MR Spectroscopy in Radiation Injury

SUMMARY: Detecting a new area of contrast enhancement in or in the vicinity of a previously treated brain tumor always causes concern for both the patient and the physician. The question that immediately arises is whether this new lesion is recurrent tumor or a treatment effect. The differentiation of recurrent tumor or progressive tumor from radiation injury after radiation therapy is often a radiologic dilemma regardless the technique used, CT or MR imaging. The purpose of this article was to review the utility of one of the newer MR imaging techniques, MR spectroscopy, to distinguish recurrent tumor from radiation necrosis or radiation injury.

N ew contrast enhancing lesions discovered on routine follow-up brain imaging at or near the site of previously treated primary brain tumor present a diagnostic dilemma. Posttreatment imaging features are often non-specific and the differentiation between recurrent tumor and radiation injury is often difficult. In attempts by investigators to improve local tumor control and the overall clinical outcome and survival for patients with primary brain tumor, new, more aggressive treatment protocols are implemented or tested. These protocols include different schemes of dosages of various chemotherapeutic agents but also different schemes of locally administered high doses of radiation.
Although these new radiation schemes have resulted in improved outcome, they have also been associated with a significant incidence of radiation injury to the brain. It is well documented that there is a relationship between increased survival and increased total dose. 1 The risk of late effects that can lead to devastating functional deficits several months to years after brain irradiation limits the total dose that can safely be administrated to patients. Recent data suggest that progressive dementia occurs in approximately 20%-50% of patients with brain tumor who are long-term survivors after treatment with large-field partial-or whole-brain irradiation. 2 The differentiation of recurrent tumor or progressive tumor from radiation injury after radiation therapy is often a radiologic dilemma, regardless of the technique used, CT or MR imaging. Most of these brain neoplasms have been subjected to radiation and/or chemotherapy, and many of the tumors do not have specific imaging characteristics that will enable the neuroradiologist to discriminate tumor recurrence from the inflammatory or necrotic change that can result from treatment with radiation and/or chemotherapy. Both entities typically demonstrate contrast enhancement. It is, therefore, often the clinical course, a brain biopsy, or imaging over a lengthy follow-up interval that enables the distinction of recurrent tumor from a treatment-related lesion, not the specific imaging itself. 3 A noninvasive tool that could differentiate these entities when a new enhancing lesion is first identified would be invaluable. MR spectroscopy might be well suited for this purpose, provided that spectra of diagnostic quality can be obtained. This noninvasive imaging can be performed by using different techniques, depending on clinical question and localization of the lesion. Spectra can be acquired by using single-voxel spectroscopy (SVS) or multivoxel spectroscopy, also referred as chemical shift imaging (CSI), with both 2D CSI and, lately, 3D CSI acquisitions. Both SVS and multivoxel techniques have lately been used in the evaluation of contrastenhancing brain lesions in patients previously treated for brain neoplasms. The decision of which sequences (SVS versus 2D CSI or 3D CSI) and which parameters (ie, TE and TR) to use depends on the location of the lesion and the choice of the brain metabolites the investigator wants to evaluate.

Other Radiologic and Nuclear Medicine Methods to Discriminate Radiation Injury from Recurrent or Progressive Tumor
MR spectroscopy is not the only method used to differentiate radiation injury from recurrent tumor. Other methods that recently have been used for this purpose but that will not be discussed in detail here are positron-emission tomography (PET), diffusion-weighted imaging (DWI), MR perfusion, and CT perfusion. [4][5][6][7][8][9] Some of these techniques seem to have a lower yield than others and have demonstrated lower sensitivity and specificity than MR spectroscopy, whereas some of them seem very promising.
