Elsevier

Magnetic Resonance Imaging

Volume 20, Issue 10, December 2002, Pages 743-757
Magnetic Resonance Imaging

Truncation artifact reduction in spectroscopic imaging using a dual-density spiral k-space trajectory

A preliminary version of this work was presented at the 9th annual meeting of ISMRM in Glasgow, Scotland.
https://doi.org/10.1016/S0730-725X(02)00608-2Get rights and content

Abstract

Truncation artifacts arise in magnetic resonance spectroscopic imaging (MRSI) of the human brain due to limited coverage of k-space necessitated by low SNR of metabolite signal and limited scanning time. In proton MRSI of the head, intense extra-cranial lipid signals “bleed” into brain regions, thereby contaminating signals of metabolites therein. This work presents a data acquisition strategy for reducing truncation artifact based on extended k-space coverage achieved with a dual-SNR strategy. Using the fact that the SNR in k-space increases monotonically with sampling density, dual-SNR is achieved in an efficient manner with a dual-density spiral k-space trajectory that permits a smooth transition from high density to low density. The technique is demonstrated to be effective in reducing “bleeding” of extra-cranial lipid signals while preserving the SNR of metabolites in the brain.

Introduction

In proton spectroscopic imaging of the human brain, truncation artifacts are particularly problematic since the intense extra-cranial lipid signal from subcutaneous fat can detrimentally contaminate spectroscopic signals in the brain. Various methods have been developed to reduce truncation artifacts in CSI. One common approach is to apodize in the k-space [1], [2], [3]. K-space apodization is equivalent to smoothing in the spatial domain. It reduces truncation artifact by reducing side-lobes of the point spread function resulting from the truncation of k-space. While this reduction is reasonably effective, the main lobe of the point spread function resulting from most apodization functions, such as the Hamming window, is broadened (by a factor of 2 at least) compared to a rectangular window of the same length [4]. To maintain desired spatial resolution when apodization is used, an expanded k-space coverage is needed. Apodization in post-processing sacrifices the SNR because it is not optimal in terms of SNR per unit measurement time. To avoid the reduction in SNR, apodization is also implemented in data acquisition [5]. An efficient implementation of such an approach was demonstrated using variable density spiral trajectory [3], where a theoretical reduction of 18 dB in magnitude for the first sidelobe is possible, without sacrificing SNR or spatial resolution, in addition to the lipid suppression achieved by inversion recovery.

Other techniques have utilized structural information from anatomic images to augment the reconstruction of spectroscopic images to reduce truncation artifacts [6], [7], [8], [9]. SLIM [6] solves for compartmental spectra assuming that the sample consists of a number of homogeneous compartments which can be identified from anatomic images. Data extrapolation techniques [7], [8], [9], [10] also use edge information from anatomic images. Although these methods reduce the truncation artifact to a large extent, their effectiveness is limited by the inherent assumptions made by these methods. Another approach is to use spatially selective excitation techniques to excite the spins well within the brain and away from the extra-cranial lipid [11], [12], [13], [14]. This approach, however, limits the extent of the available FOV, making it impossible to study the entire cerebral cortex. A method based on spatial saturation utilizes 8 pulses to saturate spins in the scalp in an octagonal fashion [15]. Although the saturation technique can be effective for single slices, it is cumbersome to use and may be limited by the RF deposition at higher fields. Taking advantage of the relatively short T1 of the lipids, others have used an inversion pulse with a proper delay to null the lipid signal [16], [17]. This approach can be robust but it assumes that dominant lipid resonances have a similar T1 values, and the inversion pulse also introduces a reduction in the signal of the metabolite peaks.

Realizing that substantial signal contamination due to truncation arises mainly from areas having strong signals, a strategy for reducing contamination from these regions based on extended k-space coverage using dual-SNR was developed [18]. In this strategy, termed as dual-SNR sampling in the rest of this work, a larger region of the k-space was covered, with the low k-space points sampled with a high SNR (i.e. more averaging) and the high k-space points sampled with a low SNR (i.e. less averaging). The high k-space data, being low in SNR, contained sufficient information for the high intensity areas, which contribute most significantly to the truncation artifact. Hence, an extended region of the k-space was covered with a slight increase in measurement time. A new reconstruction approach, selective reconstruction, was also devised to reconstruct spectroscopic images from data thus sampled [18]. The technique was demonstrated to be robust with one-dimensional implementation using variable averaging. In principle, this sampling strategy can be applied to multi-dimensional CSI. However, using variable averaging alone, the additional time required for measuring the high k-space portion of the data in multiple dimensions is prohibitively long. Sampling the high k-space with reduced repetition time [19] in addition to reduced data averaging improved this technique to some extent.

To achieve variable coverage in 2-D spectroscopic imaging, a hybrid technique combining phase-encoded CSI and fast echo-planar spectroscopic imaging was developed [20]. With its rapid sampling capability, EPSI was used to cover a large portion in k-space efficiently. Using the low SNR high k-space data sampled by EPSI for intense lipid signals only, and combining it with high SNR low k-space data provided by phase-encoded CSI, a hybrid reconstruction was made which resulted in reduction of truncation artifacts. Subsequently, Ebel et al. [21] implemented the technique using a dual-EPSI technique that acquires the central k-space using multiple averaged EPSI data. Since both methods combine data from two separate acquisitions, data consistency between these acquisitions is critical. However, data consistency is often difficult to achieve in practice and mismatches in data lead to degraded results.

