深度学习在化学交换饱和转移磁共振成像中的研究进展

Research progress of deep learning in chemical exchange saturation transfer magnetic resonance imaging

化学交换饱和转移(chemical exchange saturation transfer, CEST)是一种新型的MRI技术,其基本原理是通过水信号的减少来间接实现对特定低浓度溶质分子的检测。采集速度慢、量化速度慢、量化评估不准确等问题影响着CEST MRI在临床中的应用推广,如何改善这些问题也成为研究的重点。深度学习作为人工智能的一种新的研究方向,近几年才应用于CEST MRI技术。本文在广泛调研国内外文献的基础上,对深度学习在临床CEST MRI上应用进行了深入分析与梳理。其中,在量化方面,一方面介绍了通过给深度神经网络(deep neural network, DNN)中输入临床中采集3 T的Z谱数据,预测出高场的CEST参数,进而得到比较明显的CEST信号;另一方面介绍了DNN结合磁化转移指纹识别(magnetization transfer fingerprinting, MTF)技术的方法改善传统量化方法中拟合参数精度低和拟合效率低的问题;在加速方面,一方面介绍深度学习用于CEST MRI加速采集;另一方面介绍了深度学习用于改善传统多池洛伦兹拟合量化速度慢的问题。供对本领域感兴趣者参考及在此基础上进一步地研究开发,加速CEST MRI的临床转换。

Deep learning, a significant method of artificial intelligence, has been used for chemical exchange saturation transfer magnetic resonance imaging(CEST MRI) in recent years, the basic principle is to indirectly realize the detection of specific low concentration of solute molecules through the reduction of water signal. Problems such as slow collection speed, slow quantification speed, and inaccurate quantitative evaluation affect the application and promotion of CEST MRI in clinical practice, and how to improve these problems has also become the focus of research. As a new research direction of artificial intelligence, deep learning has only been applied to CEST-MRI technology in recent years. This method is mainly used in the quantification and acceleration aspects of CEST MRI. The quantification usage includes prediction of the high field results and quantify proton exchange rate and concentration. The acceleration studies include acceleration on acquisition and acceleration on quantification. As for the method itself, the most frequently used algorithm is convolutional neural network and deep neural networks. Other studies included the comparison among different deep learning models and establishment of deep learning models based on different MRI sequences. This paper is to review the application of deep learning in CEST MRI in detail, which can be used as reference for interested parties in this field and further research and development on this basis. Then accelerate the clinical transformation of CEST MRI.