Combining Multi-channel MP2RAGE Images with Minimized Noise

Jing Zhang¹, Bruce Bjornson², and Qing-San Xiang³

¹Applied Science Laboratory, GE Healthcare Canada, Vancouver, BC, Canada, ²Department of Pediatrics, University of British Columbia, Vancouver, BC, Canada, ³Department of Radiology, Department of Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada

Synopsis

Magnetization-prepared rapid gradient echo (MP-RAGE) has been widely used for T₁-weighted imaging. In order to overcome B₁ field inhomogeneity effect, the MP2RAGE sequence was introduced, with two complex images, GRE^TI1 and GRE^TI2, acquired at two inversion times TI₁ and TI₂. The MP2RAGE images are usually calculated from all the coils first and combined later into a final result. We propose an algorithm for multi-channel MP2RAGE image combination with minimized resulting noise.

PURPOSE

The calculated MP2RAGE image for the $$$i^{th}$$$ coil is obtained by (1): $$MP2RAGE_i=\frac{(GRE_i^{TI2}) \cdot (GRE_i^{TI1})^*}{\mid{GRE_i^{TI1}}\mid^2+\mid{GRE_i^{TI2}}\mid^2} \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space [1]$$ Images from all coils are currently combined with two methods (1): $$Method \space 1: MP2RAGE_{method1}=\frac{\sum_i\mid{GRE_i^{TI2}}\mid^2\cdot MP2RAGE_i}{\sum_i\mid{GRE_i^{TI2}}\mid^2}\space \space \space \space \space \space \space \space \space \space [2]$$ $$ Method \space 2: MP2RAGE_{method2}=\frac{\sum_i Re[(GRE_i^{TI2}) \cdot (GRE_i^{TI1})^*]}{\sum_i(\mid{GRE_i^{TI1}}\mid^2+\mid{GRE_i^{TI2}}\mid^2)}\space \space \space \space \space \space \space \space \space \space [3]$$ However, these methods lack optimization. Our goal was to find an optimal method for MP2RAGE image multi-channel combination.

METHODS

All experiments were performed on a 3T GE Discovery MR750 scanner with a GE 32-channel head coil on healthy volunteers. The acquisition parameters for MP2RAGE sequence was: $$$TR_{MP2RAGE}$$$ = 5 s, $$$TI_1/TI_2$$$ = 0.7 s/2.5 s, $$$\alpha_1=7^\circ$$$, $$$\alpha_2=5^\circ$$$, parallel imaging acceleration factor = 2. The total scan time was 6 min 25 s. Complex images from each channel were reconstructed by ARC (2). All calculated images $$$MP2RAGE_i$$$ were combined as an optimal weighted average (OWA) $$MP2RAGE_{OWA}=\sum_i w_i MP2RAGE_i\space \space \space \space \space\space \space \space \space \space \space \space \space \space \space \space \space \space [4]$$ where $$$w_i=\frac{\frac{1}{\sigma_i^2}}{\sum_i\frac{1}{\sigma_i^2}}$$$ is the normalized weighting factor. The noise variance $$$\sigma_i^2$$$ can be directly estimated from the imaginary part of $$$MP2RAGE_i$$$. The combined image in Eq. [4] can be shown to have a minimized noise (3). Similar approaches have been used in MRI for motion artifact reduction (4) and bSSFP debanding(5). The MP2RAGE-OWA images were compared to images combined with Method 1 and Method 2, in terms of noise level calculated for three different volumetric regions of interest (ROIs).

RESULTS

Figure 1 shows the representative real and imaginary parts of $$$GRE_i^{TI1}$$$ , $$$GRE_i^{TI2}$$$ and the $$$MP2RAGE_i$$$ image for the $$$i^{th}$$$ coil. The $$$MP2RAGE_i$$$ image has more noise on the anterior area due to lower coil sensitivity.

Figure 2 shows a series of $$$MP2RAGE_i$$$ images from all coils. The combined MP2RAGE-OWA image shows high resolution and very good contrast.

Figure 3 shows a representative slice of MP2RAGE images created by all three methods. The real part of all MP2RAGE images visually showed very good contrast in the brain. However, the noise varies near the surface of the brain (arrows) among the three images with more favorable result from OWA. The noise variances $$$\sigma_{MP2RAGE}^2$$$ have also been calculated (Table 1). The noise variance from the imaginary part was further decreased by significant percentages with MP2RAGE-OWA: white matter (14.7%), grey matter (22.4%), CSF (34.2%).

Figure 4 shows results under subject motion, with representative sagittal and axial MP2RAGE images combined from three different methods. The image artifact due to the motion are seen be much more reduced using the MP2RAGE-OWA. The noise variance (shown in Table 1) from the imaginary part was decreased by even more significant percentages with MP2RAGE-OWA: white matter (27.4%), grey matter (28.6%), CSF (41.0%).

DISCUSSION

It was observed that the MP2RAGE produces spatially uniform tissue contrast but amplifies noise outside of the brain and adjacent to the cortical gray matter, which can introduce errors in the automatic segmentation (6). MP2RAGE-OWA would be helpful for this application since noise in that region has been much reduced.

Recently, O’Brien et al. proposed a method to denoise the ratio image $$$MP2RAGE_i$$$ in Eq. [1] (6). Another new development MPnRAGE has also been reported to obtain different $$$T_1$$$ contrast (7). Our proposed OWA can also be combined with these techniques to further improve the image quality.

CONCLUSION

The OWA method was proposed to combine multi-channel MP2RAGE images. The MP2RAGE-OWA produced optimal result with the least noise variance and effectively reduced motion artifacts.

Acknowledgements

We thank Dan Rettmann (Rochester, GE Healthcare) for many useful suggestions on pulse sequence development. We thank Rachel Connett (Menlo Park, GE Healthcare) for her assistance in data collection. Daniel Kim and Dr. Lynne J. Williams (Vancouver, BC Children’s Hospital) are also acknowledged for their help.

References

1. Marques JP, Kober T, Krueger G, van der Zwaag W, Van de Moortele P-F, Gruetter R. MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at high field. NeuroImage 2010;49:1271–1281. doi: 10.1016/j.neuroimage.2009.10.002.

2. Brau A. New parallel imaging method enhances imaging speed and accuracy. GE Signa Pulse 2007:36–38.

3. Hedges LV, Olkin I. Statistical Methods for Meta-Analysis. 1 edition. Orlando: Academic Press; 1985.

4. Xiang Q-S, Henkelman RM. Weighted Average and Its application in Ghost Artifact Reduction. In: 9th SMRI proceedings. ; 1991. p. 222.

5. Hoff MN, Andre JB, Xiang Q-S. Performance Comparison of Analytical Solutions for bSSFP Signal Demodulation. In: ISMRM 23rd Annual Meeting, Toronto, Ontario, Canada; 2015. p. 3785.

6. O’Brien KR, Kober T, Hagmann P, Maeder P, Marques J, Lazeyras F, Krueger G, Roche A. Robust T1-Weighted Structural Brain Imaging and Morphometry at 7T Using MP2RAGE. Plos One 2014;9:e99676. doi: 10.1371/journal.pone.0099676.

7. Kecskemeti S, Samsonov A, Hurley SA, Dean DC, Field A, Alexander AL. MPnRAGE: A technique to simultaneously acquire hundreds of differently contrasted MPRAGE images with applications to quantitative T1 mapping. Magn. Reson. Med. Off. J. Soc. Magn. Reson. Med. Soc. Magn. Reson. Med. 2015. doi: 10.1002/mrm.25674.

Figures

Figure 1. Representative images from one of the receiver coils: (a) Real and imaginary parts of GRE_i^TI1 at TI₁. (b) Real and imaginary parts of GRE_i^TI2 at TI₂. (c) The calculated real and imaginary parts of the MP2RAGE_i image from Equation [1].

Figure 2. The workflow of multi-channel MP2RAGE image combination. (a) Real and imaginary parts of MP2RAGE images from each channel. (b) Real and imaginary parts of the combined MP2RAGE-OWA image.

Figure 3. The representative MP2RAGE images (real and imaginary parts) obtained from three different coil combination methods: (a) optimal weighted average, (b) method 1, (c) method 2. The result from OWA has lower noise level as indicated by the arrows and quantitative measurements from the square ROI.

Figure 4. Representative sagittal and transverse MP2RAGE images under subject motion. The images were combined from three different methods: (a) OWA, (b) method 1, (c) method 2. The OWA shows great advantages for motion artifact reduction.

Table 1. Comparison of noise variances from MP2RAGE images combined with three different methods over tissue specific ROIs. While methods 1 and 2 have similar performance, the proposed OWA yields significantly lower noise variances in all regions.

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

1917