Previous PET studies have shown that areas of radiation injury have lower glucose metabolism than normal brain tissue because they have lower cellular attenuation. 10 A previous PET review reported the sensitivity of PET to be 80%-90% and the specificity to be 50%-90% in differentiating late-delayed radiation injury from recurrent high-grade glioma. 4 Another study of 15 patients with histopathologically confirmed diagnosis reported that fluorodeoxyglucose-PET (FDG-PET) was only 43% sensitive in distinguishing recurrent tumor from radiation effect and was the least accurate when the lesion volume was Ͻ6 cc. 11 Recent studies using DWI 6 have shown that the apparent diffusion coefficient (ADC) ratios in the contrast-enhancing lesion are lower in recurrent tumor than in radiation-induced injury 6 ; however, other investigators using diffusion tensor imaging (DTI) 7 have demonstrated higher ADC values in the contrast-enhancing part of the lesion in patients with tumor recurrence than in the contrast-enhancing lesion in patients with radiation injury. That study also showed that the ADC ratios in the white matter tracts in the perilesional edema were significantly higher in patients with radiation injury compared with those with recurrent tumor and that the fractional anisotropy (FA) ratios were significantly higher in normal-appearing white matter tracts adjacent to the edema in patients diagnosed with radiation injury compared with those with recurrent tumors. 7 Thallium 201 ( 201 Tl) or 99mTc hexamethylpropyleneamine oxime single-photon emission tomography (HMPAO SPECT) or both [12][13][14] have been reported as useful techniques to discriminate tumor progression and radiation injury. It has been suggested that the combination of lowthallium and low-HMPAO uptake was associated with benign radiation change; whereas increased uptake of either agent or both was associated with recurrent/persistent tumor at biopsy and a poor prognosis. 14 However, false-positive FDG-PET and 201 Tl SPECT findings have been reported with biopsyproved radiation necrosis. 15 In a recent CT perfusion study, the investigators demonstrated significant differences between recurrent tumor and radiation necrosis; patients with recurrent tumor had higher mean normalized cerebral blood volume (nCBV) and normalized cerebral blood flow (nCBF) and shorter normalized mean transit time compared with those with radiation necrosis. 9 Newer MR imaging techniques such as dynamic susceptibility contrast (DSC) perfusion MR imaging have made it possible to obtain hemodynamic measurements such as relative CBV, relative peak height, and percentage of signal-intensity recovery (PSR) within the brain. A recent DSC MR perfusion study of 33 patients treated with stereotactic gamma knife radiosurgery who subsequently developed progressively enlarging regions of contrast enhancement within the radiation field, suggestive of tumor recurrence or radiation necrosis, found that PSR, an imaging indicator of microvascular leakiness, was the most significant variable able to differentiate retrospectively whether a progressively enhancing lesion was due to recurrent metastatic tumor or gamma knifeϪinduced radiation necrosis. 16

Effects of Radiation on Normal Brain
Before reviewing the role of MR spectroscopy in discriminating recurrent tumor from radiation injury, it is important to understand the underlying mechanisms resulting in radiation injury and its symptoms and also to understand what happens to the normal brain when it is irradiated.
The pathophysiology of radiation therapyϪinduced injury to the central nervous system (CNS) is not completely understood. Variables that might play a part include total radiation dose, size of the radiation field and radiation fraction, number and frequency of radiation doses, combination of chemotherapy and radiation therapy, duration of survival, and the age of the patient at time of treatment. 17,18 For example, it is known that Adriamycin (doxorubicin) and methrotrexate potentiate radiation effects and that young children are more susceptible to radiation than adults. Radiation injury can be divided into acute, early-delayed, and late-delayed stages. 4 It is believed that at least 3 different types of CNS tissue are affected by irradiation: the neurons, the glial cells, and the vessels. 19 Effects on the fibrinolytic enzyme system and immune mechanisms have also been suggested. 20,21 It has been shown that oligodendrocytes are very sensitive to radiation 21,22 ; and even if the neurons are less sensitive than the oligodendrocytes, 19 there is sufficient loss of cellular compo-nents to explain the brain-volume loss seen after radiation. 