N -Acetylaspartate.

NAA accounts for the majority of the NAA resonance at 2.01 ppm; this signal is the most prominent one in normal adult brain proton MRS and is used as a reference for determination of chemical shift. The NAA signal also contains contributions from other N -acetyl groups, such as N -acetylaspartyl glutamate (NAAG), N -acetylated glycoproteins, and amino acid residues in peptides. NAA is second only to Glu as the most abundant free amino acid in the normal adult brain. The function of this amino acid is not fully understood despite its early discovery in 1956 by Tallan.

From animal studies, NAA is believed to be involved in coenzyme A (CoA) interactions and in lipogenesis within the brain. Specifically, such studies suggest that NAA is synthesized in the mitochondria from aspartate and acetyl CoA and transported into the cytosol where it is converted by aspartoacylase into aspartate and acetate. Although NAA is widely regarded as a nonspecific neuronal marker, it has also been detected in immature oligodendrocytes and astrocyte progenitor cells.

Normal absolute concentrations of NAA in the adult brain are generally in the range of 8 to 9 mmol/kg, although regional and age-related variations in NAA concentration have been noted by Kreis and others. In normal adults, NAA concentrations in cortical gray matter are higher than those in white matter; in infants the concentrations in gray and white matter are similar (highly active lipid synthesis in immature white matter accounts for this difference from the adult pattern). NAA concentrations are decreased in many brain disorders, resulting in neuronal and/or axonal loss, such as in neurodegenerative diseases, stroke, brain tumors, epilepsy, and multiple sclerosis, but are increased in Canavan’s disease.

Creatine.

The main Cr signal is present at 3.03 ppm and demonstrates major contributions from methyl protons of creatine and phosphocreatine as well as minor contributions from GABA, lysine, and glutathione. A second, usually smaller Cr signal is seen at 3.94 ppm. Cr is probably involved in maintenance of energy-dependent systems in brain cells by serving as a reserve for high-energy phosphates in neurons and as a buffer in cellular adenosine triphosphate/diphosphate (ATP-ADP) reservoirs. Thus the Cr signal is an indirect indicator of brain intracellular energy stores.

The Cr signal is often used as an internal reference standard for characterizing other metabolite signal intensities because it tends to be relatively constant in each tissue type in normal brain; however, this is not always true in abnormal brain tissue, particularly in areas of necrosis. Cr concentrations in the brain are relatively high, with progressive increases noted from white matter to gray matter to cerebellum. Kreis and coworkers noted a mean absolute Cr concentration in normal adult brains of 7.49 ± 0.12 mmol/kg on the basis of a sample of 10 normal subjects, whereas Michaelis and colleagues reported a similar value of 5.3 mmol/kg. Total Cr values tend to be abnormally reduced in brain tumors, particularly malignant ones.

Choline.

The Cho resonance is present at 3.2 ppm and is attributable to trimethyl ammonium residues of free Cho as well as phosphocholine, phosphatidylcholine, and glycerophosphocholine. This signal reflects cell membrane synthesis and degradation. Thus all processes resulting in hypercellularity (e.g., primary brain neoplasms or gliosis) or myelin breakdown (demyelinating diseases) lead to locally increased Cho concentration, whereas hypomyelinating diseases result in decreased Cho levels. Kreis and coworkers have reported a mean absolute Cho concentration in normal adult brain tissue of 1.32 ± 0.07 mmol/kg. Michaelis and others have reported a similar value of 1.6 mmol/kg.

Myoinositol.

MI produces two signals noted at 3.56 ppm and 4.06 ppm. MI is the major component of the signal at 3.56 ppm, although contributions from MI-monophosphate and glycine are also present. MI is believed to be a glial marker because it is present primarily in glial cells and is absent in neurons.

A role in osmotic regulation of the brain has been attributed to MI. In addition, MI may represent both a storage pool for membrane phosphoinositides involved in synaptic transmission and a precursor of glucuronic acid, which is involved in cellular detoxification. A derivative, MI-1,4,5-triphosphate, may act as a second messenger of intracellular calcium-mobilizing hormones.

The mean absolute concentration of MI in normal brain tissues obtained in the Kreis series was 6.56 ± 0.43 mmol/kg. MI concentrations are abnormally increased in patients with demyelinating diseases, Alzheimer’s disease, and low-grade brain tumors.

Lactate.

Lac resonance is identified as a doublet (splitting into two distinct resonant signals separated by 7 Hz), produced by magnetic field interactions among adjacent protons (referred to as J-coupling ) centered at 1.32 ppm. A second Lac signal is present at 4.1 ppm but tends to be inconspicuous on spectra obtained with water suppression owing to its proximity to the water signal. Because Lac levels in normal brain tissue are absent or extremely low (<0.5 mmol/L), they are essentially undetectable on normal spectra. The presence of a visible Lac signal constitutes a nonspecific indicator of cellular anaerobic glycolysis, which may be seen with brain neoplasms, infarcts, hypoxia, metabolic disorders, or seizures. Lac may also accumulate within cysts or foci of necrosis.

