Back to all papers

Accelerating 7T Gradient-Echo Brain MRI With Generative AI Denoising.

September 26, 2026pubmed logopapers

Authors

Wang Y,Li B,Liang Y,Carlson M,DiGiacomo P,Moein Taghavi H,Maclaren J,Bell M,Pham N,Decker JH,Lv M,Liang T,Wong J,Mormino E,Henderson VW,Zaharchuk G,Rutt B,Shao W,Georgiadis M,Zeineh M

Affiliations (6)

  • Department of Bioengineering, Stanford University, Stanford, California.
  • Department of Electrical Engineering, Stanford University, Stanford, California.
  • Department of Radiology, Stanford School of Medicine, Stanford, California.
  • Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, California.
  • Department of Electrical & Computer Engineering, University of Florida, Florida.
  • Department of Radiology, Stanford School of Medicine, Stanford, California. Electronic address: [email protected].

Abstract

This study aimed to efficiently denoise short 7 T MRI acquisitions to achieve the image quality of longer scans using a generative Artificial Intelligence (AI) model. A 7T Conditional Diffusion Model (7TCDM) was trained on an in-house 7T dataset of 11 examinations consisting of multi-repetition 2D T2-weighted gradient-echo acquisitions. The model utilized native single-acquisition 2D reconstructions, using multi-repetition images as a reference to guide denoising and enhance signal-to-noise ratio and contrast. Performance was compared to the same single-acquisition reconstruction either unprocessed or enhanced by a similarly trained convolutional neural network, vision transformer, and generative adversarial network, using Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and comprehensive neuroradiologic ratings. 7TCDM was tested on 2D T2-weighted gradient-echo images from 19 participants: eight healthy controls, six individuals with mild cognitive impairment, and five with Alzheimer's disease. Referencing the multi-repetition reference standard, 7TCDM improved the single-acquisition original image by 31.7% in MSE, 4.9% in PSNR, and 7.5% in SSIM, outperforming all other models in all metrics (P < 0.001). Expert rater evaluations confirmed superior image quality, with significantly enhanced detail (P < 0.001) and contrast preservation (P < 0.001) for the hippocampi, for white matter lesions, and for small cortical veins. Generative AI denoising provides high-quality denoised images from shorter scans, increasing the feasibility of scanning patients in shorter times while preserving essential anatomical and pathological details.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.