18 Injury to the vasculature can be seen both in early-delayed and late-delayed radiation injuries, with changes in capillary permeability, resulting in edema in the acute phase, followed by vascular endothelial damage in the chronic phase of radiation injury. A previous study has demonstrated absence of tissue plasminogen activator and excess of urokinase plasminogen activator in patients with radiation-induced necrosis, which may contribute to the cytotoxic edema and the tissue necrosis seen in the acute phase of a radiation injury. 20 The incidence of radiation necrosis after conventional therapy ranges from 5% to 24%. 18 The delayed neurologic symptoms include functional and cognitive impairments, with deficits in learning, working memory, executive function, vision, motor function, and, eventually, dementia. [23][24][25][26] On conventional MR imaging, the effects of radiation on brain tissue are evident in some patients as early as 2-6 months after completion of radiation therapy as signal-intensity abnormality in white matter. 27 These changes are defined as earlydelayed radiation-induced injury. Signal-intensity abnormality in the periventricular white matter has also been observed but usually not until 12-18 months after radiation therapy. 27,28 Changes observed in animal models [29][30][31] and postmortem human brain specimens 32 include brain inflammation, demyelization of white matter, breakdown of the blood-brain barrier, and an array of neurotoxic effects. 33 A spatial relationship between the local radiation therapy dose and the changes in the brain seen on CT or MR imaging has been noted from retrospective studies. 27,34 In addition, it has been demonstrated that normal-appearing large white matter bundles such as the genu and splenium of the corpus callosum show progressive structural degradation after radiation therapy. This is evident initially in high-dose regions and later becomes evident outside the high-dose regions. 35 Recent prospective studies have used MR spectroscopy to prove that structural degradation in cerebral tissue after radiation therapy can be predicted by early changes in metabolic activity before the development of neurocognitive symptoms or anatomic changes seen on conventional MR imaging. 36,37 Significant alterations in brain metabolites, especially a decrease in N-acetylaspartate (NAA), which is considered a neuronal marker, were shown to occur in normal-appearing human brain parenchyma early during radiation treatment; interval progression of these changes was noted during at least a 6-month period. 36 The decrease in NAA has been suggested to be due to neuronal damage such as neuronal cell death due to apoptosis or neuronal dysfunction secondary to the irradiation. 38 Other explanations such as neuronal response to blood-brain barrier breakdown, edema, damaged oligodendrocytes, demyelinization, release of cytokines, and exposure to inflammatory cells have also been suggested. 19,39,40 The second important metabolite that seems to be affected by irradiation is the choline (Cho) compound, which is correlated with cell membrane biosynthesis and metabolic turnover in proliferative tissue. 41 Observations of decreases in both the Cho and the Cho compounds and a decrease in the Cho/ Creatine (Cr) ratio have been reported in irradiated brain. 36,37,42,43 A third large metabolite present in the normal spectra is Cr. Cr is a marker of energy metabolism and is commonly considered to be fairly stable under most conditions and, therefore, is often used as the denominator in metabolic ratio calculations, even if some reports question the stability of Cr in tumors, hypoxia, and other confounding factors. 41,44 Injury to the white matter outside the area of the initial brain lesion has not only been demonstrated with MR spectroscopy but is also supported by findings with DTI. For example, a recent study showed that the mean FA value decreased and the average of the mean isotropic ADC value increased significantly in normal-appearing white matter in patients treated with radiation compared with values found in normal white matter in control subjects. 45

MR Spectroscopy in Radiation Injury
