Deep Learning Translation Boosts Ultra-Low-Field Brain MRI Volumetric Accuracy

A CycleGAN model trained on real ultra-low-field and high-field MRI improved hippocampal volume agreement and brain segmentation accuracy, with better test-retest reproducibility than direct ULF segmentation.

• The model was pretrained on large open-access datasets and fine-tuned on real ULF scans, reducing domain shift from synthetic training.
• A framework to reduce input acquisition requirements also improved scan protocol efficiency while maintaining enhanced performance.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.