Changing the TE using a point-resolved spectroscopy (PRESS) sequence enables confirmation of the presence of an abnormal Lac doublet and differentiation from lipid signals; at TE = 20 and 272 milliseconds the Lac doublet projects above baseline, and at TE = 136 milliseconds it is inverted below the baseline. In encephalopathic neonates, however, a doublet very similar to Lac may be seen at 1.15 ppm, which corresponds to propan-1,2-diol and may easily be mistaken for Lac; propan-1,2-diol is a solvent commonly used for administration of anticonvulsant medications to neonates.

Glutamate, Glutamine, and GABA.

The glutamate and glutamine (“Glx”) and GABA signals are a complex set of resonances noted between 2.1 and 2.5 ppm and consist of Glu, Gln, and GABA components. Glu is an excitatory neurotransmitter involved in neuronal mitochondrial metabolism and is the most abundant amino acid in the human brain. Gln is involved in both cellular detoxification and regulation of neurotransmitter activities. GABA is a product of decarboxylation of Glu via glutamic acid decarboxylase and serves as an inhibitory neurotransmitter. Abnormalities in this signal complex have been noted in schizophrenia and epilepsy.

Lipids and Proteins.

Lipids produce multiple resonances, the most important of which are noted at 0.8 to 0.9 and 1.2 to 1.3 ppm because of methyl and methylene protons of fatty acids, respectively. Membrane lipids are not usually identified unless very short TEs are employed, because they have very short TRs and are not normally observed. Although artifactual accentuation of these signals may be seen with voxel contamination by adjacent subcutaneous fat, they can be eliminated by using outer volume suppression saturation pulses. High-grade gliomas (HGGs), meningiomas, demyelinating processes, necrotic foci, and inborn errors of metabolism can show the presence of lipids owing to the presence of lipid droplets within necrotic tumors, loosening of the lipid-rich myelin sheath, and breakdown of plasma membrane structures of cells.

Noninvasive Histologic Grading of Gliomas.

Although it is possible to clearly distinguish glial neoplasms from normal brain tissue by MRS, controversy exists regarding the reliability of MRS in distinguishing among different histologic grades of astrocytomas and other brain tumors. For example, Shimizu and associates demonstrated that through semiquantitation of MRS signal intensities as a ratio to that of an external reference, it was possible to predict tumor malignancy ; higher-grade brain tumors were associated with higher Cho/reference values and lower NAA/reference values in their series. This, in conjunction with findings from Tamiya’s group regarding high correlation between Cho/NAA and Cho/Cr ratios and cellular proliferative activity, provides substantial evidence that proton MRS may be useful in the prediction of tumor grade.

In a series involving children, linear discriminant analysis and proton MRS demonstrated an 83% success rate in establishing the correct diagnosis of histologic grade of brain tumors. In a series of 27 patients with biopsy-confirmed brain tumors, Meyerand and colleagues showed that the combination of Lac/water and Cho/water ratios obtained from regions of contrast-enhancing brain tumor (with normalization of each metabolite signal area to the area of the unsuppressed water signal) permitted differentiation of low-grade astrocytomas from anaplastic astrocytomas and GBs.

However, not all studies have confirmed the ability of proton MRS to differentiate between different tumor grades. For example, Burtscher and colleagues have described a series of 26 intracranial tumors in which MRS allowed differentiation of infiltrative processes from circumscribed lesions but did not allow differentiation of different types of lesions within each category. In a multicenter study involving 86 cases of glial tumors, Negendank and coauthors showed that each type and grade of tumor was a metabolically heterogeneous group with significant overlap in spectral NAA/Cr and Cho/Cr ratios with tumors in other categories; all tumors demonstrated abnormally decreased NAA/Cr and increased Cho/Cr ratios with respect to normal brain parenchyma. Some of the reasons for these findings in early studies (i.e., before the year 2000) regarding the inability of proton MRS to reliably differentiate various tumor grades have included inconsistencies in quality of spectroscopic data among different institutions, use of only long TE spectra, and use of only single-voxel technique; many of these problematic issues have been addressed in more recent studies.

Studies have suggested that advanced physiologic MRI methods in conjunction with MRS may be more accurate for determination of tumor grade than MRS or anatomic MRI alone. For example, Law and colleagues demonstrated that relative cerebral blood volume (rCBV) perfusion MRI combined with proton MRS metabolite ratios data improved both sensitivity and positive predictive value (PPV) for the determination of glioma grade over structural MRI alone. The combination of rCBV, Cho/Cr, and Cho/NAA resulted in sensitivity, specificity, PPV, and negative predictive value of 93.3%, 60%, 87.5%, and 75%, respectively, compared to 72.5%, 65%, 86.1%, and 44.1%, respectively, with conventional MRI alone. The low specificity was due in part to the fact that high Cho levels were found in some low-grade lesions.