Conventional MR imaging alone cannot reliably discriminate tumor recurrence/progression from the inflammatory or ne-crotic changes resulting from radiation, 3 though the latter can be associated with more specific patterns of enhancement, like "soap bubbles" or "Swiss cheese" 17 ; both recurrent tumors and radiation injury typically demonstrate contrast enhancement (Fig 1). Common and uncommon features of radiationinduced injury include enhancement at the site of the original lesion and distant from the original lesion or as multiple necrotic enhancing masses spread in the brain, simulating brain metastases (Figs 2 and 3). In a recent study of 11 patients with histology-proved radiation necrosis, the most common MR imaging finding was Swiss cheeseϪlike enhancement with feathery margins and central necrosis. 46 In addition to the contrast-enhancing lesions, high T2 signal intensity in the periventricular and deep white matter with no enhancement is a common feature in patients who have been irradiated.  Specific spectroscopic changes that occur in radiation necrosis have been reported and include slight depression of NAA and variable changes in Cho and Cr. [47][48][49][50] In addition, radiation necrosis may show a broad peak between 0 and 2 ppm, probably reflecting cellular debris containing fatty acids, lactate (Lac), and amino acids. 51 Also, other metabolites have been suggested to be present in radiation necrosis. In 1 study monitoring the progression of severe cerebral radiation injuries in the temporal lobes of 10 patients previously treated for nasopharyngeal carcinoma, an unknown resonance named Px in the 2.37-to 2.40-ppm region was detected in the affected temporal lobes in 4 of the 10 patients. 52 The resonance of Px was only confined to spectra with Lac and in patients with the highest severity grade of radiation injury. Lesions with Px had significantly higher Lac/Cr ratios and more extensive masseffect changes than lesions without Px. The authors speculated whether Px could be associated with anaerobic glycolysis producing pyruvate (2.37 ppm) or succinate (2.40 ppm) as can be seen in brain abscess formations. 52 Many of the newly occurring lesions that are subjected to MR spectroscopy do not consist only of large areas of pure tumor or radiation injury/necrosis. It has to be assumed that, more commonly, a mixture of tumor cells and tissue with radiation injury is present. This assumption is supported by a prior study of multivoxel MR spectroscopy that found that "spectral patterns do allow reliable differential diagnostic statements to be made when the tissues are composed of either pure tumor or pure necrosis, but the spectral patterns are less definitive when tissues composed of varying degrees of mixed tumor and necrosis are examined." 53 The effort to separate tumor recurrence from pure radiation damage might be more problematic when using SVS compared with using 2D CSI. SVS of an enhancing lesion that contains a small focus of recurrent tumor in a bed of much larger radiation necrosis would likely be "averaged out" such that the Cho and NAA metabolite profiles may suggest only inflammatory changes and the recurrent tumor would be missed. In addition, an enhancing lesion containing heterogeneous areas of normal CNS tissue and recurrent tumor also could be averaged into a spectral profile suggestive of only inflammatory changes, with the recurrent tumor being missed. Averaging normal or radiation-injured brain tissue with tumor tissue will tend to lower Cho/Cr and Cho/NAA ratios of the pure tumor and lower its conspicuity, the later being a matter of major clin- ical importance. With 2D CSI, the coverage of contrastenhancing tissue, surrounding tissue, and normal-appearing white matter in the contralateral hemisphere is allowed (Fig 4). This enables sampling of multiple discrete regions, which may be necessary to discern the subtle differences between tumor recurrence and radiation injury and for the identification of areas of both tumor and inflammatory changes in the same enhancing lesion.
A previous study using the 2D CSI technique reported a 97% success rate for retrospectively differentiating recurrent tumor from radiation injury, with significantly increased Cho/ NAA and Cho/Cr ratios in areas of recurrent tumor compared with areas of radiation injury and with normal adjacent brain tissue. 54 That study reported that when cutoff values of 1.8 for either Cho/NAA or Cho/Cr were used (ie, values Ͼ1.8 being diagnostic for tumor recurrence), 27 of 28 patients were retrospectively correctly diagnosed. 54 In a similar population examined with SVS, similar significant differences were found, and the use of Cho/Cr and Cho/NAA ratios allowed a correct retrospective classification in Ͼ80% of the cases. 50 These findings are in agreement with those in a previous study using multivoxel 1 H-magnetic resonance spectroscopy imaging ( 1 H-MRSI) and correlation with histologic specimens. 55 In that study, the investigators claimed that a Cho/Cr ratio Ͼ1.79 or a lipid (Lip) and Lac/Cho ratio Ͻ0.75 has sevenfoldincreased odds of being pure tumor compared with pure necrosis and the odds of the biopsy's being pure necrosis and having either the Cho/normalized Cr (nCr) values Ͻ0.89 or a Cho/normalized Cho (nCho) value Ͻ0.66 are 6 times the odds of the biopsy's being pure tumor. 