Introduction

Multi-parametric quantitative MRI provides a comprehensive understanding of tissue characteristics, promising applications in medical diagnosis and research.¹ However, the long acquisition time limits its widespread application. Fast quantitativeMRI methods based on overlapping echoes with EPI readout have been developed,^2,3 but the collected magnetic resonance images have deficiencies such as distortion and eddy current artifacts. To overcome these problems, we design a method SSFP-MOLED, which combines balanced steady-state free precession (bSSFP) sequence with multiple overlapping echo detachment (MOLED) imaging technique. Experiments confirm its ability to simultaneously yield precise T₁, T₂, T₂^*, PD, B₀, and B₁ maps.

Methods

Pulse sequence design
Figure 1 shows the proposed sequence, which consists of two consecutive SSFP-MOLED sequences. In Sequence 1, the pulse flip angle starts from α₀ and gradually decreases by a step size of ∆α. It decreases to the minimum angle α₀-n∆α after n TRs, then gradually increases to α₀ and remains at α₀, making the magnetization gradually enter a steady state. After Sequence 1 ends, an IR module is inserted, followed by the second SSFP-MOLED sequence, where the pulse flip angle remains at α₀. Throughout the sequence, the phase of each flip angle alternates between 0° and 180°.
Figure 1(b) shows the sequence in a TR period of Sequence 1. Based on the bSSFP sequence, a pair of echo-shifting gradients G_RO1 and G_PE1 is added to the frequency-encoding and phase-encoding dimensions, respectively. Due to the echo-shifting gradients, the time to form an echo is advanced or delayed, making the echo to be separated in the readout dimension. Figure 1(c) shows the sequence in a TR period of Sequence 2. It applies two pairs of echo-shifting gradients (G_RO2, G_RO3, G_PE2, and G_PE3) on the frequency-encoding and phase-encoding dimensions respectively.
The SSFP-MOLED sequence yields 6 k-spaces totally (Figure 1(d)). By applying echo-shifting gradients, several overlapping echoes will form in each k-space. The location of each echo in the k-space is determined by the size and direction of each echo-shifting gradient, and different echoes in each k-space have different parameter weightings. The three k-spaces of Sequence 1 have different T₂^* weightings, which can achieve T₂^* and B₀ quantification. The equivalent flip angles of different echoes in each k-space are different, which collects different T₂/T₁ weighted echo signals and can achieve B₁ mapping. The echoes in the three k-spaces of Sequence 2 have different T₁ weightings, and each echo in one k-space experiences different equivalent inversion time, which can achieve T₁ quantification. After T₁ is quantified, T₂ and PD maps can be further obtained.
Synthetic Data and Network training
Figure 2 shows the process of data synthesis and network training. Templates that contain different parametric maps were used for data synthesis. The Bloch simulation was performed to synthesize the SSFP-MOLED images using MRiLab.⁴ Then the synthetic SSFP-MOLED images and corresponding template were combined into a training sample pair. U-net was used for image reconstruction. The inputs were the amplitudes and phases of SSFP-MOLED images, and the outputs were parametric maps.
Experiments
Experiments were conducted on a 3T scanner (Ingenia CX, Philips Healthcare). SSFP-MOLED imaging parameters: flip angle α₀ = 35°, ∆α = 0.25°, field of view = 22×22 cm², matrix size = 256×256. For Sequence 1: TR1 = 40 ms, TE = 10/20/30 ms, G_RO1 = 0.2´A_ro, G_PE1=0.2´A_pe. For Sequence 2: TR2 = 11.12 ms, TE =2.78/5.56/8.34 ms, G_RO2 =-0.15´A_ro, G_RO3 = 0.3´A_ro, G_PE2 = 0.15´A_pe, G_PE3 = -0.3´A_pe. Here A_pe is twice the minimum G_pp gradient's absolute value,and A_ro is the area ofreadout gradient. Single slice scan time of SSFP-MOLED was 13.1 s. Reference T₁ maps were acquired using multi-shot IR-SE-EPI, T₂ and PD maps were acquired using SE, T₂^* and ∆B₀ maps were acquired using multi-echo GRE, and B₁ maps were acquired using double angle SE-EPI.

Results

Figure 3 shows the results of a healthy volunteer. The parametric maps from SSFP-MOLED are consistentwith the reference counterparts. Figure 4 shows the statistical results of 50 ROIs (10 per slice). The T₁ (R² = 0.8936), T₂ (R² = 0.8222), and T₂^* (R² = 0.9839) values from SSFP-MOLED and reference measurements are highly correlated.

Discussion and conclusion

We propose a new method for multi-parametric quantitativeMRI. It collects multiple k-spaces in one scan, and each k-space contains multiple echo signals with different parameter weightings. It can obtain accurate T₁, T₂, T₂^*, PD, B₀, and B₁ parametric maps simultaneously. The proposed method has the advantages of being fast, accurate, distortion-free, artifact-free, and no need for registration, and has great potential for clinical application.

References

1. Wang L, Gaddam S, Wang N, et al. Multiparametric mapping magnetic resonance imaging of pancreatic disease. Front Physiol. 2020; 11:8.

2. Cai CB, Wang C, Zeng YQ, et al. Single‐shot T₂ mapping using overlapping‐echo detachment planar imaging and a deep convolutional neural network. Magn Reson Med. 2018; 80(5): 2202-2214.

3. Zhang J, Wu J, Chen SJ, et al. Robust single-shot T₂ mapping via multiple overlapping-echo acquisition and deep neural network. IEEE Trans Med Imaging. 2019; 38(8): 1801-1811.

4. Liu F, Velikina JV, Block WF, et al. Fast realistic MRI simulations based on generalized multi-pool exchange tissue model. IEEE Trans Med Imaging. 2016; 36: 527-537.