El-Sherbeny and coworkers recently showed that inclusion of MI/Cr ratios in the MRS analysis in conjunction with the MRI information obtained from conventional and advanced MRI techniques added additional diagnostic value to the differentiation of the grade of the brain tumor. A number of studies have shown that MI/Cr peak area ratios over 0.6 are highly characteristic of low-grade gliomas. With the addition of MI/Cr ratios to MRS data, the sensitivity and specificity of the MRI/MRS paradigm employed to determine the grade of the lesion was 91% and 90.5%, respectively, compared to the study of Law et al., which found a 93.3% sensitivity and only a 60% specificity. The study of El-Sherbeny examined 93 patients with histologically confirmed brain tumors (31% low grade and 69% high grade).

More recently, Caulo and colleagues reported a quantitative multiparametric MRI evaluation of 110 patients with either low- or high-grade gliomas (rCBV, apparent diffusion coefficient [ADC]), MRS (Cho/Cr and Cho/NAA), and conventional MRI that incorporated the analysis of four heterogeneous MRI tumor regions. They found that taking into account the heterogeneous nature of the gliomas and combining the studies with both advanced and conventional MR techniques significantly improved the discrimination between low- and high-grade brain gliomas (sensitivity of 84.4%, specificity of 100%, and an AUC of 0.95).

A meta-analysis by Hollingworth and colleagues published in 2006 explored the overall clinical utility of MRS in the characterization of brain tumors and shed some important light on the potential role of MRS in the differentiation of various tumor types. It concluded that MRS was accurate in differentiating high- and low-grade tumors. Specifically, based on receiver operating characteristic (ROC) curve analysis of multiple studies examining sensitivity and specificity of proton MRS for differentiating high- from low-grade tumors, the study found excellent values for this purpose in the AUC in multiple studies. In studies by Devos and Lukas and colleagues, AUC values of 94% and 96% in studies using long and short TE, respectively, were obtained. In Hollingworth’s meta-analysis, two studies also showed excellent ability of proton MRS to differentiate between high- and low-grade gliomas; specifically, Herminghaus and coworkers showed in their study of 90 patients that the sensitivity and specificity of proton MRS for the differentiation of high-grade from low-grade tumors were 95% and 93%, respectively. In addition, Astrakas and associates demonstrated in their cohort of 66 patients that by using a combination of MRS parameters (namely Cho/Cr and Lac/Cr; threshold value = 1.8), an AUC value of 96% was obtained for the sensitivity and a specificity of 88% for differentiation of high-grade from low-grade tumors.

Astrocytomas typically demonstrate reduced NAA levels, moderately reduced Cr levels, and elevated Cho levels compared to normal brain parenchyma ( Fig. 17-3 and Table 17-1 ). These absolute reductions result in abnormally low NAA/Cr ratios and elevated Cho/Cr ratios (see Table 17-1 ). NAA levels in astrocytomas are reduced to 40% to 70% of the levels seen in normal brain parenchyma. Furthermore, Cho has been reported to be substantially elevated in more malignant astrocytomas, and this may be secondary to increased cellularity and cell membrane synthesis. There is evidence to suggest that the elevation of Cho is proportional to the degree of tumor malignancy. However, highly malignant primary brain tumors with extensive necrosis may actually demonstrate decreased levels of Cho. Elevated Cho levels are seen more consistently in anaplastic astrocytomas (which do not demonstrate histologic evidence of necrosis) than in glioblastoma multiforme (GBM), and ependymomas display higher Cho/Cr ratios than those noted for astrocytomas in general.

A, Grade II astrocytoma. A representative single-volume proton MR spectrum of a histologically proven low-grade glioma using a STEAM sequence for the TE = 20 msec spectrum and a PRESS sequence for the TE = 135 msec spectrum. The TR used was 1500 msec in both the TE = 20 and TE = 135 msec studies. Note the intensity of the myoinositol (3.56 ppm) and choline (3.2 ppm) resonances in the low-grade lesion compared to the higher-grade gliomas. B, Glioblastoma multiforme (GBM). Representative single-volume proton MR spectrum of a histologically proven GBM using a PRESS sequence at TE = 135 msec and TR = 1500 msec. C, TE = 30 msec proton spectrum of GBM. Note the lower myoinositol resonance compared with the grade II lesion and the elevated choline signal at both TE = 30 and 135 msec.

Lac levels may be elevated in some astrocytomas as well. However, a higher incidence of Lac in more aggressive or higher-grade astrocytomas is controversial because Lac may also be present in benign pilocytic astrocytomas. It is thought that Lac may be present across the entire spectrum of grades of astrocytomas because Lac may arise not only from hypoxia developing within the tumor itself as a result of disruption of normal vascular networks but also from necrosis or cysts within the tumor. Lac levels may be elevated in all cysts regardless of etiology.

The presence of Lac has been attributed to (1) the extent of anaerobic glycolysis, (2) the efficiency of electron transport, and (3) the rate of washout from tumor tissue. In highly vascular tumors, Lac may be rapidly removed from the tumor as a result of increased blood flow; therefore, even in high-grade vascular tumors, Lac may not be present in the MR spectra.

