Exploring the potential performance of 0.2 T low-field unshielded MRI scanner using deep learning techniques

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Low-field magnetic resonance imaging (MRI) systems are limited by physical constraints affecting image quality and speed.
• Conventional high-field MRI scanners are expensive and less portable, hindering widespread clinical adoption.

Purpose of the Study:

• To explore the potential of 0.2 T low-field unshielded MRI using deep learning to overcome hardware limitations.
• To achieve high imaging quality and speed comparable to high-field MRI systems.

Main Methods:

• Implemented active electromagnetic shielding and basic super-resolution for fast, high-quality unshielded imaging.
• Reduced acquisition time by decreasing the number of excitations for basic super-resolution.
• Analyzed cross-field super-resolution to map low-resolution images to high-resolution equivalents.
• Cascaded basic and cross-field super-resolution to enhance low-field image quality.

Main Results:

• Achieved image quality comparable to 1.5 T scanners (512x512 resolution, 0.45 mm^2 spatial resolution) on a 0.2 T unshielded system.
• Acquired single-orientation images in under 3.3 minutes.
• Demonstrated successful enhancement of low-field image quality to high-field levels.

Conclusions:

• The deep learning strategy effectively overcomes physical limitations of low-field unshielded MRI hardware.
• Rapid image acquisition with high-field-level quality is achievable on low-field systems.
• Findings support the transition towards portable, point-of-care MRI systems, advancing medical imaging technology.