Neuroimaging in Pediatric Brain Tumors: Gd-DTPA–enhanced, Hemodynamic, and Diffusion MR Imaging Compared with MR Spectroscopic Imaging

Patients

Thirty-one patients with neuroglial tumors underwent conventional MR imaging and MR spectroscopic imaging; 16 patients underwent hemodynamic MR imaging, and 12 of the 16 underwent diffusion-weighted MR imaging. The mean age of the patients was 10 years (range, 6 months to 16 years). When necessary, sedatives (midazolam hydrochloride 0.1 mg/kg, lorazepam 0.1 mg/kg, or chloral hydrate 50 mg/kg by mouth) were administered to the patients. Children were examined at initial presentation or immediately after surgery and before any adjuvant treatment (radiation therapy or chemotherapy or both). Follow-up conventional and spectroscopic MR images and hemodynamic MR images were included in this study; diffusion-weighted images obtained in four children with malignant recurrent tumors were not included. Recurrence was verified with stereotactic biopsy, as well as the clinical course. We included only follow-up MR images obtained after the acute phase of radiation (approximately 3 months) to ensure that nonspecific posttreatment effects did not affect our observations.

The pathologic classification used in our institution is the World Health Organization classification of brain tumors (7, 8). The two main categories of tumors included in this study were glial tumors and embryonal neuroglial tumors. Tumors were classified at biopsy as follows: pilocytic astrocytomas (n = 6), grade I astrocytomas (n = 2), ependymomas (n = 3), poorly differentiated (embryonal) tumors (n = 4), medulloblastomas (n = 2), pineoblastomas (n = 2), anaplastic astrocytomas (n = 2), choroid plexus carcinomas (n = 2), and gangliogliomas (n = 2). Pilocytic astrocytomas, grade I astrocytomas, and gangliogliomas were classified as benign tumors, whereas ependymomas, medulloblastomas, pineoblastomas, anaplastic astrocytomas, choroid plexus carcinomas, and other embryonal tumors were classified as malignant tumors. Therefore, we examined 10 patients with benign tumors and 15 with malignant tumors. The remaining six patients with inoperable glial tumors (eg, brain stem gliomas) did not have a histopathologic diagnosis. Informed consent was obtained from each participant’s parent or guardian before his or her inclusion in the study if the study was not clinically indicated.

To spatially correlate the histopathologic information to the MR data, we obtained biopsy samples with the guidance of a surgical navigation system to determine the location of each biopsy site on the MR images. Thus, immediately before the removal of a biopsy specimen, a multiplanar MR image of the origin of the sample was saved by using an ISG Viewing Wand (ISG Technologies, Missisauga, Ontario, Canada). Each sample was then obtained with small surgical forceps, labeled, handled separately, and submitted for routine pathologic examination.

MR Methods

MR examinations were performed with a 1.5-T, whole-body imaging and spectroscopic MR system (GE Medical Systems, Milwaukee, WI) and a quadrature head coil. Our system permitted conventional imaging, as well as echo-planar and multilevel hemodynamic MR imaging. In addition, we used our automated spectroscopic acquisition routines, which resulted in faster data acquisition (approximately 4 minutes per 2D spectroscopic data set with 1 × 1 × 1-cm resolution). These acquisitions were possible at adjacent levels (no gap). Hemodynamic and diffusion-weighted MR images were acquired at corresponding levels. Multilevel 2D MR data acquisitions with no gap were used rather than 3D methods. This approach improved the signal-to-noise ratio, because adjustments for magnetic field homogeneity and water suppression were performed in each section; with a large volume, these adjustments often fail in the clinical MR setting. The levels of hemodynamic, diffusion, and spectroscopic MR images were chosen after the conventional T1-weighted, T2-weighted, and fluid-attenuated inversion recovery (FLAIR) images were evaluated to include the tumor and surrounding tissues.

Pediatric neurologists (M.K.Z., T.Y.-P.) who were blinded to the findings of MR spectroscopy and perfusion and diffusion imaging studied the MR images. These images were then compared with the multivoxel data obtained at proton MR spectroscopy, as well as with the perfusion and diffusion images at the appropriate level. In as much as our multivoxel data included the tumor (as defined on conventional MR images) and surroundings, we avoided interjecting a bias and assumed that Gd enhancement could be used to identify the extent of the tumor. Thus, we were able to test the hypothesis that MR spectroscopy depicts tumor outside the Gd-enhancing tumor bed.

