AI Superresolution: Converting T1-weighted MRI from 3T to 7T resolution toward enhanced imaging biomarkers for Alzheimer’s disease

Sep 30, 2026

Uploaded by Asher

High-resolution 7T MRI enables in vivo visualization of fine anatomical structures selectively affected in Alzheimer’s disease, but 7T acquisition remains difficult and scarce in clinical practice. Deep learning is used to synthesize 7T-resolution T1-weighted MRI from lower-resolution 3T scans. Paired 7T/3T data from 178 BioFINDER-2 participants trains a U-Net and a U-Net-GAN, with performance assessed via image metrics and blinded expert ratings.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages3

Reading Time8 min

AccessFree to download

Abstract

Background

Method

Result

Conclusion

DOI: 10. 1002/alz70856_106600 BIOMARKERS POSTER PRESENTATION NEUROIMAGING AI Superresolution: Converting T1-weighted MRI from 3T to 7T resolution toward enhanced imaging biomarkers for Alzheimer’s disease Malo Gicquel 1 Nicola Spotorno3 Laura E. M. Wisse4 Gabrielle Flood2 Ruoyi Zhao 1 Anika Wuestefeld3 Olof Strandberg3 Yu Xiao 1 Kalle Åström2 Danielle van Westen4 David Berron5, 6 Oskar Hansson3 Jacob W. Vogel1 1 Department of Clinical Sciences Malmö, SciLifeLab, Lund University, Lund, Sweden 2 Centre for Mathematical Sciences, Lund University, Lund, Sweden 3 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Lund University, Lund, Sweden 4 Diagnostic Radiology, Department of Clinical Sciences Lund, Lund University, Lund, Sweden 5 Center for Behavioral Brain Sciences (CBBS), Otto-von-Guericke University Magdeburg, Magdeburg, Sachsen Anhalt, Germany 6 German Center for Neurodegenerative Diseases (DZNE), Magdeburg, Germany Correspondence Jacob W. Vogel, Department of Clinical Sciences Malmö, SciLifeLab, Lund University, Lund, Sweden. [Email:](Email: jacob.vogel@med.lu.se)[ jacob.vogel@med.lu.se](Email: jacob.vogel@med.lu.se) Abstract Background: High-resolution (7T) MRI facilitates in vivo imaging of fine anatomical structures selectively affected in Alzheimer’s disease (AD), including medial temporal lobe subregions. However, 7T data is challenging to acquire and largely unavailable in clinical settings. Here, we use deep learning to synthesize 7T resolution T1-weighted MRI images from lower-resolution (3T) images. Method: Paired 7T and 3T T1-weighted images were acquired from 178 participants (134 clinically unimpaired, 48 impaired) from the Swedish BioFINDER-2 study. To synthesize 7T-resolution images from 3T images, we trained two models: a specialized U-Net, and a U-Net mixed with a generative adversarial network (U-Net-GAN) on 80% of the data. We evaluated model performance on the remaining 20%, compared to models from the literature (V-Net, WATNet), using image-based performance metricsand by surveying five blinded MRI professionals based on subjective quality. For n = 11 participants, amygdalae were automatically segmented with FastSurfer on 3T and synthetic-7T images, and compared to a manually segmented “ground truth”. To assess downstream performance, FastSurfer was run on n = 3,168 triplets of matched 3T and AI-generated synthetic-7T images, and a multi-class random forest model classifying clinical diagnosis was trained on both datasets. Result: Synthetic-7T images were generated for images in the test set (Figure 1A). Image metrics suggested the U-Net as the top performing model (Figure 1B), though blinded experts qualitatively rated the GAN-U-Net as the best looking images, exceeding even real 7T images (Figure 1C). Automated segmentations of amygdalae from the synthetic GAN-U-Net model were more similar to manually segmented This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Alzheimer’s Association. Alzheimer’s & Dementia published by Wiley Periodicals LLC on behalf of Alzheimer’s Association. Alzheimer’s Dement. 2025;21(Suppl. 2):e106600. [https://doi.org/10.1002/alz70856_106600](https://doi.org/10.1002/alz70856_106600) wi[leyonlinelibrary.com/journal/alz](leyonlinelibrary.com/journal/alz) 1of3 BIOMARKERS amygdalae, compared to the original 3T they were synthesized from, in 9/11 images (Figure 2). Classification obtained modest performance (accuracy∼60%) but did not differ across real or synthetic images (Figure 3A). Synthetic image models used slightly different features for classification (Figure 3B). Conclusion: Synthetic T1-weighted images approaching 7T resolution can be generated from 3T images, which may improve image quality and segmentation, without compr

FAQs

Why is 7T MRI important but hard to use clinically in Alzheimer’s disease research?

7T MRI provides finer anatomical detail relevant to Alzheimer’s disease, including medial temporal lobe subregions. However, 7T data acquisition is challenging and largely unavailable in clinical settings.

How were 7T-resolution images synthesized from 3T in this work?

Paired 7T and 3T T1-weighted images from 178 participants were used to train two deep learning models: a specialized U-Net and a U-Net combined with a generative adversarial network (U-Net-GAN).

What evaluation outcomes were reported for the synthesized images?

Image metrics indicated the U-Net as top performing, while blinded experts rated the GAN-U-Net as producing the best-looking images, even better than real 7T. Amygdala segmentations from the synthetic GAN-U-Net were more similar to manual ground truth in 9/11 cases, and downstream classification showed modest accuracy (~60%) without differing between real and synthetic images.

Sep 30, 2026

Uploaded by Asher