Development and Application of a Deep Convolutional Neural Network Noise Reduction Algorithm for Diffusion-weighted Magnetic Resonance Imaging
Development and Application of a Deep Convolutional Neural Network Noise Reduction Algorithm for Diffusion-weighted Magnetic Resonance Imaging
复制标题
DOI：
10.4283/jmag.2019.24.2.223
复制
发表时间：
2019-06-01
影响因子：
0.5
通讯作者：
Lee, Youngjin
中科院分区：
文献类型：
--
作者：
Han, Dong-Kyoon;Kim, Kyuseok;Lee, Youngjin
弥散加权成像(DWI)在诊断医学领域中经常用于检测各种人类疾病。在离散余弦变换中，为了达到较高的疾病检测精度，噪声抑制是非常重要的。在这项研究中，我们开发了一种深度卷积神经网络(Deep-CNN)去噪算法，并通过在1.5-T和3.0-T磁共振系统上进行模拟和真实实验来评估其在DWI中的有效性。实验结果验证了所提出的用于DWI的Deep-CNN算法的有效性。与已有的非局部均值(NLM)算法相比，本文提出的Deep-CNN算法取得了更好的量化结果。实验结果表明，与已有的DWI算法相比，本文提出的Deep-CNN算法具有更高的降噪效率和图像可见性。
Diffusion-weighted imaging (DWI) is frequently used in the field of diagnostic medicine to detect various human diseases. In DWI, noise suppression is very important for achieving high detection accuracy of diseases. In this study, we develop a deep convolutional neural network (Deep-CNN) noise reduction algorithm and evaluate its effectiveness in DWI by performing both simulations and real experiments with a 1.5- and a 3.0-T MRI system. The results validate the proposed Deep-CNN algorithm for DWI. Compared with previously developed non-local means (NLM) algorithms, the proposed Deep-CNN algorithm achieves superior quantitative results. In conclusion, the quantitative results verify that the proposed Deep-CNN algorithm has higher noise reduction efficiency and image visibility than previously developed algorithms for DWI.