The value of Lac identification following radiation therapy or surgery is controversial, but the Lac concentration may be proportional to the original tumor size. Lac has been reported following radiation therapy and surgery, including stereotactic biopsy, although postsurgical porencephaly may lead to artifactual increases in Lac levels because cerebrospinal fluid (CSF) is known to be rich in Lac.

The presence of elevated lipid levels has also been used to differentiate low-grade from high-grade neoplasms. Some studies suggest that mobile lipids tend to be present in higher-grade neoplasms, with highest levels noted in GB; however, although high levels of mobile lipids may be specific for anaplastic astrocytoma or GB, their absence in MR spectra does not exclude the possibility of such high histologic grades. Lipids may originate from tumor cells within high-grade astrocytomas, macrophages along the tumor periphery, or areas of necrosis. Lipids may be present as a result of microscopic cellular necrosis, which may not be apparent on conventional MRIs.

There are inherent limitations of histopathologic grading of gliomas that make it a somewhat imperfect gold standard for tumor assessment. For example, there is the intrinsic sampling error associated with current stereotactic biopsy, which is frequently based on enhancing components of tumor, when HGGs are known to infiltrate nonenhancing parenchyma following the vascular channels of white matter tracts. Furthermore, in adults the biological behavior of low-grade astrocytomas with identical histologic grades is heterogeneous, with more than half of low-grade tumors eventually recurring as or evolving into more aggressive tumors. Some investigators believe that differences in MRS ratios of various metabolites, like differences in glucose metabolism in positron emission tomography (PET), may be of prognostic significance even if they are not of diagnostic value in such cases of low-grade neoplasms. Physiologic MRI, such as proton MRS, not only allows whole lesion evaluation but also allows for evaluation of metabolic abnormalities beyond the tumor margins.

Proliferative Glioma Population: Cho/NAA Ratio, ADC, and rCBV Parameters.

At present, the major role of MRSI in the surgical treatment planning for HGG is in guiding the neurosurgeon to regions that show high metabolic activity on biopsy (regions with elevated Cho levels and low NAA levels relative to normal brain tissue). In a study conducted by Gupta and colleagues, 18 glioma patients were preoperatively examined with H1-MRSI and DWI along with conventional MR studies. This study was conducted to correlate quantitative presurgical H1-MRSI and DWI results with quantitative histopathologic characteristics of resected tumor tissue. The primary hypotheses to be tested were: “(1) glioma choline correlates with cell density, (2) glioma apparent water diffusion coefficient (ADC) correlates inversely with cell density, and (3) glioma choline signal correlates with cell proliferative index” as measured by MIB-1 immunostaining. The tumor-to–contralateral normalized Cho signal ratio (nCho = Cho _{tumor voxel} /Cho _{contralateral normal volume} ) and the ADC from resected tumor regions were determined from the preoperative MRI studies. The results showed that cell density correlated linearly with nCho but inversely with ADC values. No correlation was found between nCho and cell proliferative capacity.

However, in a similar study conducted by McKnight et al., using a normalized Cho/NAA (CNI) and Cho/Cr (CCrI) index instead of nCho index, they found that the normalized Cho/NAA but not the Cho/Cr ratios correlated significantly with the proliferation index of the tumors as determined by MIB-1 immunostaining ( P < 0.001) and cell density measurements ( P < 0.001) of the biopsy-obtained voxel sample. This finding suggests that voxels containing elevated normalized Cho/NAA ratios have the highest probability of being a region that has “high cell density, high proliferative fraction, and/or high ratio of cell proliferation to cell death.”

As discussed previously, for surgical, chemotherapeutic, and radiation therapy clinical patient treatment decisions it is important to be able to delineate and identify, as confidently as possible, the various heterogeneous regions of the tumor so that the correct treatment plans can be applied to these regions. Thus the more noninvasive parameters or changes in the values of the parameters one can use to identify the tumor ROIs, the more accurate and more reliable the tumor map will be, which can be used to devise the most effective treatment paradigm for the patient. In the case of proliferative tumor regions, addition of the rCBV maps obtained from PWI studies can be used to further identify and delineate the major proliferating regions of the tumor. Advanced MRI techniques have been developed that allow noninvasive study of tumor vascularity. Dynamic susceptibility-weighted contrast-enhanced perfusion MRI, used widely in clinical practice, relies on the T2* signal intensity change that occurs when the contrast agent passes through the tissues. This change in signal intensity allows calculation of the rCBV, and this parameter has been correlated with tumor vascularity. Studies have shown that gliomas that have higher rCBV also have higher mitotic activity, as judged from the histopathologic determination of the number of mitotic cells found in tissue sections from stereotactic biopsies obtained from the perfusion MRI study. More recently, Price et al. has shown that the rCBV in HGGs linearly correlated with cellular proliferation, as judged by MIB-1 immunostaining and histopathologic analysis of resected tumor section analyzed from preoperative PWI; r = 0.66, P < 001). Thus, using the correct thresholds for the Cho/NAA ratios, ADC, and rCBV for proliferating tumor regions, and coregistering this information with the anatomic MRIs can be used to improve the delineation and more accurately assess tumor regions having high proliferative capacity.

