| Title |
Synthetic Fluid Attenuated Inversion Recovery Image from Multi-echo Gradient-recalled Echo Image via Deep Neural Network |
| Authors |
박지용(Jiyong Park) ; 김동현(Dong-Hyun Kim) |
| DOI |
https://doi.org/10.5573/ieie.2019.56.4.93 |
| Keywords |
FLAIR ; MRI ; Deep learning ; |
| Abstract |
MRI is a non-invasive imaging device that can represent a variety of contrasts. With the development of magnetic resonance imaging, a great deal of progress has been made in clinical research and basic science research in brain research due to its excellent sensitivity and diversity of applications. In this paper, we have studied FLAIR (fluid attenuated inversion recovery) images and mGRE (multi-echo gradient-recalled echo), which are used clinically. In this paper, we have developed a synthetic FLAIR image through deep neural network by scanning only mGRE image. Scanned 17 volunteers in their 24 to 29 years old. Of the 17 subjects, 15 were used as train data and 2 were used as test data. Echo 11 mGRE images were input into the deep neural network and the corresponding FLAIR images were labeled. Data preprocessing process and various parameters were adjusted to optimize and the performance changes of depth neural networks were compared with the number of echoes. As the image evaluation method, the structural similarity (SSIM) were observed. As a result, the SSIM values of the 2 test subjects showed a reliable similarity of 0.8764 and 0.8451. In this study, we tried to overcome limitations and problems of previous synthetic MR study and showed better results. But it is only for normal subjects, we should test the brain images of patients. |