Chemical Shift Imaging (CSI)

How is chemical shift imaging performed?  

2D CSI ¹H brain MRS

Chemical Shift Imaging (CSI), also known as MR Spectroscopic Imaging (MRSI), refers to a family of multi-voxel techniques that utilize phase-encoding in whole or in part for spatial localization. The multi-voxel spectra can be obtained in 1D (a column of voxels), 2D (a plane of voxels), or 3D (block of voxels) mode. 

​CSI phase-encoding can be combined with any type of excitation and signal generation method. In the first example (below left) a 3D CSI sequence with FID sampling is illustrated. Here the entire sensitive volume of the coil is first excited by a non-selective RF-pulse with data sampling after each phase-encoding step. The second example (below right) illustrates a 2D CSI sequence based on PRESS with sampling of a spin-echo. Here spectra are simultaneously obtained from a 2D slab of voxels whose thickness is determined by the slice-select gradient (Gz) and whose in-plane dimensions are determined by the field-of-view and number of phase-encoding steps. 

---

3D CSI sequence using nonselective volume excitation and stepped phase-encoding gradients along all three axes (common method for ³¹P MRS)

2D-PRESS CSI sequence with slice-selective excitation pulses in 3 planes with stepped phase-encoding gradients along 2 axes (common method for ¹H brain MRS). Diagram has been simplified by leaving out preparatory water/fat saturation modules and crusher gradients)

​The relative advantages and disadvantages of multi-voxel CSI methods have been described more completely in a prior Q&A.  In brief, CSI offers both a larger total coverage area and higher spatial resolution than single-voxel methods. The potential for a wide coverage area allows evaluation of large, heterogenous lesions, while smaller size of individual voxels is advantageous for small or irregularly shaped lesions. 

The major disadvantages of multi-voxel CSI include: 1) Longer set-up and imaging time; 2) difficulties obtaining homogenous shim over the entire region; 3) lower signal-to-noise and spectral quality for individual voxels; 4) spectral contamination from adjacent voxels.

Of all these limitations, imaging time constraints are perhaps the most critical. To complete an imaging cycle for spatial localization, an entire set of phase-encoding gradients must all be stepped through in a nested fashion. Using a TR of 2.0 sec, an [8x8x8] 3D-CSI study would therefore require 2x8x8x8 = 1024 sec, or approximately 17 min to perform — a value at or beyond the tolerance limit of most patients for holding still during an MR exam. Various strategies for reducing scan time are presented in the Advanced Discussion, together with other information on the CSI technique.

Advanced Discussion (show/hide)»

Reducing Image Acquisition Time in Multi-voxel MRS

Multi-slice CSI

Multi-slice spectroscopy uses the same strategy employed in conventional multi-slice imaging. At long TR's considerable "dead time" exists after acquisition of spectra from the first slice. This allows additional 2D-CSI slabs to be excited while the first slice is recovering. 

Multi-slice CSI using spin-echo sampling. For long TR's "dead time" after acquiring data from first slab (orange) allows additional slabs (blue and pink) to be excited.

The number of possible slices is limited by the time for the additional RF-pulses and separate and data acquisition. For TR ≤ 1500 ms, multislice CSI is not usually possible. Extending TR to 2000 generally allows 2 slices to be obtained, while at TR = 2500 ms, 3 slices can be acquired.

Multi-slice CSI is advantageous for spectral volumes that have large in-plane dimensions and relatively small through-plane dimensions. Through-plane coverage can also be extended by placing small gaps between the slices. Because the slices are independently acquired, slice definition is superior to 3D-CSI methods. Independence between slices can be even further improved by slice interleaving (this requires 2 separate acquisitions, however). 

Turbo Spectroscopic Imaging (TSI)

Turbo spectroscopic imaging (TSI)

Turbo spectroscopic imaging (TSI) uses a strategy similar to turbo (fast) spin-echo imaging. Instead of generating just a single echo after excitation, multiple 180°-pulses are used to produce additional echoes. Each of these echoes is separately phase-encoded. Imaging time is reduced by the turbo-factor (number of echoes) for the sequence, which can be as high as 4-6 for long TR/TE sequences but is typically only 2-3.