Hypoxic/Pseudopalisading Tumor Population: MRS Metabolic Profile and the Role of PET Imaging.

Two other cell populations within the tumor that can lead to recurrence and/or resistance to therapy also need to be evaluated: (1) the quiescent pseudopalisading migratory hypoxic tumor population adjacent to necrotic regions and (2) the active infiltrating but low mitotically active tumor population migrating from the main tumor body. If neither of these populations are either resected or targeted for further treatment, they will lead to repopulation/recurrence and spread into other areas of the brain. In the case of the hypoxic/pseudopalisading population, this population is normally quiescent and if not surgically resected will be more resistant to standard radiotherapy treatments normally applied following surgery; this will lead to repopulation and spread of the tumor. In addition to their intrinsic resistance to radiation damage, this hypoxic tumor population is often refractory to the cytotoxic actions of conventional chemotherapeutic agents. Thus it is important to try to identify the location of this population for the design of augmented surgical, radiotherapy, and chemotherapy treatment plans.

MRSI and conventional MRI, in conjunction with positron emission tomography (PET) of [F-18] fluoromisonidazole uptake (F18-FMISO) may be of use to map out the regions containing the hypoxic/pseudopalisading tumor cells. F18-FMISO is a fluoronitroimadazole compound that was one of the first PET agents used for the pretherapy detection of hypoxic tumor populations in patients with tumors outside the CNS. With the advent of PET-MR clinical scanners, PET studies can be readily registered into the MRI and MRSI studies to produce anatomic maps combined with functional maps showing the distribution of proliferating and quiescent tumor populations in hypoxic regions. F18-FMISO PET can be used to image tumor hypoxia, because tumor cells show increased F18-FMISO uptake, and once taken up, it is exclusively trapped in hypoxic tumor cells because of the reduced cellular oxygen environment of these cells. F18-FMISO is cleared in normoxic tumor cells.

The presence of hypoxia appears to be a common feature of HGGs that accumulate F18-FMISO. The hypoxic regions found in HGG have been shown to be associated with the regions adjacent to areas of necrosis where the hypoxic pseudopalisading cells reside. Recently Bruehlmeier et al. found that uptake and retention of F18-FMISO was consistently found in the periphery of the lesion adjacent to necrotic regions of the HGG.

MRSI can add support to the F18-FMISO PET findings because it has been shown that regions showing low Cho and NAA levels relative to normal brain normally indicate necrosis, astrogliosis, or regions with macrophage infiltration. However, these regions that appear to be necrotic may also contain quiescent pseudopalisading migratory tumor cells adjacent to necrotic centers, especially if they show spectra (obtained at TE = 30 milliseconds) that have elevated Cho/NAA and Cho/Cr ratios, presence of lipid/Lac resonances, but overall lower levels of metabolites relative to normal tissue ( Fig. 17-8 ). This spectral profile in fact may represent the most aggressive tumor population within HGGs and the major population responsible for the resistance and repopulation of tumors.

A and B, MRSI volume showing necrosis (Lip/Lac) and tumor spectrum (elevated Cho/Cr levels).

In a study conducted by Martin et al. to determine the efficacy of MRSI in guiding biopsy, they found that in 17 out of 21 patients who had histologically confirmed biopsy tumor samples, the Cho levels were elevated relative to normal levels. However, in four patients with low Cho levels similar to those observed in necrotic regions, histologic evaluation of the biopsy sampled showed tumor. The representative MRSI spectra of one of these patients ( Fig. 17-9 ) shows relatively elevated Cho/Cr and Cho/NAA ratios and an elevated lipid/Lac resonance, suggestive of a necrotic region with actively proliferating tumor cells.

This patient had prominent lipid resonance and otherwise reduced metabolic levels on both SVS and MRSI. Although a modest degree of Cho elevation is evident on both spectra, it is difficult to definitively infer the presence of tumor. However, on biopsy this area was confirmed to be recurrent glioblastoma.

(From Martin AJ, et al: Preliminary assessment of turbo spectroscopic imaging for targeting in brain biopsy. AJNR Am J Neuroradiol 22:959–968, 2001.)

Infiltrative Glioma Population: Cho/Cr and Cho/NAA Ratio Profiles, rCBV, and Fractional Anisotropy.

