Two NSA or not two NSA: does perforator artery detection in white matter benefit from signal averaging?
Lennart Geurts1, Sander Brinkhof1, Peter R. Luijten1, and Jaco J.M. Zwanenburg1

1Radiology, UMCU, Utrecht, Netherlands

Synopsis

Because cerebral perforating arteries have sub-millimeter diameters and slow blood flow velocities, their blood flow velocity and pulsatility measurements are challenging and limited by noise and partial volume effects. Our previously reported acquisition method used two signal averages (NSA) to increase the signal-to-noise ratio (SNR). We show that decreasing NSA, and thereby reducing scan time by half, has little effect on vessel detection. The NSA=1 coefficients of repeatability (CoR) found in this study are similar to previously published NSA=2 CoR`s. Subject motion and small vessel size likely play together to cause a sub-optimal benefit from increased imaging time.

Introduction

Recently, it was shown that blood flow velocity and pulsatility can be measured in perforating arteries of the white matter (WM) at 7T [1]. Because these perforators have very small diameters and slow blood flow velocities, these measurements are challenging and limited by noise and partial volume effects. The previously reported acquisition method used two signal averages (NSA) to increase the signal-to-noise ratio (SNR). However, we found that subject motion between scans can be large compared to the size of WM perforators. This raised doubt on how much SNR is really gained by signal averaging. Besides that, the identification of arteries was based on both velocity and magnitude [1]. Simulations showed that more vessels are detectable by velocity than by magnitude criteria (data not shown). The aim of this study was to investigate the benefit from signal averages and magnitude selection on the detection of WM perforators.

Methods

Seven healthy subjects were included. A single slice 2D phase contrast (PC) MRI acquisition was adapted from a previously published version and performed on a 7T MRI system (Philips Healthcare) with 32 channel receive head coil (Nova Medical) [1]. Before and after PC acquisition, a T1-weighted (T1w) 3D TFE image was acquired for WM segmentation (Table 1). To compare NSA=1 with NSA=2 acquisitions and investigate the repeatability of NSA=1 acquisitions concurrently, NSA was reduced to 1 and the number of dynamics was increased to 2. In post processing, the real and imaginary parts were computed from the modulus and phase images of the two dynamics, and seperately combined to create a surrogate NSA=2 image.

The published method for perforator identification [1] was automated with Matlab (The Mathworks Inc.). Vessels in the acquisition with the least number of identified vessels were paired to the nearest vessel in the repeated measurement. A pair was approved only if their distance was smaller than 2 mm. The number of vessels identified without matching were compared between the first NSA=1 and the NSA=2 images and tested using a paired two-sample t-test. The SNR of vessels in the NSA=2 image were plotted against their matched vessels in a NSA=1 image, where a slope of √2 would signify full benefit of signal averages. The performance of magnitude based selection over velocity based selection was assessed by calculating coefficients of repeatability (CoR) for the mean velocity (Vmean) and pulsatility index (PI) in the NSA=1 images. The standard deviation (SD) of noise was approximated with the SD over the cardiac cycle. The criterium for velocity selection was significant flow velocity (|v|>2 SDVnoise) in every cardiac phase. The criterium for magnitude selection was significant contrast (Mvessel-Mtissue>2 SDMnoise) in the mean magnitude image.

Results

Slightly more vessels were detected in the NSA=2 images compared to the first NSA=1 images (52 versus 46 vessels, p = 0.046), (Table 2). For the vessels that were detected in both images, the velocity SNR increased with a factor of 1.2 (Figure 1 and Table 2).

Selecting vessels based on magnitude did not improve the CoR`s of Vmean or PI, in fact they worsened slightly (Table 2). Velocity selection alone detected 35% more vessels (30 versus 46, p = 0.006), which were discarded by the magnitude criterium.

Discussion

With a doubling of scan time due to NSA=2, only an increase of 12% of the number of identified vessels was observed. While image SNR is expected to increase with √2, the SNR of matched vessels increased only by a factor of 1.2. The CoR`s found for NSA=1 images are slightly better than the CoR`s for NSA=2 images previously published [1], however in the previous study subjects were repositioned between scans. These results confirm our suspicion that subject motion compared to the size of the imaged vessels counteracts the benefits of averaging in this particular application.

Selecting vessels based on magnitude did not improve the CoR`s of Vmean or PI, while excluding 35% of vessels detected by velocity. This shows that magnitude selection imposes stricter boundaries than necessary, as previously suggested by simulations [1]. Because velocity has to be significant for each cardiac phase, the probability of identification by chance is 0.05n cardiac phases, which rapidly decays to very small probabilities.

Conclusion

We conclude that the increased scan time with NSA=2 is not as beneficial as expected, we suspect this is due to subject motion. Furthermore, selecting perforators based on magnitude data does not improve repeatability of the measurements. With these results in mind we suggest not to use signal averaging or magnitude selection when imaging WM perforators with 2D PC-MRI.

Acknowledgements

This work was supported by the European Research Council, ERC grant agreement n°337333.

References

[1] Bouvy WH, Geurts LJ, Kuijf HJ, et al. Assessment of blood flow velocity and pulsatility in cerebral perforating arteries with 7-T quantitative flow MRI. NMR Biomed. 2015 Apr27.

Figures







Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1835