The potential of 256 channel high density receive arrays in combination with 2DCAIPIRINHA at 7T
Arjan D. Hendriks1, Cezar B.S. Alborahal2, Michel G.M. Italiaander2, Dennis W.J. Klomp1,2, and Natalia Petridou1

1Department of Radiology, University Medical Center Utrecht, Utrecht, Netherlands, 2MR Coils B.V., Drunen, Netherlands


There is an overall drive to high speed, high resolution brain MRI. High density receive arrays have proven to be a very valuable tool in this process. However, alternative acquisition strategies like 2DCAIPIRINHA have shown to be highly effective in acceleration. We investigated whether a 256ch high density receive array with or without 2DCAIPIRINHA can still outperform a 32ch array with 2DCAIPIRINHA.


The brain is organized at multiple spatial scales, ranging from small neuronal populations to brain-wide networks. To understand how the brain operates we need to measure dynamic processes across these spatial scales, concurrently.

Present-day efforts are geared towards accelerated imaging techniques (SENSE [1], CAIPIRINHA [2], SMS [3]). These techniques have significantly advanced our capability to image whole brain function at a resolution in the order of millimeters in a few seconds (for example 1.25 mm isotropic in 2 sec at 7T for human connectome project [4]). At high fields, the increased SNR and sensitivity offers the advantage of increased resolution. However, very high resolution scans still have a penalty of increased scan time for whole brain imaging (for example 0.55 mm isotropic in 6 sec [5]) or restrictions to small areas in the brain.

The performance of accelerated imaging methods depends on the coil used. High density receiver arrays allow for higher acceleration factors with reduced g-factors and very high resolutions [5]. It may be questioned whether alternative acceleration techniques can improve acceleration even further. Which is why, in this work, we investigate 2DCAIPIRINHA [6] performance in an extension of the high density array, that is a 256 channel coil (Figure 1) for whole brain imaging.


In this study, while in the process of installing the receivers, simulations are performed to investigate the possible acceleration performance of a virtual 256 channel coil in combination with an advanced parallel imaging sequence, i.e. 2DCAIPIRINHA for 3D datasets.


All measurements were performed on a healthy volunteer in a 7 Tesla Achieva system (Philips, Cleveland, USA). To simulate the performance of a 256ch whole brain receive array, two 16ch receive arrays (MR Coils BV, Drunen, the Netherlands) were shifted 8 times to different positions on the head (Figure 2). For each position a reference scan was acquired, yielding a total of 8 reference scans, each containing signal from 32 channels. For comparison purposes, a reference scan was also acquired with the standard 32 channel headcoil (Nova Medical, USA). The SENSE reference scans were acquired as follows: 3D gradient echo, acquired interleaved with receiver arrays and volume coil; TE/TR= 1.22/8.0 ms, 2x2x2 mm3 voxel, 20x20x20 cm3 FOV, acquisition time= 1min. After scanning, sensitivity maps were constructed from the reference scans. The resulting 8 sets of 32 channels were combined to form a virtual 256ch coil. The sensitivity maps were calculated using ReconFrame (Gyrotools, Zurich, Switzerland). To correct for head displacement in between reference scans, images acquired for each coil position were aligned to the images of the first position using AFNI. Alignment parameters (rotation & translation) were calculated, using the reconstructed magnitude images from the reference scan. Afterwards these parameters were applied to the sensitivity matrices per coil position.

The g-factor (geometry factor, noise amplification factor) was calculated from the sensitivity maps according to Pruessman et al. [1]. Multiple g-factor calculations were performed for different combinations of SENSE acceleration factors and different 2DCAIPIRINHA undersampling patterns (Figure 3).


The calculated g-factor maps (Figure 4) are shown for different acceleration factors, for both the standard 32ch head coil (Figure 4, top) and the virtual 256ch coil (Figure 4, bottom). The undersampling patterns shown are 2DSENSE (left) and 2DCAIPIRINHA (right). Different acceleration factors are displayed in feet-head (FH) and anterior-posterior (AP) direction. When considering a g-factor of less than 2 (to avoid excessive noise amplification), the maximum achievable acceleration for the standard 32ch headcoil is: 9 (SENSE 3x3) and 12 (CAIPIRINHA 4x3_FH2). For the virtual 256ch headcoil this is: 24 (SENSE 4x6) and 28 (CAIPIRINHA 4x7_AP3).

Discussion and conclusion

The results show that a virtual 256ch head coil allows for more than 2-fold acceleration performance compared to the standard 32ch head coil. In addition, for both the standard head coil and the virtual 256ch coil, a 2DCAIPIRINHA sampling pattern results in significant improvements for a high SNR or acceleration performance. Overall, a spectacular maximum acceleration factor of 28 is estimated. Note that the reference scan of the virtual 256ch coil is acquired consecutively, excluding incorporation of the full noise correlation matrix in the simulation. However, coil coupling is expected to be highest between neighboring coils, which is taken into account in this simulation by partially filling the noise correlation matrix (diagonally) with the noise coupling measured from the 32ch sets. In conclusion, the virtual 256ch head coil shows great acceleration performance. Together with 2DCAIPIRINHA, a peak acceleration factor of 28 can be achieved, showing great potential for high density receive arrays combined with fast parallel imaging acquisition methods.


The Netherlands Organization for Scientific Research (NWO), grant number: ALW-834.14.004


[1] K.P. Pruessmann et al. 1999 Magn Reson Med 42:952–962

[2] F.A. Breuer et al. 2005 Magn Reson Med 53:684–691

[3] M. Barth et al. 2015 Magn Reson Med [Epub ahead of print] 26-08-2015

[4] K. Ugurbil et al. 2013 NeuroImage 80: 80–104

[5] N. Petridou et al. 2013 NMR Biomed 26: 65–73

[6] F.A. Breuer et al. 2006 Magn Reson Med 55:549–556


Figure 1: An illustration of the development process of the 256ch coil. Currently, the 2x16ch coil arrays are widely in use (left). Next to that 64 elements (middle) of the 256ch coil (right) are constructed and are in testing and optimization phase.

Figure 2: 2x16ch coil arrays are shifted to 8 different coil positions on the head, creating a virtual 256ch coil. The other side of the brain is covered simultaneously by the second 16ch array. At each position a reference scan was made.

Figure 3: Example of the different possible undersampling patterns using 2DSENSE (black) and 2DCAIPIRINHA (red). Note that the 2DCAIPIRINHA method is very similar to the 2DSENSE method, however it contains an extra voxel shift in its sampling patterns to distribute aliasing more homogenously over an image.

Figure 4: g-factor maps for the standard head coil (top) and the virtual 256ch coil (bottom), using 2DSENSE (left) and 2DCAIPIRINHA (right) undersampling patterns, Rtotal (RFHxRAP_CAIPIshift). The maximum achievable acceleration, for g-factors <2, is 9(SENSE 3x3) and 12(CAIPIRINHA 4x3_FH2) for the head coil. For 256ch: 24(SENSE 4x6) and 28(CAIPIRINHA 4x7_AP3).

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