Advancing the Volumetric Analysis of Ultra-Low-Field Brain MRI Using Image-to-Image Translation

Area of Science:

• Neuroimaging
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Ultra-low-field (ULF) MRI promises accessible neuroimaging but suffers from low signal-to-noise ratio (SNR) and resolution.
• Deep learning (DL) for ULF enhancement often uses synthetic data, leading to domain shift issues.
• Accurate brain volume analysis is crucial for population health research and clinical applications.

Purpose of the Study:

• To develop a DL framework for enhancing ULF MRI quality using real subject-matched ULF and high-field (HF) MRI data.
• To improve the accuracy and reliability of ULF-derived brain volume measurements, particularly for subcortical structures.
• To mitigate domain shift errors common in DL models trained on synthetic data.

Main Methods:

• A CycleGAN framework was employed for image-to-image translation between ULF and HF MRI.
• The model was pretrained on large open-access datasets and fine-tuned on real ULF scans.
• Downstream volumetric analysis assessed agreement with HF measurements and test-retest reproducibility.

Main Results:

• The DL framework significantly improved hippocampal volumetric agreement and brain segmentation accuracy compared to existing methods.
• Test-retest reproducibility of DL-enhanced ULF images surpassed direct segmentation on ULF scans.
• The framework demonstrated improved accuracy and reliability in ULF-derived brain volume measurements.

Conclusions:

• The developed DL framework substantially enhances the accuracy and reliability of ULF-derived brain volume measurements.
• This approach offers a viable solution for improving ULF MRI analysis, particularly for subcortical structures like the hippocampus.
• The framework holds potential for advancing accessible neuroimaging and population-level brain health research.