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

Ultra-low-field (ULF) MRI offers a promising path to accessible neuroimaging, with potential to address global healthcare disparities and advance population-level brain health research. However, the inherently low signal-to-noise ratio (SNR), reduced spatial resolution, and altered tissue contrasts relative to conventional high-field (HF) scans are significant barriers to ULF analysis and interpretation. While deep learning (DL) approaches have been proposed to enhance ULF image quality, many rely on synthetic training data due to the lack of available subject-matched ULF and HF scans, introducing potential "domain shift" errors when applied to real acquisitions. Here, we present a DL framework trained on real ULF and HF-MRIs to address these limitations and improve ULF-derived brain volume analysis.

A CycleGAN framework was developed for image-to-image translation across field strengths, while mitigating the need for large subject-matched ULF- and HF-MRIs. This approach enabled pretraining on large open-access MRI datasets followed by fine-tuning on real ULF scans. Model performance was evaluated through downstream brain volumetric analysis, assessing volumetric agreement with HF-derived measurements and test-retest reproducibility. We additionally explored a framework to reduce input acquisition requirements, improving scan protocol efficiency while preserving enhanced performance.

The proposed methods significantly improved hippocampal volumetric agreement and brain segmentation accuracy between ULF- and HF-MRI compared with existing strategies. Test-retest reproducibility for DL-enhanced images was superior to that of direct segmentation on ULF scans.

The proposed framework substantially improved the accuracy and reliability of ULF-derived brain volume measurements, particularly for subcortical structures such as the hippocampus.