When targeting tumor biopsies or defining tumor treatment volumes, the infiltrating tumor population must be taken into account. To determine the usefulness of MRSI in assessing the degree of tumor infiltration, Ganslandt et al. performed MRSI studies on 7 patients with untreated grade 2 and 3 gliomas. The MRSI data sets were fused with the 3D MRI data sets and integrated into a frameless stereotactic system for image-guided interactive neuronavigation surgery. Tissue samples were obtained from three regions based on the Cho/NAA ratios found in the MRSI study: (1) normal-appearing brain region, (2) zone between spectroscopically pathologic and suspicious for infiltrating tumor volumes, and (3) maximum spectroscopically pathologic tumor volumes. The implementation of the MRSI data set into the frameless stereotactic system was successful in all cases, and stereotactic biopsies were obtained from the MRSI-defined regions. In this study, a relationship between the tumor cell density and Cho/NAA metabolic changes were found (60%-100% in the maximum pathologic areas to 5%-15% in the border zones suspicious for infiltrating tumor). Another important finding of this study was that the tumor areas defined by the metabolic maps, which were confirmed histopathologically, exceeded the T2-weighted abnormal signal change in all cases by 6% to 32%. However, in four patients, biopsies obtained with normal Cho/NAA ratios showed the presence of tumor infiltration. It was suggested that the false-negative finding in these patients was due to the low resolution of MRSI with respect to the glioma borders. Another explanation could be that the Cho/NAA ratio reflects primarily the proliferative capacity of the tumor cells and not necessarily the migratory capacity. Thus in infiltrative tumor zones, Cho/NAA levels may be normal or only slightly elevated because tumor cells are not dividing and/or displacing or destroying normal neural cells. Highly infiltrative cells normally only proliferate at vascular branch points, therefore in infiltrative white matter tumor regions the Cho/NAA ratios may not be elevated. However, the Cho/Cr ratios may be elevated. This is shown by Wright to represent an infiltrative spectral tumor pattern.

The infiltrating tumor pattern of gliomas is closely associated with the expression of matrix metalloproteinase 2 (MMP-2). Invasion of glioma cells involves the attachment of invading cells to the extracellular matrix (ECM), disruption of the ECM components, and subsequent penetration of the invading cell into adjacent brain structures. This is accomplished in part by the secretion of MMPs by the invading glioma cell to break down the ECM barrier, which impedes the infiltrating tumor cells from extending. Zhang et al. found a significant correlation between the Cho/NAA and Cho/Cr ratios and MMP-2 expression; it appears that the higher the ratios, the more invasive/infiltrative the tumor. This is in agreement with the image-guided MRSI brain tumor biopsy studies of Croteau et al. Histologic analyses of tissue samples obtained from the MRSI-guided biopsy sites found that using contralateral Cr and Cho levels for normalization or ipsilateral NAA appeared to correlate with the degree of tumor infiltration. They found that the degree of tumor infiltration increased with increasing Cho/NAA ratio and Cho/Cr ratios (minimal level of tumor infiltration had Cho/NAA = 1.09 ± 0.12 and Cho/nCr = 1.62 ± 0.09; maximum level of infiltration had Cho/NAA = 2.01 ± 0.18 and Cho/nCr = 1.98 ± 0.18).

Using the Cho/Cr and Cho/NAA threshold ratios defined in the Croteau study for minimal level of tumor infiltration may be useful in delineating the more metabolic and proliferative tumor populations ( Fig. 17-10 ) from the less proliferative infiltrating tumor ones ( Figs. 17-11 and 17-12 ). Figure 17-11 shows a tumor that appears to be infiltrating into areas beyond the Gd T1-weighted enhancing regions of tumor. The Cho/Cr peak area ratios in these areas are 1.5 or greater. The region outside the enhancing area also shows Cho/NAA peak area ratios of 1.0 or less. This suggests that in areas outside the Gd-enhancing or T2-weighted hyperintense areas, in which Cho/Cr ratios greater than 1.5 and Cho/NAA levels of less than 1.0 occur (see Fig. 17-12 ), there may be regions containing a mixture of normal cells and infiltrating tumor cells with lower rates of proliferation. In the area of enhancement or extending just beyond this area, where the MRSI Cho/Cr ratios are greater than 1.5 and the Cho/NAA ratios are greater than 1.0, are areas of active tumor proliferation and infiltration (see Fig. 17-11 ). The spectral pattern observed for the infiltrative tumor process is consistent with the infiltrative tumor spectral pattern found by Wright et al. ( Fig. 17-13 ).

MRSI of proliferative glioma pattern. Note both Cho/Cr and Cho/NAA patterns are similar. (Threshold ratios: Cho/Cr > 1.5 and Cho/NAA > 1.0.)

Infiltrative tumor pattern. Note the proliferative tumor population (with Cho/Cr ratios > 1.5 and Cho/NAA > 1.0) is located primarily in enhancing region of tumor, whereas the predominantly migratory population (with elevated Cho/Cr ratio > 1.5 and Cho/NAA < 1.0) is outside the enhancing region.

Spectral pattern of (left) normal and (right) infiltrative tumor process (non–Gd-enhancing region).

