Eugene Joo ’27

WUJUR (2025) | Cite this work

Abstract

Accelerated MRI scanning often results in lower resolution, introducing image distortions and loss of critical details. This study explores the application of machine learning to super-resolve Fluid-Attenuated Inversion Recovery Magnetic Resonance Imaging (FLAIR MRI) scans and restore those missing details. Specifically, we test models based on the U-Net architecture trained with two down sampling methods: removing two out of every 3 slices and linearly interpolating in image space and cropping the top and bottom thirds slices in k-space, the raw frequency data. Our experiments demonstrate that k-space down sampling consistently outperforms image-space methods in both Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) metrics, while also achieving superior reconstruction of fine details. Additionally, a multi-slice input approach using three adjacent slices was shown to further improve results by providing additional spatial context. This work allows for the restoration of MRI scans collected in an accelerated manner, significantly enhancing image quality and detail for improved diagnostic confidence.

Main

This article is viewable by PDF download.

References

References for this work are viewable by PDF download.

Supplemental material

There is no supplemental material provided by the author(s) for this work.

About this article

Citation (APA)

Joo, E. (2025). Restoring resolution in accelerated FLAIR MRI via U-Net super resolution. WUJUR, 2(1), 16-19.

A DOI link will soon be supplied for this article and added to the above citation.

Author Information

Eugene, Joo. Washington University in St. Louis.
Corresponding Author. Send correspondence to e.joo@wustl.edu.

Additional Information and Declarations

There is no additional information or declarations provided by the author(s).

Peer Review

WUJUR thanks the anonymous Peer Reviewers who contributed to the review of this work.

Rights and Permissions

Individual articles © 2025 by their respective authors. Illustrations and images remain the property of their respective authors or rights holders and are reproduced here with permission or under fair use.

The author(s) may grant certain license permissions to the public.

Copyright © 2025. Washington University Journal of Undergraduate Research. All rights reserved for compilation, design, and editorial content.