1.3Partial Volume Effects Due to Low Resolution

Metabolite concentrations from lesions smaller than the grid resolution will be affected by the concentration in the surrounding tissue and changes may be masked, i.e., choline concentrations will be underestimated while NAA concentrations will be overestimated. Also, special care should be taken when nominal matrix size (i.e., the number of phase encoding steps in each direction before extrapolation by adding zeroes) is rather small (<16 × 16), since this causes signiÞcant blurring due to the poor point spread function leading to ÒbleedingÓ of signal intensity between adjacent voxels. Signal bleeding also becomes signiÞcant when the grid resolution (resolution after adding zeroes) exceeds the nominal resolution signiÞcantly; thus, digital resolution enhancement by more than a factor of 2 should be avoided. Partial volume effects should deÞnitely be taken into account when the absolute quantiÞcation of spectroscopic data is considered.

1.4Evaluation of Metabolite Concentrations

Spectroscopic data reßect the concentration of a subset of brain metabolites. The accuracy of the related information depends crucially on the approach used for data quantiÞcation. Generally, the spectrum is evaluated by measuring the area under the metabolite signals. This can be done either by numerical integration of metabolite peaks in phased (real) or magnitude (modulus) spectra or by using more sophisticated tools which basically perform a nonlinear Þt of the entire spectrum. Depending on the tool, the Þt is performed in the time domain using constraints (jMRUI (Naressi et al. 2001; Vanhamme et al. 1997), an ofßine tool which requires export of the data to an external workstation) or frequency domain (most processing tools which are provided by the vendor and operate on the scanner console; LCModel (Provencher 1993), ofßine data evaluation). All methods report signal intensities which are proportional to the respective metabolite concentration in the volume of interest (VOI). Conversion of the hard- ware-speciÞc units to absolute concentrations (i.e., mMol/l) requires a set of correction factors which depend on the used pulse sequence, hardware parameters like signal ampliÞcation and coil loading, relaxation times (T1, T2) of the metabolites, as well as fractions of GM, WM, and CSF in the VOI (partial volume effects). Hardware parameters can be corrected for by using either the so-called phantom replacement method (Michaelis et al. 1993) or scaling relative to the water signal (Barker et al. 1993). The water must be recorded in a separate measurement, either as a separate MRSI data set which has to be corrected for T1 and T2 relaxations or by an imaging sequence with proton density contrast. Relaxation terms for metabolite signals from regular (healthy) tissue are available in

several publications, but they may be changed in tumor tissue (TrŠber et al. 2004; Hattingen et al. 2007; Isobe et al. 2002). Further, the presence of contrast agents can lead to a decrease of signal intensity between 10 and 15 % (Smith et al. 2000; Sijens et al. 1997; Murphy et al. 2002). Correction for partial volume effects requires at least one more additional imaging sequence and further calculations. A rather quick method which only takes into account the CSF fraction was described by Horsk‡ et al. (2002), while analysis of GM, WM, and CSF fraction requires tissue segmentation which can be very time consuming. Therefore, a thorough data evaluation in terms of absolute concentrations should be reserved for research studies aimed at metabolic differences between different groups of patients (e.g., different tumor entities) and longitudinal studies, while for diagnostic purposes a semiquantitative approach just comparing metabolite intensities from tumor tissue and normal-appearing tissue from the contralateral side may be sufÞcient. Immediate information of the extent of change of metabolite concentrations or their ratios can be visualized in the MRSI metabolite map (Figs. 1 and 3). However, one should be aware of artifacts (see below).

1.5Artifacts in Metabolite Maps

Spectroscopic imaging data are frequently visualized as metabolite maps, i.e., for each metabolite the concentration is displayed either as a grayscale image or as a color-coded overlay on an anatomical image. While this provides the most intuitive picture of the results, special care should be taken when interpreting these maps. Local Þeld inhomogeneities due to calciÞcation or deposits of paramagnetic hemosiderin which occur in the vicinity of areas with former bleeding can shift and distort signals, spoiling the data analysis algorithm applied to obtain the signal intensities for the speciÞc metabolites. Especially for voxels crucial for diagnostic decision (e.g., with highest choline), the choline hot spots or Cho/NAA signal intensities require an inspection of the entire spectrum to exclude excessive line broadening and baseline distortions which usually prohibit a reasonable signal analysis by integration or Þtting routines, leading to false values for metabolite concentrations or their ratios. Intense lipid signals originating from necrotic areas as well as from fat deposits in the skull base, soft tissue, and orbit can also distort the baseline. These lipid signals can even appear in the spectra and should not be misinterpreted as tumor necrosis (Fig. 4a). An excellent description how to judge the quality of the spectra is given by Kreis (2004). Rapidly growing tumor cells typically have marked increase of glycolytic rates even if oxygen is abundant (Warburg effect (Warburg 1956), see below), and lactate is considered as a marker for increased glycolysis. Lactate in tumor tissue coincides with the lipid signal but