From Low Field to High Value: Robust Cortical Mapping From Low-Field MRI.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Three-dimensional reconstruction of cortical surfaces from MRI is crucial for brain morphometric analysis.
• High-field MRI (HF-MRI) is standard but limited in availability.
• Low-field MRI (LF-MRI), especially portable systems, offers accessibility but faces challenges with existing analysis tools due to lower SNR and resolution.

Purpose of the Study:

• To develop and validate a machine learning method for 3D cortical surface reconstruction and analysis from portable LF-MRI scans.
• To create a tool that works "out of the box" without retraining for diverse LF-MRI contrasts and resolutions.

Main Methods:

• A 3D U-Net model trained on synthetic LF-MRI data to predict signed distance functions of cortical surfaces.
• Post-processing geometric steps to ensure topologically accurate reconstructions.
• Evaluation using paired HF-/LF-MRI scans from 65 subjects and validation on postmortem LF-MRI data.

Main Results:

• The method robustly recovers cortical surfaces across various LF-MRI acquisitions.
• A 3mm isotropic T2-weighted scan (under 4 min) showed strong agreement with HF-derived surfaces (surface area r=0.96, parcellations Dice=0.98, gray matter volume r=0.93).
• Cortical thickness estimation showed moderate correlation (r=0.70), sensitive to resolution and anisotropy.

Conclusions:

• The developed machine learning method significantly advances the feasibility of cortical surface analysis using portable LF-MRI systems.
• The tool demonstrates robustness across different LF-MRI sequences and contrasts, paving the way for wider accessibility of brain morphometry.
• Public availability of the tool facilitates broader research and clinical applications.