Improved banding removal for high resolution bSSFP imaging of the inner ear using SENSE
Eliana NessAiver1, Dan Zhu1, Ari Meir Blitz2, and Daniel Herzka1

1Biomedical Engineering, The Johns Hopkins University School of Medicine, Baltimore, MD, United States, 2Radiology, The Johns Hopkins Hospital, Baltimore, MD, United States

Synopsis

High resolution imaging of the inner ear is a desirable tool for the diagnosis and treatment of inner ear pathologies. In particular, balanced steady state free precession images have a good balance of high SNR, fast imaging times, and novel tissue contrast which yields satisfactory differentiation of inner ear structures. However, it suffers from banding artifacts in areas of field inhomogeneity. While common clinical practice is to combine two images that are 180º phase cycled from one another, which shifts the bands to different locations in each image, this yields only partial mitigation of the artifact. This study applies parallel imaging techniques to acquire four phase cycled images in a similar timeframe to the original two-image acquisition, in order to produce a combined volume with superior banding removal at little to no extra cost over current clinical practice.

Purpose

To improve CISS artifact removal in high resolution inner ear imaging with minimal cost in increased scan time.

Introduction

High resolution imaging of the inner ear is imperative for the diagnosis and treatment of inner ear pathologies. MRI provides excellent contrast for the complex tissues in this region of the brain. In particular, balanced steady state free precession (bSSFP) images yield a good balance of high SNR and short imaging times, coupled with novel tissue contrast which allows for satisfactory differentiation of inner ear structures. However, due to the many air-tissue boundaries in this complicated region, local field inhomogeneities lead to the characteristic banding artifacts often seen in bSSFP images. These bands occur wherever the magnetic field inhomogeneities cause an off-resonance spin precession of ± π/TR. The high resolution 3D scans which are crucial for this application necessitate increased TR time, which compounds the banding problem by increasing spin precession per TR and thus shrinking the distance between successive dark bands. Common clinical practice is to combine two images that are 180º phase cycled from one another, which shifts the bands to different locations in each image, a technique named constructive Interference in steady-state (CISS) [1]. This combination mitigates the severity of the banding but does not eliminate them entirely. This study applies parallel imaging techniques to acquire four phase cycled images in a similar timeframe to the original two-image acquisition, to produce a combined volume with superior banding removal with minimal increase in scan time relative to current clinical standards.

Motivation

Theoretical signal magnitudes for a variety of realistic imaging and tissue parameters were simulated using Mathematica (Wolfram Research, Inc.). For a given set of parameters, the full range of possible values repeats every n*π magnetic field off-resonance precession. Although the shape of this profile is highly parameter dependent, for most tissue and imaging applications there is an approximately smooth region of high signal with a sharp dip - representing the band artifacts - that appears every ± π. (Figure 1) A ratio of the difference between maximum and minimum signal in this profile to the signal average can be called the “ripple” associated with those parameters [2]. The goal of debanding is to achieve a ripple of near 0%. [2-4] Simulations of the profiles resulting from a two image root sum of squares (RSS) combination for values approximating gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) [5] were compared with those of a four-image combination and improvement of 28-80% was seen.

Methods

Four 3D phase cycled images (RF offset ΙΈ = 0º, 90º, 180º, 270º) were acquired using a 3T imaging system (Achieva, Philips Medical Systems,The Netherlands) and a 31 channel head coil. Imaging parameters were: TE = 3.68 ms, TR = 7.36 ms, α = 35º, resolution = 0.49 x 0.49 x 0.5mm, with the imaging region positioned obliquely to cover the relevant structures of the inner ear. Scan time was 12 min 34s. The data were retrospectively under-sampled by a factor of 2, and an auto calibration region of 32 x 32 lines was acquired for calculation of sensitivity profiles. The four phase cycled images were then combined using the root-sum-squares (RSS)method and qualitatively compared to a fully sampled two-image RSS combined volume.

Results and Discussion

While SENSE reconstruction of the four phase cycled images suffered a slight SNR loss, as expected, a comparison to the fully sampled image showed this loss to be very tolerable - likely due to the high g-factor made possible by the 31 coil channels. The combined images regain some of that SNR and a comparison with the fully sampled two-image volume shows a noticeable reduction in band severity (Figure 2), including the areas of the internal auditory canals (Figure 3) In areas with extreme T1/T2 values, as seen in the fluid of the orbit, the banding removal is clearly incomplete but still benefits from reduced banding severity.

Conclusion

This work demonstrates a clinically feasible method for improving the appearance of CISS banding artifacts without significant increases in scan time. Conjugate gradient SENSE can effectively be employed to acquire 4 phase cycled images, instead of the usual 2, in order to further mitigate off resonance bands. Higher acceleration factors in more than one dimension must be explored to determine the appropriate balance between reduction of scan time and reduction of banding artifacts.

Acknowledgements

No acknowledgement found.

References

[1] Casselman JW, Kuhweide R, Deimling M, Ampe W, Dehaene I, Meeus L. Constructive interference in steady state-3DFT MR imaging of the inner ear and cerebellopontine angle. AJNR. American journal of neuroradiology. 14(1):47-57. 1993.

[2] Bangerter NK, Hargreaves BA, Vasanawala SS, Pauly JM, Gold GE, Nishimura DG. Analysis of multiple-acquisition SSFP. Magnetic resonance in medicine. 51(5):1038-47. 2004. [3] Xiang Q-S, Hoff MN. Banding artifact remo val for bSSFP imaging with an elliptical signal model. Magnetic resonance in medicine. 71:927–933. 2014.

[4] Quist B, Hargreaves BA, Daniel BL, Saranathan M. Balanced SSFP Dixon imaging with banding-artifact reduction at 3 Tesla. Magnetic resonance in medicine. 74(3):706-15. 2015

[5] Bieri O, Scheffler K. On the origin of apparent low tissue signals in balanced SSFP. Magnetic resonance in medicine. 56(5):1067-74. 2006.

Figures

Figure 1.

The magnitude profile of two phase cycled images compared to the RSS 4 phase cycles combination, using T1/T2 parameters similar to gray matter, α = 45 degrees.


Figure 2.

Comparison of heavily banded original bSSFP image to a 2-image fully sampled RSS combination and a 4-image, SENSE reconstructed RSS combination.

Gray box indicates region in Figure 3.


Figure 3. Band indicated by arrow in the first two images is gone in the 4-image combination.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1336