Self-supervised learning for denoising of multidimensional MRI data

Authors
Kang, BeomguLee, WonilSeo, HyunseokHeo, Hye-YoungPark, Hyunwook
Issue Date
2024-11
Publisher
John Wiley & Sons Inc.
Citation
Magnetic Resonance in Medicine, v.92, no.5, pp.1980 - 1994
Abstract
PurposeTo develop a fast denoising framework for high-dimensional MRI data based on a self-supervised learning scheme, which does not require ground truth clean image.Theory and MethodsQuantitative MRI faces limitations in SNR, because the variation of signal amplitude in a large set of images is the key mechanism for quantification. In addition, the complex non-linear signal models make the fitting process vulnerable to noise. To address these issues, we propose a fast deep-learning framework for denoising, which efficiently exploits the redundancy in multidimensional MRI data. A self-supervised model was designed to use only noisy images for training, bypassing the challenge of clean data paucity in clinical practice. For validation, we used two different datasets of simulated magnetization transfer contrast MR fingerprinting (MTC-MRF) dataset and in vivo DWI image dataset to show the generalizability.ResultsThe proposed method drastically improved denoising performance in the presence of mild-to-severe noise regardless of noise distributions compared to previous methods of the BM3D, tMPPCA, and Patch2self. The improvements were even pronounced in the following quantification results from the denoised images.ConclusionThe proposed MD-S2S (Multidimensional-Self2Self) denoising technique could be further applied to various multi-dimensional MRI data and improve the quantification accuracy of tissue parameter maps.
Keywords
DIFFUSION TENSOR MRI; NOISE; ARTIFACTS; magnetization transfer contrast (MTC); quantitative MRI; self-supervised learning; denoising; diffusion
ISSN
0740-3194
URI
https://pubs.kist.re.kr/handle/201004/150203
DOI
10.1002/mrm.30197
Appears in Collections:
KIST Article > 2024
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML

qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE