当前位置: X-MOL 学术Appl. Magn. Reson. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Compressed Sensing MRI Reconstruction Using Generative Adversarial Network with Rician De-noising
Applied Magnetic Resonance ( IF 1.1 ) Pub Date : 2021-08-24 , DOI: 10.1007/s00723-021-01416-0
Mrinmoy Sandilya 1 , S R Nirmala 2 , Navajit Saikia 3
Affiliation  

Compressed sensing magnetic resonance imaging (CS-MRI) has been a widely studied field in biomedical signal processing owing to its success and practical MR machines integrated with the technology already rolling on in market. Since then, a decade-long research in the field has paved its way into deep-learning techniques being incorporated into the field to simplify some of the complex, theoretical issues that the original setup pose, without compromising efficiency. These machine-learning methods are particularly useful in context of input signal adaptability. The data acquisition process of MRI is noisy in nature with various types of noises associated, such as Rician noise, Gaussian Noise, and motion artifacts like breathing artifacts. However, the repercussions of noise has been scarcely incorporated into the study of most of these methods. In this context, training a model directly with the obtained data is somewhat inefficient. This paper proposes a Generative Adversarial Network (GAN)-based technique for obtaining a concise and more representative model which attempts to be a more robust noise immune network. Keeping in mind increased efficiency and reduced reconstruction time, the proposed method attempts to address the problem of MR image reconstruction. The results obtained from the proposed method has been compared with traditional CS-MRI and other contemporary methods using objective parameters such as PSNR, SSIM index, BRISQUE and FID score, and subjective parameters such as mean opinion score and LPIPS.



中文翻译:

使用生成对抗网络和 Rician 去噪的压缩感知 MRI 重建

压缩传感磁共振成像 (CS-MRI) 因其成功和实用的 MR 机器与市场上已经推出的技术相结合而成为生物医学信号处理中一个广泛研究的领域。从那时起,该领域长达十年的研究铺平了道路,将深度学习技术纳入该领域,以在不影响效率的情况下简化原始设置带来的一些复杂的理论问题。这些机器学习方法在输入信号适应性方面特别有用。MRI 的数据采集过程本质上是嘈杂的,伴随着各种类型的噪声,例如 Rician 噪声、高斯噪声和呼吸伪影等运动伪影。然而,噪声的影响几乎没有被纳入大多数这些方法的研究中。在这种情况下,直接用获得的数据训练模型效率有点低。本文提出了一种基于生成对抗网络 (GAN) 的技术,以获得一个简洁且更具代表性的模型,该模型试图成为一个更强大的噪声免疫网络。考虑到提高效率和减少重建时间,所提出的方法试图解决 MR 图像重建的问题。使用客观参数(如 PSNR、SSIM 指数、BRISQUE 和 FID 分数)和主观参数(如平均意见分数和 LPIPS),从所提出的方法获得的结果与传统的 CS-MRI 和其他当代方法进行了比较。本文提出了一种基于生成对抗网络 (GAN) 的技术,以获得一个简洁且更具代表性的模型,该模型试图成为一个更强大的噪声免疫网络。考虑到提高效率和减少重建时间,所提出的方法试图解决 MR 图像重建的问题。使用客观参数(如 PSNR、SSIM 指数、BRISQUE 和 FID 分数)和主观参数(如平均意见分数和 LPIPS),从所提出的方法获得的结果与传统的 CS-MRI 和其他当代方法进行了比较。本文提出了一种基于生成对抗网络 (GAN) 的技术,以获得一个简洁且更具代表性的模型,该模型试图成为一个更强大的噪声免疫网络。考虑到提高效率和减少重建时间,所提出的方法试图解决 MR 图像重建的问题。使用客观参数(如 PSNR、SSIM 指数、BRISQUE 和 FID 分数)和主观参数(如平均意见分数和 LPIPS),从所提出的方法获得的结果与传统的 CS-MRI 和其他当代方法进行了比较。

更新日期:2021-08-24
down
wechat
bug