55 Also multivoxel 3D proton MR spectroscopy ( 1 H-MR spectroscopy) has been used in the assessment of recurrent contrast-enhancing areas at the site of the previously treated gliomas. 56 In that study, the investigators found the Cho/NAA and Cho/Cr ratios to be significantly higher in recurrent tumor than in radiation injury, whereas the NAA/Cr ratios were lower in recurrent tumor than in radiation injury. They also noted that the Cho/Cr and Cho/NAA ratios were significantly higher in radiation injury than in normal-appearing white matter; however, the NAA/Cr ratios were significantly lower in radiation injury than in normal-appearing white matter. When they used receiver operating characteristic analysis, the resulting sensitivity, specificity, and diagnostic accuracy of 3D 1 H-MR spectroscopy were 94.1%, 100%, and 96.2%, respectively, based on the cutoff values of 1.71 for Cho/Cr or 1.71 for Cho/NAA or both as tumor criteria. 56 Another ratio that has been used in the attempt to diagnose radiation necrosis is the Cho/Lip or Lac ratio. 57 The authors of that study found that in cases of radiation necrosis, a high lipid-dominant peak was observed from the central nonenhanced region, along with a low Cho peak and a low NAA peak. The positive predictive value of a Cho/Lip or Lac ratio Ͻ0.3 and the positive predictive value of a Cho/Cr Ͻ2.48 for diagnosing radiation necrosis were 100% and 71.4%, respectively. 57 They concluded that it is possible to differentiate statistically radiation necrosis from metastatic brain tumor by using the Cho/Cr ratio or the Cho/Lip or Lac ratio. However, they found no significant difference between glioblastoma and radiation necrosis by using the Cho/Cr ratio. 57 Similar figures were seen in another study by the same group in which the authors could differentiate ring-enhancing lesions as "space-occupying radiation necrosis" from ring-enhancing metastasis in all 6 cases by using MR spectroscopy. 58 With recurrent/residual tumor, the pathologic spectra consistent with the presence of tumor (ie, markedly elevated Cho and depressed NAA) can be identified not only in voxels placed within the contrast-enhancing lesion but also in voxels outside the contrast-enhancing lesion, as demonstrated in 38% of the cases in a recent study. 54 It has been suggested that the radiation dosage plays a role in the changes in metabolic ratios. An increase in Cho/ Cr ratios after radiation therapy, proportional to radiation dose, has been reported from studies using SVS 43,50,51 ; a decrease in NAA or a reduction in NAA/Cr ratios or both af- ter radiation therapy are also a frequently reported finding (Fig 5). [41][42][43]47,48,50,51,59

Measurements of Metabolites and Ratio Calculations
One of the major problems when reviewing the literature and comparing existing MR spectroscopy data is the use of various ways of calculating metabolite ratios. Several different methods have been used when trying to differentiate recurrent tumor from radiation-induced injury, and there is currently no consensus in the literature on which type of differently calculated ratios can best differentiate tumor recurrence from radiation change. Standard metabolite ratios calculated from data obtained within the diagnostic voxel (eg, Cho/NAA) have been used in many institutions 54,60-64 and sometimes compared with similar ratios obtained from the contralateral hemisphere. 65 Alternatively, normalized ratios have been calculated by using a variety of definitions such as the following: 1) using 1 metabolite from the diagnostic voxel in the numerator and the other (or same) metabolite from the contralateral hemisphere in the denominator (eg, Cho/normalized NAA [nNAA]), 53,55,63,66 2) dividing a metabolite peak area measured in the lesion by the sum of all the spectrum peak areas in the same lesion, 41 and 3) dividing a metabolite peak integral in the lesion by the same metabolite peak in the "normal brain" and comparing it with standard metabolite ratios. 67 Other more tumor-specific ways of spectrum evaluation, like the Cho-NAA index, have been used to assess the presence of tumor recurrence. 68 In a recent study, the 2 most common ways of calculating ratios were compared: Normalized ratios in which the metabolite from the normal contralateral hemisphere was used as denominator were compared with "nonnormalized" ratios obtained from metabolites in the lesion (both as the numerator and the denominator). 