For CSI with multiple dimensions using the dual-SNR approach, it is desirable to have a method that collects all the data in a single acquisition so that data matching is not required. In this paper, we describe such a technique, which is based on dual-density sampling of the k-space using spiral trajectories. A higher sampling density corresponds to a higher SNR as a result of effective averaging of the sampled points during reconstruction. In keeping with the dual-SNR strategy, the k-space trajectory is designed with high density at low k-space and lower density at high k-space. Spiral trajectory was first used for spectroscopic imaging several years ago [22], and variable density spiral was introduced recently to achieve apodization during acquisition for spectroscopic imaging [3]. For the present work, spiral trajectory is used to achieve a smooth transition from high-density to low-density sampling in k-space during the same acquisition.

Section snippets

Theory

In in vivo spectroscopic imaging only a limited region in k-space is covered due to low metabolite concentrations and time constraints. For coverage of k-space within a circle with radius |k⃗|max, the reconstructed spatial distribution can be expressed as M̃0(r)=∫ ∫ M0(k)circ(|k|/|k|max)exp(j2πrk)d2k where the integration is over the entire k-space and circ(|k⃗|/|k⃗|max) is circular sampling window. The circular sampling window leads to a convolution of the true image with a point spread

K-space trajectory design strategy

The objective of the design was to expand k-space coverage with the desired dual-density without increasing the data acquisition time substantially. To achieve this goal the k-space trajectory was designed according to the desired k-space coverage illustrated in Fig. 2. Low k-space, the extent of which is determined by the desired spatial resolution for metabolite signals, was sampled using the highest possible density achievable under system and design constraints with the aim of achieving

Processing

The acquired data was processed in three steps:

  • 1.

    The sampled k-space data were temporally aligned into isotemporal k-space planes.

  • 2.

    Spatial reconstruction was performed by gridding the spiral k-space data to a Cartesian grid and applying selective reconstruction to k-space at each time point.

  • 3.

    Spectral reconstruction was carried out by taking a 1D FT of the FID for each pixel in the spatial image.

Results

Fig. 8 shows the effectiveness of the temporal alignment procedure for off-resonance correction in spatial reconstruction. Spatial images shown in the figure were reconstructed from the data at t = TE before spectral reconstruction. These images correspond to an integration over the entire spectral bandwidth and contain mainly lipid signals because the data were obtained with water suppression. Without temporal interpolation (Fig. 8a), due to off-resonance artifact, the extra-cranial lipid

Discussion

The present method extends the k-space coverage with a dual-density spiral acquisition, which is a more efficient method for implementing the dual-SNR acquisition strategy [18] compared to previous implementations. In the dual-SNR strategy, the fact that lipid signals have high intensities is utilized such that the outer k-space data is acquired with a low SNR strategy, which corresponds to low-density in the present method. With the dual-density acquisition, selective reconstruction allows us

Summary

This paper describes a data acquisition method for truncation artifact reduction in spectroscopic imaging using the principle of dual-SNR sampling. The approach uses dual-density spiral sampling. The technique was demonstrated experimentally to be successful in significantly reducing “bleeding” of extra-cranial lipid signals while preserving the SNR of the metabolites in the brain.

Acknowledgements

The authors thank Dr. Yasser M. Kadah, Dr. Gregory J. Metzger, and Dr. Xiaodong Zhang for providing the pulse sequences and reconstruction software, which were used as a starting point of this work.

References (40)

  • D.L. Parker et al.

    Gibbs artifact removal in magnetic resonance imaging

    Med Phys

    (1987)
  • X. Hu et al.

    SLIMspectral localization by imaging

    Magn Reson Med

    (1988)
  • X. Hu et al.

    Technique for reduction of truncation artifacts in chemical shift images

    IEEE Trans Med Imaging

    (1991)
  • M.A.S. Patel et al.

    A robust algorithm for reduction of truncation artifact in chemical shift images

    IEEE Trans Med Imaging

    (1993)
  • M.S. Patel et al.

    Selective data extrapolation for chemical shift imaging

    Proc Intl Soc Magn Reson Med

    (1994)
  • Z.P. Liang et al.

    A generalised series approach to MR spectroscopic imaging

    IEEE Trans Med Imaging

    (1991)
  • P.A. Bottomley

    Spatial localization in NMR spectroscopy in vivo

    Ann NY Acad Sci

    (1987)
  • J. Frahm et al.

    Localized high-resolution proton NMR spectroscopy using stimulated echoesinitial applications to human brain in vivo

    Magn Reson Med

    (1989)
  • J.H. Duijn et al.

    Human brain infarctionproton MR spectroscopy

    Radiology

    (1992)
  • P.R. Luyten et al.

    Metabolic imaging of patients with intracranial tumorsH-1 MR spectroscopic imaging and PET

    Radiology

    (1990)
  • Cited by (0)

    Work supported by NIH (grants RO1MH55346, RR08079, R03MH59245).

    View full text