Hemodynamic and diffusion-weighted MR imaging were performed by using echo-planar and line-scan techniques, respectively (9, 10). Hemodynamic images (rCBV maps) and calculated diffusion images (ADC maps) were generated on a remote workstation (Sun Microsystems, Mountain View, CA). Hemodynamic MR imaging was performed by using a single-shot echo-planar T2*-weighted method during a compact bolus injection of Gd-DTPA, with a 1.5-T MR system (9). A clinical dose (0.1 mmol/kg, 0.2 mL/kg) of gadolinium-based contrast agent was administered intravenously by using an automatic injector with a power injector rate of 1–2 mL/s. To ensure the quality of the hemodynamic MR images, the intravenous line was placed before the patient entered the MR unit. Twenty-five images were acquired in nine adjacent axial sections (10 mm thick) within 1.25 minutes. Typical parameters for hemodynamic MR imaging were 300/40/1 (TR/TE/NEX), with a FOV of 40 cm.

The diffusion-weighted MR imaging method used in our study has been described previously (10). Briefly, diffusion-weighted imaging was performed by using a line-scan method to acquire axial images (FOV, 20 × 15 cm; effective section thickness, 7 mm; gap, 0 mm; acquisition matrix, 128 × 128 interpolated to 256 × 192; b = 5 and 750 s/mm² as the maximum b value applied along the three orthogonal directions). The minimum imaging time was 5 minutes 49 seconds. ADC maps and trace images extrapolated to a b value of 1000 s/mm² were generated offline for each of the 10 locations. The trace images were calculated as the cubed root of the product of the three diffusion axes.

Proton MR spectroscopic imaging was performed by using multivoxel chemical shift imaging with point-resolved spectroscopy (PRESS) and volume preselection (5). Briefly, after a 50–100-mL volume was selected and after shimming and water suppression adjustments were made, a large data set was obtained by using phase-encoding gradients in two directions. The following parameters were used: 1000/65 (TR/TE), 16 × 16 phase-encoding matrix, 160-mm FOV, section thickness of 10 mm, 1250-Hz spectral width, two averages, and 512 points. Data sets of 1–1.2-cm³ resolution were acquired. The decision to use a TE of 65 milliseconds was made for the following reasons. In our study, we were not as interested in the lactic acid detection as in the presence of lipids, which might be important, because lipids are related to tumor necrosis or apoptosis; this, in turn, is a determining factor of tumor activity. Our pretest hypothesis was that a TE of 65 milliseconds provides us with the opportunity to 1) null lactic acid; 2) increase sensitivity in lipid detection, and 3) prevent diffusion artifacts and water-suppression failures with PRESS performed at TEs shorter than 65 milliseconds. Thus, the four prominent peaks of biologic importance in our studies were those of NAA, Cho, tCr, and lipids (and/or lactate). Data were processed on a workstation (Sun Microsystems) by using analysis software (SAGE; GE Medical Systems) and software developed in-house by using interactive data language (IDL) 5.3 (Research Systems, Boulder, CO). The data sets were apodized with a 1.0-Hz lorentzian filter, Fourier transformed in the time domain and two spatial domains, and phased (first automatically and then manually if necessary) by using the SAGE spectroscopic analysis software (GE Medical Systems) . Then, a baseline estimator was used to subtract the broad components of the baseline before peak area calculations. Finally, the areas of selected metabolite peaks were estimated by using the PIQABLE algorithm developed in IDL. Metabolite images were generated and stored as tiff files on a SPARC workstation (Sun Microsystems) and then transferred to a Macintosh workstation (Apple Computer, Cupertino, CA). We used imaging editing software such as NIH Image (National Institutes of Health, Bethesda, MD) and Photoshop (Adobe, San Jose, CA) to overlay the metabolite images on the corresponding anatomic images.

Normalized MR spectroscopy–derived parameters (Cho, lipids) were compared with rCBV and ADC values. We normalized these parameters by dividing their values from within tumor regions by the mean value of tCr in normal tissue in each patient, which can be used as a reasonable internal standard (11). Thus, our spectroscopy-derived parameters were expressed in arbitrary units. We also normalized values in the same tumor regions of interest on rCBV maps with values obtained from within the middle cerebral artery territory in the same patient. The rCBV values were also expressed in arbitrary units. ADC values were expressed as a number × 10⁻² in units of square millimeter per millisecond. Percent enhancement in the tumors was calculated by subtracting the enhancement within normal tissue from the enhancement within tumor tissue and by dividing this by the enhancement within normal tissue and multiplying the result by 100. Multiple measurements were obtained in each patient to account for brain tumor heterogeneity.