---

Because the later echoes suffer T2-decay, their data is used to fill the outer regions of k-space while the primary echo is used to fill the center.  TSI is most useful for metabolites that have long T2-values, such as ​NAA and Choline. The need for recording multiple echoes limits data acquisition time and reduces spectral resolution. J-coupling modulations as a function of echo times may also become problematic, with spatial ringing and blurring of lactate maps, for example.  The quality of water suppression also changes along the course of the echo train affecting spectral quantification. An additional limitation concerns increased RF-power deposition in tissues secondary to the use of multiple 180°-pulses. 

Echo-Planar Spectroscopic Imaging (EPSI)

Echo-Planar Spectroscopic Imaging (EPSI)

Echo-planar spectroscopic imaging (EPSI) is an exception to the general principle that readout gradients must not be used during collection of the MRS signal. In EPSI an oscillating gradient collects a single line of k-space at different time points, thus allowing simultaneous acquisition of both spatial and spectral information. Phase-encoding along the other axes permits 2D or 3D localization. 

---

This rapid traversal of k-space EPSI results in significantly shortened examination times. However, it is not widely available commercially and demands high performance gradients. Ghosting errors can occur due to phase imbalances between the positive and negative gradient lobes. Spectral bandwidth and signal-to-noise are also slightly inferior to conventional CSI methods.

A variation of EPSI using oscillating gradients along two axes is known as Spiral-MRSI. Spiral-MRSI repeatedly samples k-space along a curved trajectory from the center of k-space outward. It holds some advantages compared to EPSI in that the center of k-space is repeatedly sampled, reducing phase errors.

Parallel Imaging and Multi-Band

Parallel-encoding techniques (SENSE, GRAPPA) are commonly applied to MRSI studies of all types. Acceleration (R-) factors of 2-3 are possible in each dimension reducing imaging times by factors of 4-6 (depending on available signal-to-noise). Parallel encoding is generally performed in two in-plane directions, but with appropriate coil geometry 3D acceleration is possible.

A major problem with parallel imaging in MRS concerns artifacts relating to incomplete unfolding of the spectroscopic signal. Specifically, lipids outside the field-of-view (e.g. from the scalp) may contaminate the central spectra. Outer volume and chemical shift suppression techniques are therefore required when parallel imaging is considered.

Simultaneous Multislice Imaging (Multi-Band) provides another approach to reducing scan time. Although not yet widely used in spectroscopy, it holds great promise for the near future.

Limited k-space sampling

Another strategy commonly used to reduce acquisition time in MR spectroscopy is to sample only a limited region of k-space. Typically a central circular or elliptical area is chosen for data collection, while the outer 25% of k-space points are filled with zeros. Data in the central region of k-space can be further weighted to improve performance.

Although imaging time is shortened commensurate with the degree of reduced k-space filling, this increases both spectral line width and spectral contamination. A common method to reduce these problems is to filter the measurement data prior to Fourier transformation. A popular choice is the Hamming filter that causes the measurement data at the edge of a voxel to decay gradually and symmetrically. By doing so, the effective voxel size may be 50% or more of the nominal voxel size.

---

References
     Keevil SF. Spatial localization in nuclear magnetic resonance spectroscopy. Phys Med Biol 2006; 51:R579-636.
     Öz G, Alger JR, Barker PB, et al. Clinical proton MR spectroscopy in central nervous system disorders. Radiology 2014; 270:658-679.
     Pohmann R, von Kienlin M, Haase A. Theoretical evaluation and comparison of fast chemical shift imaging methods. J Magn Reson 1997; 129:145-160.
     Posse S, Otazo R, Dager SR, Alger J. MR spectroscopic imaging: principles and recent advances. J Magn Reson Imaging 2013; 37:1301-1325.
     Skotch A, Jiru F, Bunke J. Spectroscopic imaging: basic principles. Eur J Radiol 2008; 67:230-239.
     Zhu H, Barker PB. MR spectroscopy and spectroscopic imaging of the brain. Methods Mol Biol 2011; 711:203-226

---

Related Questions
     How do you choose between a single and multi-voxel technique?
     If frequency-encoding cannot be used to determine spatial position, how do you localize an MRS signal?  

---

←  Previous Question

Next Question  →

↑ Complete List of Questions ↑