Synopsis

Keywords: Machine Learning/Artificial Intelligence, Quantitative Imaging, water/fat separation

Most of the water/fat separation techniques need to acquire multiple images with different echo time, which usually take long acquisition time. Multiple overlapping-echo detachment (MOLED) imaging can shorten acquisition time by acquiring multiple MR echo signals in the same k-space. Here, a new method for fast water/fat separation T₂* mapping with MOLED technology was proposed, which can obtain quantitative T₂* maps and M₀ maps of water and fat in a single shot. In vivo experiment demonstrates that the proposed method can obtain accurate quantitative parameter values and accurate separation of water and fat images.

Introduction

Water/fat separation is important for MRI applications. Additionally, T₂* relaxation time of the water and fat signals can provide valuable information for biomedical research and clinical diagnosis of diseases.¹ However, most of the water/fat separation techniques need to acquire multiple images with different echo time, which usually take long acquisition time, and may be susceptible to physiological motions, such as Dixon-based methods.^2,3
Multiple overlapping-echo detachment (MOLED)⁴ imaging provides an innovative idea for accelerating MR parametric mapping by acquiring multiple MR echo signals with different signal evolutions in the same k-space. Here, we proposed a new method for fast water/fat separation T₂* mapping with MOLED technology. With MOLED technology, water and fat signals with different T₂* weightings can be obtained in a single shot, which allows us to obtain quantitative T₂* maps and M₀ maps of water and fat without additional scans. Experimental results show that the proposed method can obtain accurate quantitative parameter values and accurate separation of water and fat images.

Methods

Pulse sequence design: The MOLED sequence is shown in Figure 1. Signals excited by four sequential RF pulses are moved to different locations in k-space by shifting gradients (G_ROi,G_PEi,i=1,2,3,4). Echoes with different evolution time are acquired and overlapped in the k-space. GRE-EPI readout is used in two echo trains. Consequently, these echoes contain different information of water and fat, and convey T₂* information with different weights, hence allowing water/fat separation and T₂* quantification. There are eight echoes in two echo trains totally, providing enough information for multi-peak water/fat separation. The TEs of these echoes range from 5.45 ms to 43 ms, while the T₂* value of water and fat in the leg region is generally distributed within 20-35ms. The appropriate TE range ensures the accurate T₂* quantification.
Signal model: The nth echo S_n with echo time TE_n can be expressed as:
$$S_{n}=(W\times{e^{\frac{-TE_{n}}{T_{2,w}^*}}}+\sum_{m=1}^6\alpha_{m}e^{i2\pi\delta{f_{m}}TE_{n}}\times{F}\times{e^{\frac{-TE_{n}}{T_{2,f}^*}}})e^{i2\pi\phi{TE_{n}}}$$
where W and F represent the complex water and fat signals. A six-peak fat model is used in this work, and the relative amplitude α_m of each peak as well as their frequency shift δf_m are taken as a prior.⁵ φ(Hz) is off-resonance frequency induced by main magnetic field inhomogeneity. $$$T_{2,w}^*$$$is the T₂* of water and $$$T_{2,f}^*$$$ is the T₂* of fat. Here, we assume all fat peaks have equal T₂*.
Water/Fat separation using neural network: The training data were generated using synthetic data and MRiLab software, as shown in Figure 2a. The MOLED sequence was used to stimulate water and six fat peak components to get their respective MR signals. The image of each component was obtained and added together to get the final MOLED image. U-Net was employed, as shown in Figure 2b. The inputs of U-Net are the real and imaginary parts of original MOLED image as well as shifted MOLED image to correct chemical shift artifact along the PE dimension, and the outputs are T₂* maps and M₀ maps of water and fat, respectively.
Experiments: This study was approved by the IRB at Zhongshan Hospital of Xiamen University. MOLED sequence was performed in the leg region on a 3T scanner (Ingenia CX, Philips Healthcare), using a 32-channel abdomen receive coil. Imaging parameters were as follows: flip angle of excitation pulse = 30°, TR = 6000 ms, ESP = 0.664 ms, FOV = 22×22 cm², slice thickness = 5 mm, imaging matrix size= 128×124. The traditional mDixon-Quant technique and GRE sequence were used to obtain reference images. Single slice scan time of MOLED, mDixon-Quant and GRE were 65, 448 and 5514 ms, respectively.

Results

Figure 3 shows the water/fat separation results of two different slices (a and b). The second row shows the M₀ mapping results. The water/fat separation results of MOLED are accurate and show clear texture, matching well with mDixon. Besides, the chemical shift artifacts along the PE dimension are corrected. The third row displays the T₂* mapping results. Especially, the third image in the third row is the T₂* map of MOLED obtained by adding the T₂* maps of water and fat. The fourth row shows the reference T₂* map obtained by GRE sequence and the T₂* error maps. In slice #1 (Figure 3a), the average errors of the maps of MOLED and mDixon are 6.61%, 14.41%, respectively, and in slice #2 (Figure 3b), the corresponding values are 6.01%, 12.30%, respectively. This indicates that MOLED can obtain more accurate T₂* maps of water and fat than mDixon.

Discussion and conclusion

In this work, we proposed a new method for fast water/fat separation T₂* mapping based on MOLED technology, which allows us to obtain T₂* maps and M₀ maps of water and fat in a single shot. Multi-peak fat model, synthetic data, and U-Net were used in image reconstruction. The proposed method can obtain accurate quantitative parameter values and accurate separation of water and fat images.

Acknowledgements

This work was supported by the National Natural Science Foundation of China under grant numbers 82071913 and 11775184.