著者

Naoto Fujita Suguru Yokosawa Toru Shirai Yasuhiko Terada

出版者

Japanese Society for Magnetic Resonance in Medicine

雑誌

Magnetic Resonance in Medical Sciences (ISSN:13473182)

巻号頁・発行日

pp.mp.2023-0031, (Released:2023-07-28)

参考文献数

45

---

Purpose: Deep neural networks (DNNs) for MRI reconstruction often require large datasets for training. Still, in clinical settings, the domains of datasets are diverse, and how robust DNNs are to domain differences between training and testing datasets has been an open question. Here, we numerically and clinically evaluate the generalization of the reconstruction networks across various domains under clinically practical conditions and provide practical guidance on what points to consider when selecting models for clinical application.Methods: We compare the reconstruction performance between four network models: U-Net, the deep cascade of convolutional neural networks (DC-CNNs), Hybrid Cascade, and variational network (VarNet). We used the public multicoil dataset fastMRI for training and testing and performed a single-domain test, where the domains of the dataset used for training and testing were the same, and cross-domain tests, where the source and target domains were different. We conducted a single-domain test (Experiment 1) and cross-domain tests (Experiments 2–4), focusing on six factors (the number of images, sampling pattern, acceleration factor, noise level, contrast, and anatomical structure) both numerically and clinically.Results: U-Net had lower performance than the three model-based networks and was less robust to domain shifts between training and testing datasets. VarNet had the highest performance and robustness among the three model-based networks, followed by Hybrid Cascade and DC-CNN. Especially, VarNet showed high performance even with a limited number of training images (200 images/10 cases). U-Net was more robust to domain shifts concerning noise level than the other model-based networks. Hybrid Cascade showed slightly better performance and robustness than DC-CNN, except for robustness to noise-level domain shifts. The results of the clinical evaluations generally agreed with the results of the quantitative metrics.Conclusion: In this study, we numerically and clinically evaluated the robustness of the publicly available networks using the multicoil data. Therefore, this study provided practical guidance for clinical applications.

著者

Yo Taniguchi Suguru Yokosawa Toru Shirai Ryota Sato Tomoki Amemiya Yoshihisa Soutome Yoshitaka Bito Hisaaki Ochi

出版者

Japanese Society for Magnetic Resonance in Medicine

雑誌

Magnetic Resonance in Medical Sciences (ISSN:13473182)

巻号頁・発行日

pp.mp.2021-0045, (Released:2022-07-30)

参考文献数

21

被引用文献数

3

---

Purpose: MR parameter mapping is a technique that obtains distributions of parameters such as relaxation time and proton density (PD) and is starting to be used for disease quantification in clinical diagnoses. Quantitative susceptibility mapping is also promising for the early diagnosis of brain disorders such as degenerative neurological disorders. Therefore, we developed an MR quantitative parameter mapping (QPM) method to map four tissue-related parameters (T1, T2*, PD, and susceptibility) and B1 simultaneously by using a 3D partially RF-spoiled gradient echo (pRSGE). We verified the accuracy and repeatability of QPM in phantom and volunteer experiments.Methods: Tissue-related parameters are estimated by varying four scan parameters of the 3D pRSGE: flip angle, RF-pulse phase increment, TR and TE, performing multiple image scans, and finding a least-squares fit for an intensity function (which expresses the relationship between the scan parameters and intensity values). The intensity function is analytically complex, but by using a Bloch simulation to create it numerically, the least-squares fitting can be used to estimate the quantitative values. This has the advantage of shortening the image-reconstruction processing time needed to estimate the quantitative values than with methods using pattern matching.Results: A 1.1-mm isotropic resolution scan covering the whole brain was completed with a scan time of approximately 12 minutes, and the reconstruction time using a GPU was approximately 1 minute. The phantom experiments confirmed that both the accuracy and repeatability of the quantitative values were high. The volunteer scans also confirmed that the accuracy of the quantitative values was comparable to that of conventional methods.Conclusion: The proposed QPM method can map T1, T2*, PD, susceptibility, and B1 simultaneously within a scan time that can be applied to human subjects.