A-C, The three independent components (ICs) generated from the 1517 voxels, normalized as described by Wright and coworkers. D, The median of all absorption spectra, with IC ₁ as the component with the highest coefficient ( g ₁ , necrotic tumor core), illustrated with the 25% and 75% quartiles above and below (dashed lines). E and F, The median and quartiles of spectra for the highest-coefficient components of IC ₂ and IC ₃ , respectively ( g ₂ [infiltrative tumor growth] and g ³ [normal brain]). D-F, Key metabolite resonances, indicated as Lip and lac (lipids and lactate), Cho (cholines), Cr (creatine), and NAA ( N -acetylaspartate); g ₂ ( E ), the median and quartiles of spectra for the highest-coefficient IC ₂ (mixture of normal and infiltrative tumor pattern); g ₃ ( F ), the median and quartiles of spectra for the highest coefficient IC ₃ (normal brain pattern).

(From Wright AJ, et al: Pattern recognition of MRSI data shows region of glioma growth that agrees with DTI markers of brain tumor infiltration. Magn Reson Med 62:1646–1651, 2009.)

Cho/NAA ratios may be low (<1.0) in areas of tumor infiltration, because the normal neuronal cells are not being eliminated in these regions by actively proliferating tumor cells, as would be expected in regions in and peripheral to the Gd-enhancing area where Cho/NAA ratios are greater than 1.0. Figure 17-10 shows an HGG that appears to be well circumscribed. The Cho/Cr ratio is greater than 1.5 only in or adjacent to the enhancing region where the Cho/NAA peak area ratio is greater than 1.0. It does not appear to extend beyond this region as defined by the Cho/Cr peak area ratio of 1.5 or greater for tumor threshold (namely, where the Cho/Cr is > 1.5, it is highly probable tumor is present in this voxel). This observed Cho/Cr and Cho/NAA ratio pattern may suggest that the tumor is expanding via a more proliferative rather than an infiltrative pathway; this may be due to the more cortical location of this lesion, which predominantly affects white matter.

Di Constanza et al. used an MR multiparametric approach to assess the extent and malignancy of cerebral gliomas. Thirty-one patients (21 high grade and 10 low grade) with gliomas underwent conventional MRI, H1-MRSI, DWI, and PWI studies prior to surgery and histologic examination of resected tissue. ROIs with high Cho (>1.3; Cho _volume on tumor side/Cho _{contralateral normal volume} ) and or abnormal Cho/NAA greater than 1.0 (normalized to ratio found in contralateral volume) were considered tumor or infiltrated tumor on the basis of previous neuropathologic correlation studies.

From this multiparametric study of high-grade lesions, three tumor regions could be delineated: (1) Gd-enhancing tumor region (margin and mass), (2) perienhancing abnormal region containing tumor and edema, and (3) normal-appearing perienhancing region containing infiltrated tumor and normal brain parenchyma. The high-grade lesions containing the bulk tumor mass and enhancing margins (1) showed very high mean Cho/NAA ratios greater than 4.0 and Cho/Cr greater than 2.0 (normalized relative to contralateral normal region), high mean ADC values (1.24 ± 0.31 × 10 ⁻³ mm ² /sec in bulk tumor vs. 0.77 ± 0.05 × 10 ⁻³ mm ² /sec in normal brain region), and mean elevated rCBV of 3.70 in tumor vs. 0.98 in normal tissue. In the perienhancing abnormal-appearing region containing tumor and edema (2), the normalized mean Cho/NAA was found to be 3.3 and Cho/Cr was 1.76, the mean tumor ADC value was 0.92 ± 0.17 × 10 ⁻³ mm ² /sec, and mean tumor rCBV was 1.68 ± 0.41. In the normal-appearing perienhancing region (3), the normalized mean infiltrated tumor Cho/NAA ratio was 1.55 and Cho/Cr ratio was 1.28, the mean ADC of this region was 0.86 ± 0.08 × 10 ⁻³ mm ² /sec, and the mean rCBV of this region was 1.28 ± 0.16.

The primary change in the ratios of Cho/NAA in the normal-appearing infiltrated tumor region of a high-grade lesion is due to a change in the Cho levels and not to changes in the level of NAA. The mean normalized NAA levels in infiltrated tumor regions were 0.94 ± 0.23 vs. 0.96 ± 0.14 for the levels found in the normal contralateral side. In the bulk tumor mass, the mean NAA level was 0.31 ± 0.11, and in the perienhancing abnormal region (2) the mean NAA level was 0.58 ± 0.22. Thus in both regions 1 and 2 (as per above), the increased changes observed in the metabolite ratios appear to be due to decreased levels of NAA and increased Cho. This suggests higher tumor proliferative capacities in these regions as measured by increased Cho levels and loss of NAA levels. NAA is a neuronal marker, so the loss of NAA suggests either loss of neuronal activity or destruction due to tumor proliferation and invasion into these regions. However, in region 3 the almost normal level of NAA suggests that changes observed are due primarily to infiltrating tumor cells that are not proliferating rapidly. This interpretation is consistent with the changes in rCBV observed in tumor regions (2-3) of this study. Price et al. showed that high-grade brain lesions that have increased mitotic activity also have higher rCBV values. These investigators showed that rCBV correlated significantly with MIB-1 labeling index and not with tumor cell packing, which suggested that increased tumor vascularity as measured by rCBV was more dependent on cellular proliferation than just the number of cells present.

