Introduction

Estimating the $B_1^+$-field is essential for many quantitative MRI methods and with the advent of ultra high-field MRI systems the demand of fast an accurate methods has become even more crucial.
In Bloch-Siegert shift based $B_1^+$-mapping an off-resonance RF pulses is applied which induces a phase shift proportional to the peek amplitude $B_{1p}$ of the RF pulse ¹. Because the shift is proportional to the RF power deposition, the method is SAR demanding, requiring long acquisition times. To increase the efficiency, the frequency of the RF-pulse can be decreased². Another possibility is a dense sampling of the k-space center in combination with variational image reconstruction³. Further, acquisition can be accelerated with 2D spiral and EPI readouts ⁴.
Here, we investigate the potential of faster image acquisition with a 3D segmented EPI sequence. To exploit errors associated with faster acquisition, the number of interleaves and the block-size of the k-space center were varied. This work shows that with an adequate choice of interleaves and block-size, whole-brain 3D $B_1^+$-maps can be acquired in 5 to 10 seconds.

Methods

Bloch-Siegert shift: Assuming that the off-resonance frequency $\omega_{RF}$ of the RF pulse with peak amplitude $B_{1p}$ is much larger than the field inhomgoneities $\Delta \omega_{B0}$ ( $\omega_{RF} >> \Delta \omega_{B0}$) and $|\gamma B_{1p}| << \omega_{RF}$, the Bloch-Siegert phase shift $\phi_{2BS}$ is given by¹:

$$\begin{equation} \phi_{2BS} =B_{1p}^2 \int_0^T \frac{(\gamma B_{1n}(t))^2}{\omega_{RF}} dt = B_{1p}^2 K_{BS}\end{equation}$$
wher $B_{1n}(t)$ is the normalized pulse envelope and the constant $K_{BS}$describes the pulse properties.

Sequence: We implemented a 3D segmented EPI with a variable number of interleaves $N_{Intlv}$ in phase encoding direction $k_y$ ⁵. For each shot, one partition $k_z$is encoded. In addition, a navigator echo prior and after the echo train and after was integrated to measure time-dependent $\Delta \omega_{B0}$ fluctuations. Before the imaging-part, one shot was acquired without phase encoding gradients to correct for Nyquist ghosts. Furthermore, 20 preparation pulses were executed to approach faster a steady-state. Because of the variable implementation of $N_{Intlv}$, the sequence can be scaled to a standard RF-spoiled gradient-echo sequence, and up to the acquisition of one partition in a single shot. In addition, a variable block-size in the k-space center can be selected to measure $n_y$ time $n_z$ out of $N_y$ times $N_z$ fully sampled k-space lines.

In-vivo measurements: Brain MRI of a healthy subject were carried out on 3T MR system (Magnetom Vida, Siemens Healthcare, Erlangen, Germany) with the following sequence parameters: binominal -1 1 water excitation pulse with $\alpha = 20^\circ$ and RF-spoiling increment of $90^\circ$ ⁶, $4ms$ Gaussian saturation pulse with nominal $\phi_{BS,nominal} = 46^\circ$, $TR = 45ms$, $BW=2000Hz/px$,$FOV=256x256x192mm^3$ and matrix size$N_x=64$, $N_y=64$ and $N_z=48)$.
First, fully sampled k-space ($n_y=64, n_z=48$) images were acquired with different $N_{Intlv}$. Second, the block-size was varied ($n_y= [32,16,12]$ and $n_z=[24,12,8,6]$), each with variable $N_{Intlv}$.

Image reconstruction: For all EPI data-sets ($n_y \neq N_{intlv}$), Nyquist-correction prior image reconstruction was applied. To combine the individual coil images and to estimate $\phi_{2BS}$, each coil image for the positive $\omega_{RF}$ was multiplied by the complex conjugate of the negative $\omega_{RF}$ coil image and summed up.

Results and Discussion

Figure 1 shows the reconstructed images for different $N_{intlv}$ (top), the estimated $B_1^+$ maps (middle), and the absolute error with respect to the gradient-echo images ($N_{intlv} = 64$). For the fully sampled k-space, visually a good correspondence between $B_1^+$-maps from $N_{intlv}=64$ with an acquisition time $TA=280s$ to $N_{intlv}=4$ ($TA=18s$) can be observed. Further reduction of $N_{intlv}$ leads to an increase of the mean absolute error (MAE). For $N_{intlv}=1$, distortions become more severe, which is also reflected in the $MAE$ by $5.2\%$.
The results for different block-patterns for varying $N_{intlv}=1$ are depicted in Figure 2. When comparing the quality of $B_1^+$-maps and the $MAE$ with the fully sampled results (Figure 1), it can be observe that the $MAE$ tends to be smaller for faster $TA$.

Conclusion

We have demonstrated that with segmented 3D EPI highly accelerated whole-brain $B_1^+$-maps can be acquired. To improve image quality, and potential further acceleration, variational reconstruction methods will be implemented and the navigator signals will be incorporated to compensate temporal $\Delta \omega_{B0}$ fluctuations⁷.

Acknowledgements

No acknowledgement found.