Reduced Scan Time in Multi-Echo Gradient Echo Imaging Using Two-Stage Neural Network
- Author:
Ji-Su YUN
1
;
Jong-Yun BAEK
;
Dong-Hyun KIM
Author Information
- Publication Type:Original Article
- From:Investigative Magnetic Resonance Imaging 2024;28(4):174-183
- CountryRepublic of Korea
- Language:English
-
Abstract:
Purpose:Multi-echo gradient echo (mGRE) images are used to acquire and analyze multiple echo signals. As the number of acquired echoes increases, more information on the voxel decay changes can be obtained, facilitating myelin water fraction (MWF) mapping.However, an increase in the acquired echoes leads to an increase in scan time. In this study, we developed a workflow to reduce the scan time using a two-stage neural network approach, which extrapolates additional echo images using mGRE data.
Materials and Methods:In Stage 1, a pseudo-T1 map was estimated using a U-net network combined with a simple signal model to correct the bias between two mGRE acquired with different scan parameters. The pseudo-T1 map was used to generate an initial echo time (TE) image from the mGRE images. In Stage 2, subsequent TE images were predicted from the initial echo image generated using a trained prediction network. The results were quantitatively compared with those obtained using a fitting algorithm. The MWF mapping results were then compared.
Results:The proposed model exhibited better root mean square error, structural similarity index measure, and peak signal-to-noise ratio, as well as a higher correlation with the MWF analysis compared to the fitting algorithm.
Conclusion:These results demonstrate that the proposed network can effectively reduce the scan time for mGRE image acquisition.
