Rapid 2D 23 Na MRI of the calf using a denoising convolutional neural network.

Autor: Baker RR; UCL Centre for Medical Imaging, University College London, London, UK; UCL Centre for Translational Cardiovascular Imaging, University College London, London, UK. Electronic address: r.baker.17@ucl.ac.uk., Muthurangu V; UCL Centre for Translational Cardiovascular Imaging, University College London, London, UK. Electronic address: v.muthurangu@ucl.ac.uk., Rega M; Institute of Nuclear Medicine, University College Hospital, London, UK. Electronic address: marilena.rega@nhs.net., Walsh SB; Department of Renal Medicine, University College London, London, UK. Electronic address: stephen.walsh@ucl.ac.uk., Steeden JA; UCL Centre for Translational Cardiovascular Imaging, University College London, London, UK. Electronic address: jennifer.steeden@ucl.ac.uk.
Jazyk: angličtina
Zdroj: Magnetic resonance imaging [Magn Reson Imaging] 2024 Jul; Vol. 110, pp. 184-194. Date of Electronic Publication: 2024 Apr 19.
DOI: 10.1016/j.mri.2024.04.027
Abstrakt: Purpose: 23 Na MRI can be used to quantify in-vivo tissue sodium concentration (TSC), but the inherently low 23 Na signal leads to long scan times and/or noisy or low-resolution images. Reconstruction algorithms such as compressed sensing (CS) have been proposed to mitigate low signal-to-noise ratio (SNR); although, these can result in unnatural images, suboptimal denoising and long processing times. Recently, machine learning has been increasingly used to denoise 1 H MRI acquisitions; however, this approach typically requires large volumes of high-quality training data, which is not readily available for 23 Na MRI. Here, we propose using 1 H data to train a denoising convolutional neural network (CNN), which we subsequently demonstrate on prospective 23 Na images of the calf.
Methods: 1893 1 H fat-saturated transverse slices of the knee from the open-source fastMRI dataset were used to train denoising CNNs for different levels of noise. Synthetic low SNR images were generated by adding gaussian noise to the high-quality 1 H k-space data before reconstruction to create paired training data. For prospective testing, 23 Na images of the calf were acquired in 10 healthy volunteers with a total of 150 averages over ten minutes, which were used as a reference throughout the study. From this data, images with fewer averages were retrospectively reconstructed using a non-uniform fast Fourier transform (NUFFT) as well as CS, with the NUFFT images subsequently denoised using the trained CNN.
Results: CNNs were successfully applied to 23 Na images reconstructed with 50, 40 and 30 averages. Muscle and skin apparent TSC quantification from CNN-denoised images were equivalent to those from CS images, with <0.9 mM bias compared to reference values. Estimated SNR was significantly higher in CNN-denoised images compared to NUFFT, CS and reference images. Quantitative edge sharpness was equivalent for all images. For subjective image quality ranking, CNN-denoised images ranked equally best with reference images and significantly better than NUFFT and CS images.
Conclusion: Denoising CNNs trained on 1 H data can be successfully applied to 23 Na images of the calf; thus, allowing scan time to be reduced from ten minutes to two minutes with little impact on image quality or apparent TSC quantification accuracy.
Competing Interests: Declaration of competing interest None.
(Copyright © 2024 The Authors. Published by Elsevier Inc. All rights reserved.)
Databáze: MEDLINE
Externí odkaz: