Generative AI Model Denoises Short 7T Brain MRI Scans to Match Longer-Acquisition Quality

Generative AI denoising of short 7T brain MRI improved MSE by 31.7% and SSIM by 7.5% versus unprocessed images (p<0.001), achieving quality comparable to longer scans.

• The 7T Conditional Diffusion Model outperformed convolutional neural network, vision transformer, and generative adversarial network in all image quality metrics.
• Expert raters rated denoised images as having significantly enhanced detail and contrast preservation in the hippocampi, white matter lesions, and small cortical veins (p<0.001).

Automated summary

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