T MRI has limitations in visualizing subtle brain structures.
• Super-resolution (SR) techniques aim to improve MRI image quality.
• Self-supervised learning (SSL) offers advanced feature representation for complex tasks.

Purpose of the Study:

• To develop and validate a Barlow Twins-based superresolution diffusion network (BTS-Net).
• To enhance 3T human brain MRI to 7T-like quality (BTS-7T) using SSL within a latent diffusion model (LDM).
• To evaluate the impact of BTS-Net on image quality, anatomical fidelity, and volumetric analysis.

Main Methods:

• Constructed a paired 3T-7T brain MRI database from 50 healthy adults.
• Employed Barlow Twins SSL within an LDM for SR from 3T to BTS-7T.
• Assessed image quality (PSNR, SSIM, NRMSE) and 3D structural fidelity in 14 brain regions via VBM.
• Validated BTS-Net on an external dataset of 10 healthy participants.

Main Results:

• BTS-7T MRI demonstrated superior image quality compared to 3T MRI across all metrics in both datasets.
• BTS-7T images showed improved anatomical fidelity and comparable volumetry to ground truth.
• Higher agreement in regional brain volumes was observed with BTS-7T, particularly in hippocampus, putamen, and amygdala.

Conclusions:

• BTS-Net effectively enhances 3T brain MRI to 7T-like resolution, improving both qualitative and quantitative analyses.
• The developed network shows potential for detecting subtle morphological changes in early neurodegenerative conditions.
• Further validation with larger patient cohorts is recommended for clinical adoption.