Deep learning MRI super-resolution quality depends on feature loss layer selection

Researchers have explored how different layers in feature-based loss functions impact the quality of deep learning-based super-resolution for brain diffusion MRI. They found that using deeper layers in VGG16 networks introduced grid-like artifacts in the super-resolved images and diffusion parameters, while the shallowest layer produced results consistent with ground truth, even at a 9-fold resolution increase. The study highlights the critical need to carefully select contributing layers in feature-based losses to avoid artifacts and ensure accuracy in diffusion MRI applications. AI

IMPACT Highlights the importance of specific architectural choices in deep learning for medical imaging accuracy.

RANK_REASON Academic paper detailing a specific technical finding in deep learning for medical imaging. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Deep learning
• Human Connectome Project
• VGG16
• Brain Diffusion MRI

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →