Selective Channel Combination of 3D Phase Offset Corrected 7T MRI Phase Images
Shaeez Usman Abdulla1, David C Reutens1, Kieran O'Brien2, and Viktor Vegh1

1Centre for Advanced Imaging, University of Queensland, Brisbane, Australia, 2Siemens Healthcare Australia Pty. Ltd., Australia


A number of MRI applications rely on accurate phase images. At ultra-high field, the limitations of various processing techniques and the absence of a volume reference coil make the combination of multi-channel signal phase challenging. We propose a method which combines phase offset corrected signal phase data across a selection of channels. We evaluated method performance at two different gradient recalled echo MRI echo times. We qualitatively and quantitatively studied the combined phase quality in distinct brain regions. We found that using a subset of channels leads to improved phase images than when all channels are used in the combination.


The rapid increase in the development of MRI phase imaging at ultra-high field can be attributed to improved grey-white matter contrast and to advances in frequency shift and quantitative susceptibility mapping. These methods rely on phase images and have enabled studies on iron deposition, calcifications and contrast agent distribution in the brain.1-4 A standard method for multi-channel signal phase combination in systems that do not have a dedicated body coil or volume reference coil is lacking.1,5 A number of methods have been proposed to combine phase data, and we have shown that phase images can be improved by using only a selection of channel data.1,2,4 The noise in phase images has been shown to deviate from the Gaussian distribution when signal-to-noise ratio is low in signal phase, implying that the use of magnitude images for the creation of sensitivity maps for phase data combination may not lead to noise optimal results.4 To be able to standardise multi-channel phase combination at ultra-high field, we propose the use of 3D phase offset corrected MRI phase images and combining them in a selective manner.1,4


A multi-echo 3D GRE-MRI non-flow compensated data was collected on a 7T ultra-high field whole-body MRI research scanner (Siemens Healthcare, Erlangen, Germany) with a 32 channel head coil (Nova Medical, Wilmington, USA) using the following parameters: TE1 = 4.98ms, TE2 = 8.11ms and TE5 = 17.5ms, TR = 48ms, flip angle = 15o, voxel size = 0.75mm $$$\times$$$ 0.75mm $$$\times$$$ 0.75mm and matrix size = 280 $$$\times$$$ 242 $$$\times$$$ 160.

Fig 1 shows the selective combine pipeline. Individual channel 3D phase offset maps were computed as described for MCPC-3D, and the MCPC-3D method was used for comparison wherein all channels are magnitude weighted and combined to generate a combined phase image.1 The median filter size in the 3D phase offset computation was set to 15 $$$\times$$$ 15 $$$\times$$$ 15. For selective combine, individual channel phases were subtracted to obtain phase noise images, their standard deviation was calculated, and the result was Gaussian filtered to obtain matrices on which channel selection was based.4 The TE2 (see Fig 1) voxel phase was averaged based on the channel ranking provided by channel selection matrix entries. Scripts were implemented in MATLAB®. Tissue phase maps were obtained from the combined phase images using FSL brain masks with iHARPERELLA in STI Suite.6,7 We performed a region-by-region analysis, and regions are shown in Fig 2. We tested for (T) difference of means using a t-test, (U) changes in distribution mean and shape using the Kolmogorov-Smirnov test, (N) changes in shape only using the normalised Kolmogorov-Smirnov test and evaluated (H) entropy. The p-values provided for each test are with respect to the MCPC-3D result. We expect a lower value for entropy to correspond to a lower level of image noise.


In Fig 3 the selectively combined phase images are provided alongside the result obtained using MCPC-3D. Fig 4 shows qualitative images for each of the regions highlighted in Fig 2 are shown. In Fig 5 we present quantitative results wherein statistics are provided for differences between MCPC-3D and selective combine.


The 3D phase offset correction step in Fig 1 results in a combined phase image which has artefacts at locations depicted by the arrow. These lines disappear as the number of channels used in the combined phase image increases. We attribute the presence of these lines to not having perfect correspondence between 3D offset corrected phase images, the effect of which averages out with increased number of channels. Our results depicted in Fig 4 imply that selective combine can achieve improvements in image quality using less than the total number of channels. The entropy measure indicates that noise does not necessarily decrease by increasing the number of channels used in the combined phase image. Unlike in MCPC-3D, selective combine does not use magnitude signal weighting of the individual 3D phase offset corrected images prior to combining them. Based on these results, it appears magnitude weighting may not sufficiently supress regions of high phase noise (compare 32 channel result to MCPC-3D in Fig 5).


We have shown qualitatively and quantitatively that the use of a subset of channels can result in combined phase images with improved image quality. Improvement in phase image quality may directly result in an improvement in frequency shift and quantitative susceptibility maps.


The authors acknowledge the facilities of the National Imaging Facility at the Centre for Advanced Imaging, University of Queensland. We also acknowledge funding from the Australian Research Council (DP140103593).


1. Robinson S, Grabner G, Witoszynskyj S, Trattnig S. Combining phase images from multi-channel RF coils using 3D phase offset maps derived from a dual-echo scan. Magnetic resonance in medicine. 2011 Jun 1;65(6):1638-48.

2. Parker DL, Payne A, Todd N, Hadley JR. Phase reconstruction from multiple coil data using a virtual reference coil. Magnetic resonance in medicine. 2014 Aug 1;72(2):563-9.

3. Sood S, Urriola J, Reutens D, O’Brien K, Bollmann S, Barth M, Vegh V. Echo time-dependent quantitative susceptibility mapping contains information on tissue properties. Magnetic Resonance in Medicine. 2016 May 1.

4. Vegh V, O'Brien K, Barth M, Reutens DC. Selective channel combination of MRI signal phase. Magnetic resonance in medicine. 2015 Nov 1.

5. Robinson SD, Bredies K, Khabipova D, Dymerska B, Marques JP, Schweser F. An illustrated comparison of processing methods for MR phase imaging and QSM: combining array coil signals and phase unwrapping. NMR in Biomedicine. 2016 Jan 1.

6. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE, Niazy RK. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage. 2004 Dec 31;23:S208-19.

7. Li W, Avram AV, Wu B, Xiao X, Liu C. Integrated Laplacian-based phase unwrapping and background phase removal for quantitative susceptibility mapping. NMR in Biomedicine. 2014;27(2): 219-27.


Fig 1. Shown is the pipeline used to selectively combine multi-channel phase data. The phase data from all the channels at both echoes were unwrapped and combined according to produce a 3D offset map for each channel. Phase offset maps were median filtered and then subtracted from the second echo unwrapped phase, the result of which were phase offset corrected images. The second echo corrected phase images were used to create channel selection maps to produce the combined phase image.

Fig 2. Five regions of size 25 $$$\times$$$ 25 voxels were used in the quantitative analysis for slice number 100 in the axial plane.

Fig 3. Shown are the combined phase images at two echo times reconstructed with selective combine using varying number of channels. Phase images reconstructed with MCPC-3D are shown for comparison. Scale is in radians. The arrows show regions where phase offset corrected images have a residual difference between them. This artefact averages out as number of channels used in the combine increases.

Fig 4. Qualitative results for the five regions as a function of the number of channels used in selective combine compared to MCPC-3D. In R1 and TE = 8.11ms, combining using all channels provides better image contrast than MCPC-3D, suggesting weighting using magnitude images can lead to image deterioration. This is also the case for R4 and TE = 17.5ms, wherein additional contrast can be gained by using a smaller number of channels (16 to 24). The apparent difference between MCPC-3D and selective combine in R1 and TE = 8.11ms is due to a shift in region mean values.

Fig 5. Quantitative analysis for R1 and TE = 8.11ms and R4 and TE = 17.5ms (dashed boxes in Fig. 4). (T) is t-test for difference of means, (U) is the Kolmogorov-Smirnov (KS) test, (N) is the normalised KS test and (H) is entropy. In R1 and R4, entropy was lower with 32 channels in comparison to MCPC-3D, confirming qualitative findings. For R4, entropy for 16 to 24 channels was lower than when using 32 channels, also confirming qualitative findings. Results for R1 suggest that the region mean value could shift as a consequence of how phase images were combined.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)