Accelerated 3D myelin water imaging using joint spatio‐temporal reconstruction

Abstract

Purpose To enable acceleration in 3D multi‐echo gradient echo (mGRE) acquisition for myelin water imaging (MWI) by combining joint parallel imaging (JPI) and joint deep learning (JDL). Methods We implemented a multistep reconstruction process using both advanced parallel imaging and deep learning network which can utilize joint spatiotemporal components between the multi‐echo images to further accelerate 3D mGRE acquisition for MWI. In the first step, JPI was performed to estimate missing k‐space lines. Next, JDL was implemented to reduce residual artifacts and produce high‐fidelity reconstruction by using variable splitting optimization consisting of spatiotemporal denoiser block, data consistency block, and weighted average block. The proposed method was evaluated for MWI with 2D Cartesian uniform under‐sampling for each echo, enabling scan times of up to approximately 2 min for 2mm×2mm×2mm$2\ {\rm mm} \times 2\ {\rm mm} \times 2\ {\rm mm}$ 3D coverage. Results The proposed method showed acceptable MWI quality with improved quantitative values compared to both JPI and JDL methods individually. The improved performance of the proposed method was demonstrated by the low normalized mean‐square error and high‐frequency error norm values of the reconstruction with high similarity to the fully sampled MWI. Conclusion Joint spatiotemporal reconstruction approach by combining JPI and JDL can achieve high acceleration factors for 3D mGRE‐based MWI.

JUST‐Net: Jointly unrolled cross‐domain optimization based spatio‐temporal reconstruction network for accelerated 3D myelin water imaging

“…This approach enhances the denoising capability of the network by representing a more complex function. 23 The second advantage lies in the decomposition of the spatio-temporal domain, which facilitates optimization and feature extraction. [43][44][45][46] In the reconstruction of complex-valued data, feature maps are independently extracted from the real and imaginary channels and then concatenated.…”

“…To compare and analyze the performance of the proposed method, we implemented ESPIRiT, 54 frequency only JUST-Net (JUST-f-Net), JLORAKS, 12 KIKI-net, 31 image only JUST-Net (JUST-i-Net), and JL + JDL-net. 23 We employed the ESPIRiT reconstruction method with specific parameters, including a kernel size of [7,7], a cutoff variance set at 0.0002, and a threshold of 0.9. Our experimental approach involved a total of 20 iterations.…”

“…[19][20][21][22] The application of neural network in various aspects has been used to achieve higher acceleration of the mGRE-based MWI in numerous studies. [23][24][25] However, previous convolutional neural network (CNN)-based methods, trained only on the image domain, resulted in missing or distorted details in the reconstructed images. 26,27 Consequently, the reconstructed MWI maps were affected by these residual artifccelerated mGRE images.…”

“…Inspired by the success of neural networks, 16–18 many researchers have recently used deep learning network as an approach to the reconstruction problems, and the combinations of PI and deep learning have been presented as an alternative solution with high fidelity 19–22 . The application of neural network in various aspects has been used to achieve higher acceleration of the mGRE‐based MWI in numerous studies 23–25 …”

“…To be more specific, the quality of the reconstructed MWI map was significantly influenced by the artifacts in the early echo images due to the short $$$ {T}_2^{\ast } $$$ properties of the myelin water components 28,29 . Furthermore, the impracticality of using PI methods such as joint low‐rank modeling of local k‐space neightborhoods (JLORAKS) for clinical real‐time applications is highlighted by the extensive reconstruction time required, which amounted to approximately 17 h per subject 23 . This significant elevation in JLORAKS reconstruction time becomes particularly pronounced when dealing with multi‐echo data acquisition, leading to a notable disadvantage associated with the processing of such data in actual clinical settings.…”