收录:
摘要:
As brain investigation progresses, the need has become urgent from acquiring the higher resolution neuroimaging data to give a more detailed interpretation. In particular, the technological development and innovation of the Magnetic Resonance Imaging (MRI) machine, through increasing the magnetic field from low(such as 3T) to high (such as 7T), has revealed significant advantages regarding the image quality enhancement, etc. Currently, due to the limitations of hardware, physics and physiology, the large-scale acquisition of the high-resolution MRI neuroimages is still running on the road. Hence, enhancing the quality of the low-field MRI data is critical by using the advanced artificial intelligence technology. In this study, we propose a novel image enhancement framework, namely SR-MRI, trying to improve the quality of the low-resolution neuroimage: (1) combining with the Real-ESRGAN deep learning model; (2) bridging the 3T-MRI and the 7T-MRI within the same analysis scale; and (3) systematically comparing multiple evaluation indicators, including Brenner, SMD, SMD2, Variance, Vollath, Entropy, and NIQE. The experimental results suggest that the edge, fineness and texture features of the low-resolution neuroimages are restored to a great extent from the SR-MRI framework. In addition, the evaluation results of multiple indicators show that the processed 3T-MRI can achieve the similar effect from the 7T-MRI machine.
关键词:
通讯作者信息:
电子邮件地址:
来源 :
BRAIN INFORMATICS (BI 2022)
ISSN: 0302-9743
年份: 2022
卷: 13406
页码: 184-194
归属院系: