RF Shimming for High Field MRI using Multi-channel Receive-Signals
Abhinav V. Sambasivan1, Lance DelaBarre2, Emad S. Ebbini1, Thomas J. Vaughan1,2, and Anand Gopinath1

1Electrical and Computer Engineering, University of Minnesota-Twin Cities, Minneapolis, MN, United States, 2Center for Magnetic Resonance Research, UMN-Twin Cities, Minneapolis, MN, United States

Synopsis

Counteracting the effects of B1 heterogeneities has been a major challenge for High field MRI systems. We propose here, a receiver-based approach called the Receive-RF Shimming (Rx-RFS) algorithm for multichannel MR systems which offers potential advantages in terms of reducing image acquisition time and mitigating SAR concerns. RX-RFS involves computing an optimal spatially-varying weight vector for combining the images from different receive elements. The reconstructed images (using Rx-RFS) exhibit enhanced contrast and more uniform signal levels when compared to standard reconstruction schemes throughout the entire Field-of-View. Rx-RFS also offers clinicians the flexibility to obtain local reconstructions at arbitrary Regions-of-Interest.

Purpose

High-field MRI systems (7T and higher) offer the advantage of enhanced signal-to-noise ratios and better resolution, but are very sensitive to $$$B_1$$$ inhomogeneities. Current mitigation methods like RF-shimming[1] involves adjusting the relative magnitudes and phases of the transmitted RF signals by solving an optimization problem to improve $$$B­_{1}^{+}$$$ homogeneity and/or reduce specific absorption rate (SAR). This results in increased overall image acquisition time. We propose here, a receiver-based approach, the Receive-RF-Shimming (Rx-RFS) for multi-channel MR systems to overcome the aforementioned effects. Being an entirely receiver-side technique, the Rx-RFS algorithm is devoid of SAR concerns, and requires only a single image scan, thus reducing acquisition times. Rx-RFS algorithm employs the transmit and receive element directivity patterns at each pixel to compute a spatially-varying weight vector for combining the complex image data from different receiving elements. Directivity of a coil element determines the fraction of the element excitation power that reaches a particular image pixel, and also the fraction of the pixel protons' emitted power received by a particular coil element. By using the array directivities, we go beyond previously proposed reconstruction schemes including the adaptive reconstruction method[2] and the RSS (Root of Sum-of-Squares) method[3] which do not leverage this information.

Methods

For an $$$n_c$$$-element MR multi-channel coil, the complex image pixel value in the image space, obtained from the $$$p$$$-th receiver can be modeled using the relationship between the MR signal and the spin density function $$$\rho$$$ as,$$I_p(x,y)=\mathbf{h^+}(x,y)\mathbf{m_{tx}}D_p^-(x,y)\rho(x,y)+\nu(x,y),\quad\quad\quad(1)$$where $$$D_p^-(x,y)$$$ is the $$$p$$$-th receiver directivity at location $$$(x,y)$$$, $$$\mathbf{h^+}(x,y)=[D_1D_2\dots D_{n_c}]$$$ is the array directivity at $$$(x,y)$$$, and $$$\mathbf{m_{tx}}$$$ is the $$$n_c$$$x$$$1$$$ transmit excitation vector (assuming the $$$n_c$$$ elements are operating in transmit-receive mode). Also the noise function $$$\nu(x,y)$$$ is assumed to be independent complex-valued Additive White Gaussian Noise (AWGN). Equation (1) can be vectorized (for $$$p=1,2,\dots,n_{c}$$$) and written more compactly as, $$$\mathbf{I}(x,y)=\mathbf{H}(x,y)\rho(x,y)+\nu(x,y)$$$, where $$$\mathbf{H}(x,y)=\mathbf{h^+}(x,y)\mathbf{m_{tx}}\mathbf{h^-}(x,y)$$$, and the receiver directivities at a point $$$(x,y)$$$ are encapsulated by the vector $$$\mathbf{h^-}(x,y)$$$. When $$$\nu(x,y)$$$ is AWGN with variance $$$\sigma_{\nu}^2$$$, the spin density function can be estimated by a linear minimum-mean-square estimator[4] of $$$\rho(x,y)$$$ given $$$\mathbf{I}(x,y)$$$,$${\hat{\rho}(x,y)}={\cfrac{\mathbf{H}^{H}}{\left\|\mathbf{H}\right\|^{2}_{2}+\hat{SNR}^{-1}(x,y)}}{\mathbf{I}}(x,y).\quad\quad\quad(2)$$Here, $$$\hat{SNR}=\hat{\sigma}^2_{\rho}/\hat{\sigma}^2_{\nu}$$$ is a local estimate of the SNR. It is clear that, Equation (2) yields a SNR-regularized spatial inverse filter with respect to the array directivity vector. Equation (2) is applied at all points in the FOV where the estimated SNR is greater than a certain threshold and at all other points we reduce the pixel value by a constant factor. This SNR thresholding prevents noise amplification in pixels outside the skull where there is no signal component. The MRI data was collected from a 16-channel TEM coil, (24cm×20cm elliptical array shown in Figure 1) 7T MRI system, with TR/TE = 40/4.08ms and 25×25cm2 FOV. The transmitter and receiver directivities were estimated by computational methods using the Remcom software.

Results

The above method can be applied on a pixel-by-pixel basis, as given in Equation (2) to yield a reconstruction over the entire field-of-view (FOV). More interestingly, the proposed method offers the flexibility of choosing arbitrary Regions-of-Interest (ROIs) while performing the reconstructions. For example, since at 7T, the array point-spread-function $$$psf(x,y)$$$, is well approximated by a Gaussian function in the XY plane (spatial co-ordinates) with main lobe spanning several pixels, the directivity $$$\mathbf{H}(x,y)$$$ obtained by focusing the receive-beam at a point of interest $$$(x_0,y_0)$$$ can be easily recomputed by point-wise multiplication of the measured/estimated directivity with $$$psf(x-x_0,y-y_0)$$$. Figure 2 shows a montage of 6 such images obtained by considering different focal points. Each of the individual images locally exhibit high contrast revealing anatomically details which are not observable otherwise. Also Figure 3 provides a comparison between the adaptive reconstruction method[2] and the Rx-RFS algorithm (applied over the entire FOV). The conventional reconstruction (without the Rx-shimming) in Figure 2(a) shows darker regions (with lower SNR) near the center of the brain. Reconstruction using the Rx-RFS algorithm shown in Figure 2(b) clearly exhibits enhanced signal levels and uniform contrast throughout the image (FOV). Anatomical details like the ventricle structure are much clearer in the reconstruction using the Rx-RFS algorithm.

Conclusion

Incorporating the directivities of array elements for image reconstruction in multi-channel MRI systems produces improved results when compared to existing techniques. This method is unaffected by SAR concerns, and helps counteract the effects of $$$B_1^+$$$ heterogeneities and produce images with uniform contrast. The improvements demonstrated above can be achieved with no additional changes to the existing hardware or imaging sequences of current MRI systems. If the element directivities are computed beforehand and tabulated, such a reconstruction scheme is extremely practical and beneficial.

Acknowledgements

University of Minnesota Grant-in-Aid fund and the NIH-EB0006835

References

[1] J. T. Vaughan, et al. “RF image optimization at 7T and 9.4T”, Proc. of the 13th Annual Meeting of the ISMRM, p. 953, 2005

[2] D. O. Walsh, et al. “Adaptive Reconstruction of Phased Array MR Imagery”, MRM, vol. 43, no. 5, pp. 192-225, 2000

[3] P. B. Roemer, et al. “The NMR Phased Array”, MRM, vol. 16, pp. 682-690, 2000

[4] Ali H Sayed. “Adaptive Filters”, Wiley Interscience, 2008

Figures

Figure 1: 16 channel TEM elliptical coil

Figure 2: Montage of Reconstructions with 6 focal points situated at different locations in the head

Figure 3: Reconstructed brain images (Full FOV) obtained from

(a) Adaptive reconstruction described by D.O.Walsh, et al.,

(b) Rx-RFS Algorithm




Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1019