External

Contralateral hemisphere

Simple, no coil loading errors

Extra scan time required, sensitive to B0 or B! errors, can't be used in global/diffuse diseases, or in midline structures

External Standard

Reference concentration exactly known, no coil loading errors

Extra scan time required, sensitive to B0 or B! errors, may cause susceptibility effects, can't be used with all pulse sequences

Phantom Replacement

Reference concentration exactly known, no extra patient scan time required

Coil loading correction required, requires stable system

2.1. Units and Tissue Compartmentalization

There are several different units which can be used to express in vivo tissue metabolite concentrations. From Eq. [3], it can be seen that the in vivo concentration will be expressed in the same units as the reference concentration, since the ratio of signal intensities is unitless. Therefore, for instance, in the external reference approach, if the external reference concentration is measured in millimolar (i.e. millimoles solute per liter solution) then the in vivo concentration will be returned in the same units (equivalent to millimoles per liter brain volume). Traditionally, metabolite concentrations in tissue determined by conventional biochemical techniques are more often expressed in units such as millimoles per kg tissue wet or dry weight. To convert millimolar to millimoles per kg wet weight, it is necessary to divide by the tissue density (1.05 kg/liter for normal brain). To convert to millimoles per dry weight, it is necessary to know the wet/dry weight ratio. Note that these approaches assume that the entire volume of sample is occupied by solid brain tissue; in practice, this may well not be the case, since the large voxel sizes used for in vivo MRS often contain appreciable cerebrospinal fluid (CSF) contamination.

CSF normally contains much lower levels of metabolites than brain; therefore CSF contamination (without appropriate correction methods) will lead to underestimation of brain metabolite concentrations. Fortunately, several methods now exist for estimating voxel CSF content and applying appropriate correction factors. One method makes use of measuring the voxel water signal as a function of multiple different echo times (17). Since CSF has a much longer T2 than brain water, bi-exponential fitting of the echo signal versus time can estimate the relative fractions of brain and CSF water. This information can then be used to estimate true tissue metabolite concentrations. Alternatively, MR imaging based segmentation methods can also be used to estimate brain and CSF volumes within the MRS voxel. One particularly simple approach is to use long echo time (for instance, TE = 500 msec) fast-spin echo (FSE) MRI, which essentially only contains signal from the long T2 CSF, the brain signal having decayed to the noise level at this TE (18). More sophisticated approaches use multiple FSE scans with different contrast in combination with appropriate multi-spectral post-processing methods, in order to estimate not only CSF content, but also fractional gray and white matter content as well (19). An example of brain and CSF segmentation using rapid FSE imaging is shown in Figure 1.

Figure 1. An example of multi-spectral brain and CSF segmentation for use with MRSI. Top row: rapid fastspin echo sequences are acquired with different degrees of T2-weighting (proton density (TE 20 msec), T2-weighting (TE 100 msec), CSF only (TE 500 msec), and a T1-weighted (TI 500, TE 20 msec) . Through the use of region of interest measurements of individual tissue types and subsequent "Eigenimage" filtering (50), these images can be processed to generate maps (bottom row) corresponding to pure CSF, gray matter and white matter. This information can be used to correct metabolite concentrations determined by MRS(I) for partial volume effects.

Figure 1. An example of multi-spectral brain and CSF segmentation for use with MRSI. Top row: rapid fastspin echo sequences are acquired with different degrees of T2-weighting (proton density (TE 20 msec), T2-weighting (TE 100 msec), CSF only (TE 500 msec), and a T1-weighted (TI 500, TE 20 msec) . Through the use of region of interest measurements of individual tissue types and subsequent "Eigenimage" filtering (50), these images can be processed to generate maps (bottom row) corresponding to pure CSF, gray matter and white matter. This information can be used to correct metabolite concentrations determined by MRS(I) for partial volume effects.

2.2. Determination of Peak Areas

Equation [1] indicates that the signal induced in the receiver coil is directly proportional to the NAA concentration. From the Fourier transformation, it can be shown that the area under the curve in the spectrum (frequency domain) is equal to the amplitude of the first point of the time domain signal ("free induction decay", or FID). Therefore, quantitative analysis either requires the determination of peak areas in the frequency domain, or direct time-domain analysis to estimate the amplitude of the first data point of the different frequency components that comprise the FID.

A variety of methods have been developed for frequency domain analysis (3,6,20). The simplest approach is simply to use numerical integration, although this method will work poorly when spectral overlap occurs. More sophisticated analysis methods include parametric curve-fitting routines, using various model functions (e.g. Lorentzian, Gaussian, Voigt, others (21,22)) and fitting algorithms (simplex, non-linear least squares, etc..). The most sophisticated method, and one which is becoming widely used, is the so-called linear combination model ("LCModel") that fits the spectrum as a linear combination of the pure compound spectra known to exist in the spectrum (4). The LCModel is particularly attractive for several reasons; (a) it makes full use of all the resonances in the molecule, (b) it is fully automated and user independent, including both baseline and phase correction, (c) with appropriate calibration data, it can give absolute metabolite concentrations, and an estimate of the uncertainty. An example of a LCModel analysis of a short echo time spectrum recorded at 3 Tesla in the normal human brain is shown in Figure 2.

Figure 2. An example of the "LCModel" analysis of a short echo proton spectrum from the anterior cingulated gyrus of the normal human brain recorded at 3 Tesla. The original data (black line) is overlaid with the results of the linear combination analysis (red line), with the difference (noise) plotted at the top of the display. Metabolite concentrations (determined using brain water as an internal intensity reference), their % standard deviations, and other imformation is tabulated on the right hand side.

Figure 2. An example of the "LCModel" analysis of a short echo proton spectrum from the anterior cingulated gyrus of the normal human brain recorded at 3 Tesla. The original data (black line) is overlaid with the results of the linear combination analysis (red line), with the difference (noise) plotted at the top of the display. Metabolite concentrations (determined using brain water as an internal intensity reference), their % standard deviations, and other imformation is tabulated on the right hand side.

A variety of time-domain fitting methods have been proposed, usually using parametric models based on the exponentially decaying oscillations (10,23). One of the main perceived advantages of time-domain fitting is that it can avoid artifacts that may be induced by Fourier transformation of incomplete of partially corrupted time-domain data (e.g. missing or incorrect data points at the beginning of the FID). While this is true, since the properties of the Fourier transform are well known, it is usually possible to account for these problems using appropriate functions in frequency domain fitting as well, so that the choice of method may depend as much on numerical convenience, rather than any fundamental difference in approach.

2.3. Literature Review of Human Brain NAA Concentrations

There have been numerous papers over the last 12 years on the determination of NAA concentrations in different regions of the human brain (4,6,10,12,15,24-34). Generally, NAA concentrations have been found to be in the range of 7-13 mM in the adult human brain, with relatively small regional variations. It should be remembered that the signal at 2.02 ppm has components from both NAA and N-acetyl aspartyl glutamate (NAAG), which is typically present at about 1 mM concentration (35). Since these compounds are difficult to resolve, in most papers the quoted "NAA" concentration usually is in fact the sum of NAA and NAAG.

NAA concentrations are low at birth, and increase rapidly over the first one to two years of life (36). NAA appears to relatively stable in adults, with some investigators finding no change with age, while others find slight decreases in NAA in elderly subjects (6,37,38).

Was this article helpful?

0 0

Post a comment