Low-grade lesions demonstrated the same trend with respect to Cho/NAA, Cho/Cr, ADC, and rCBV values in bulk tumor versus infiltrated tumor regions. The bulk of the tumor showed much higher values for all parameters compared to the normal-appearing adjacent infiltrated tumor regions, but in both the bulk tumor and the infiltrated tumor regions, values were found to be higher than in the contralateral normal brain regions.

Tumor infiltration into white matter can induce widening of white matter fiber bundles, which affects the directionality of the motion of water molecules within the white matter fiber bundles. Changes in the directionality of water molecules within white matter tracts can be measured with DTI. By employing at least six diffusion gradient directions, compared to three gradient directions, DWI studies enable not only the macroscopic detection and imaging of the highly anisotropic water diffusion found in white matter bundles but can also detect changes in the anisotropic motion of water (i.e., directionality) within white matter due to tumor invasion, which cause disruption of white matter bundles. A DTI study generates a diffusion tensor D for each image voxel, and from each D voxel value, the magnitude and directionality of water diffusion can be calculated. The magnitude is known as the mean diffusivity (MD) and is mathematically equivalent to the ADC obtained from a conventional DWI study. The directionality is known as fractional anisotropy (FA).

In a 2006 publication, Stadlbauer et al. used DTI to retrospectively correlate changes in FA and MD with the degree of tumor cell infiltration determined histologically. Histopathologic findings of 77 MRI-guided stereotactic biopsies in 20 glioma patients were correlated with the FA and MD values found at the biopsy sites. FA was found to be inversely correlated to percent tumor infiltration and was found to be better than MD for assessment and delineation of tumor infiltration. For low tumor cell numbers as shown in Figure 17-14 , the logarithmic regression lines showed a strong decrease in FA (see Fig. 17-14A ) and an increase in MD (see Fig. 17-14B ) followed by an asymptotic plateauing trend for high tumor cell number. From these findings, the investigators hypothesized that their findings “reflect the infiltrative invasion mechanism of gliomas”—namely, initial enlargement of the extracellular space as the tumor invades, leading to a logarithmic decrease in FA and a logarithmic increase in MD but preserving the normal microstructure of the white matter fiber bundles in the border zone. This hypothesis is consistent with the data of Di Constanza et al. in which the NAA levels remained relatively unchanged while the Cho levels increased by almost 50% in suspected tumor-infiltrated normal-appearing ipsilateral white matter. This lack of change in NAA reflects only the initial phase of tumor infiltration into these regions, which induces only mild disruption of white matter bundles. As more tumor cells infiltrate and proliferate, there is a decrease in the extracellular space (e.g., edema) and a decrease in directionality of diffusion due to tumor-induced derangement and destruction of white matter microstructures, which leads to decreases in NAA along with increased Cho levels.

Scatterplot of fractional anisotropy (FA) and mean diffusivity (MD) vs. percentage TI. A, Linear regression (solid line) of FA versus percentage TI for all biopsies sampled. Regression analysis resulted in R = 0.796 and P < 0.001. Linear regression fit is represented by FA = 0.28-1.5 × 10 ⁻³ percentage TI. B, Linear regression (solid line) of mean diffusivity versus percentage TI for all biopsies sampled. Regression analysis resulted in R = 0.521 and P < 0.001. Linear regression fit is represented by mean diffusivity = 1.17-5.9 × 10 ⁻³ percentage TI. Dotted lines = 95% CIs.

(From Stadlbauer A, et al: Preoperative grading of gliomas by using metabolite quantification with high-spatial-resolution proton MR spectroscopic imaging. Radiology 238:958–969, 2006.)

A number of investigators have proposed the combined use of proton MRSI and DTI to map out the infiltrative regions of gliomas. Wright et al. performed independent component analysis of MRSI data from human gliomas to segment tissues into tumor core, tumor infiltration, and normal brain. DTI analysis was then used to confirm the correctness of the MRSI segmentation pattern (see Fig. 17-13 ). They found that the MD and FA data found for these regions were consistent with previous studies that compared anomalies in isotropic and anisotropic diffusion images to determine regions of potential glioma infiltration ( Table 17-2 ). They showed that coefficients of independent components from their analysis could then be used to create colored images overlaid onto conventional anatomic images to visualize regions of infiltrative tumor growth.

MRSI and DTI both provide markers that represent infiltrative growth, and the two techniques are complementary. DTI can detect abnormalities with high spatial resolution, and MRSI can provide validation of the presence of proliferating and/or infiltrating tumor cells from the metabolic profile. The combination of markers will provide greater certainty in correctly identifying regions of tumor infiltration, which can then be treated appropriately.