69 That study demonstrated that 2 non-normalized ratios, Cho/NAA and NAA/Cr, were significantly correlated with tumor recurrence, whereas only 1 of the 6 normalized ratios tested, Cho/nNAA, significantly correlated with tumor recurrence. Non-normalized Cho/NAA was the best discriminator for predicting tumor recurrence, with a sensitivity of 86%, a specificity of 90%, a positive predictive value of 92%, a negative predictive value of 81%, and 88% correctly classified. Non-normalized Cho/ NAA performed the best (area under the receiver operating characteristic analysis [ROC] curve ϭ 0.92), followed by NAA/Cr (area under the ROC curve ϭ 0.85). Cho/nNAA was the only ratio to yield significant values for tumor recurrence with a sensitivity of 73%, specificity of 40%, a positive predictive value of 64%, a negative predictive value of 50%, and 60% correctly classified; the area under the ROC curve was 0.77. 69 One major drawback of most of the studies in this field, regardless of the ways of calculating the metabolite ratios, is the limited correlation with histopathology. From 1 of the few larger reports with correlating histopathology, it is clear that MR spectroscopy is a helpful tool in differentiating pure necrosis from normal tissue and from pure tumor; however, MR spectroscopy has problems in differentiating when the specimen is a mixture of recurrent tumor and radiation necrosis. 55 That study also found that combining the information from ADC values with the MR spectroscopy ratios did not help much when trying to distinguish specimens of mixed tumor and the addition of ADC did not increase the accuracy in identifying pure tumor or pure necrosis compared with MR spectroscopy alone. 56

Need for Prediction Models for Clinical Decision Making
Although previous studies have shown that MR spectroscopy findings affect the clinical management of patients with brain tumor, 62,70 the question remains whether MR spectroscopy alone or combined with other methods can affect clinical management of patients with new contrast-enhancing lesions after radiation.
Although specific ratio cutoffs may be helpful in defining the ability of the MR spectroscopy to detect statistically significant differences between patients with recurrent tumor and those with radiation injury/necrosis, a more clinically applicable measure of the utility of a test, in this case MR spectroscopy, is needed. The clinical relevance of a positive or negative test may be estimated by the corresponding likelihood ratio. 71 Although likelihood ratios are effective tools in guiding selection of a diagnostic test, clinical decision making at the level of the individual patient relies ultimately on the assignment of posttest probability-that is, the probability of disease, given the test result. Posttest probabilities of recurrent tumor can be explicitly quantified for the individual patient by using a prediction model. Recently such a model has been proposed, 72 which uses alterations in the ratios of standard brain metabolites to predict the probability of tumor recurrence in patients previously treated for brain tumors with new contrastenhancing lesions. 72 Because the model predicts the probability of tumor recurrence, it allows classification of patients into different clinical management strategies by using ranges of posttest probability, which may facilitate clinical decision making.
In the future, prediction models combining multiple metabolic ratios, with or without clinical data, may prove to be an even more effective decision-making tool, resulting in reduction of the number of patients subjected to unnecessary invasive procedures or treatment.
In the future, not only are more sophisticated prediction models needed but also larger multicenter studies validating such models and larger prospective studies using new methods like 3D MR spectroscopy and target-guided biopsies with histologic confirmation to reach a consensus about the true value of MR spectroscopy as a method of differentiating recurrent tumor from radiation injury. Future investigations should be focused on trying to determine if MR spectroscopy alone can be the tool that is needed or if the technique should be used only in combination with other radiologic methods to obtain the highest sensitivity, specificity, and accuracy.

Conclusions
MR spectroscopy is presently one of the noninvasive radiologic methods used to distinguish recurrent tumor and radiation injury in patients previously treated with radiation for brain neoplasm. Still, despite a considerable volume of research in the field, no consensus exists in the community regarding ratio calculations, the accuracy of MR spectroscopy to identify radiation necrosis, and the accuracy of MR spectroscopy in differentiating radiation necrosis from tumor recurrence or the true value of the method in clinical decision making.