Biostatistical Analysis

Results of the Kolmogorov-Smirnov test of normality indicated that Cho, rCBV, ADC, and percent enhancement values did not significantly depart from a gaussian distribution. Therefore, the Pearson product-moment correlation coefficient, r, and a linear regression analysis were used to evaluate the strength of relationships. Stepwise multiple linear regression was performed to identify independent predictors of percent enhancement. Because the data for lipids were heavily skewed and because they did not conform to a normal distribution, a natural logarithmic transformation was used to normalize the lipid variable. For correlational analysis, the number of measurements included 139 for percent enhancement, Cho, and lipids; 68 for rCBV; and 48 for ADC. To account for the multiple measurements within the same patient, we also used a nested analysis of variance approach to test the relationship among the variables (12). SAS statistical software (version 6.12; SAS Institute, Cary, NC) was used to analyze the data. All reported P values are two tailed, and P values of less than .05 were considered to indicate statistically significant differences.

Choline-Containing Compounds

In our study, Gd-DTPA enhancement on conventional MR images did not correlate with Cho findings (Fig 1A). Meanwhile, Sijens et al (13, 14) reported that, for large lesions (with a single-voxel proton MR spectroscopic method in 15 patients), Cho findings were significantly correlated with percent tumor enhancement. The lack of agreement between our findings and those of Sijens et al may be due to different tumor histologic features, because they examined adult patients with brain metastasis of breast carcinoma. Also, their single-voxel approach and their analysis, which included regions of necrosis, led to the interpretation that their correlation was due to a decrease in cellularity and Gd-DTPA delivery to necrotic areas (14). We studied pediatric brain tumors, which are known to have histopathologic characteristics that differ from those of adult tumors. Also, we used a multivoxel method with 1 × 1 × 1-cm resolution, which better addressed tumor heterogeneity, instead of a single-voxel approach in which viable tumor and necrosis are averaged over a volume of interest that is typically much larger than 1 cm³. Viable tumor growth and necrosis have different spectral patterns (5, 6).

Elevated Cho levels were detected in both enhancing (Fig 2) and nonenhancing (Fig 3) tumors. Published reports of proton MR spectra in adults (6, 11, 15) and children (5, 16–18) with untreated brain tumors and in brain metastases have also indicated high levels of Cho. Although compounds contributing to the Cho signal in vivo have been suggested (19), the particular compound that increases with neoplasia remains essentially unknown. Moreover, Cho has been associated with increased cellularity (17, 20–22), membrane synthesis, and rapid cell turnover or proliferative activity (15, 23, 24). Cho was proposed to be the most reliable indicator of malignancy in gliomas (15); also, it has been shown to increase with benign brain tumors in children (17, 25).

Our patients who underwent follow-up MR examination had elevated Cho levels and recurrence before the appearance of contrast-enhancing regions on conventional MR images (Figs 5 and 6). In other studies (6, 15, 26), highly elevated Cho levels on the posttreatment proton MR spectra were thought to indicate areas of recurrence or progressive disease or malignant transformation or both rather than areas of radiation necrosis. On the other hand, increases in Cho levels have been observed both in animals and after irradiation of normal brain tissue (27) and in patients with brain tumor shortly after radiation treatment (28). They have been attributed to the breakdown of membrane lipids with an elevation in the level of water-soluble mobile phosphocholine (29) and with demyelination of the brain caused by the destruction of oligodendrocytes and/or myelin with the subsequent release of Cho (30, 31). Nevertheless, in our study, we included the patients with follow-up MR findings obtained only after the acute phase of radiation (approximately 3 months) to ensure that nonspecific postradiation effects did not affect our observations about Cho.

Our finding that no correlation existed between Cho values and percent tumor enhancement and that regions with high Cho levels existed beyond the contrast-enhancing regions (Figs 2–4) agree with those of other investigators who studied adult glioblastoma multiforme (6). Therefore, we believe that percent enhancement, although it coincides with blood-brain barrier breakdown, does not necessarily indicate an early phase of tumor growth, which seems to be better detected with proton MR spectroscopy. We also believe that Cho mapping adds value to conventional MR imaging, especially if it is combined with lipid mapping.

Lipids

In our study, peaks at 1.33 ppm, which we identified as lipid peaks, were not detected in all patients. We considered that these peaks primarily consisted of lipids and secondarily lactate, because our acquisition protocol, with a TE of 65 milliseconds, was not optimized to depict lactate. Some colleagues (32–34) have focused on lactate and tailored their protocols accordingly; they showed that lactate is an indicator of tumor metabolic activity, while others have disputed the role of lactate as an indicator of malignancy (35, 36). Thus, the role of lactate at in vivo MR spectroscopy of tumors cannot be established until the appropriate pulse sequences and protocols for lactate detection are implemented in the clinical setting (36–38). In our study, a TE of 65 milliseconds was chosen because this was the shortest TE that could be implemented to achieve T2 weighting and sensitivity to detect metabolites with short TEs, such as lipids. Peaks at the 0.8–1.8-ppm range are probably due to aliphatic compounds (lipids) and other macromolecules (proteins). Recently, certain investigators (21, 39–42) have increasingly considered lipids to be important indicators of tumor cellular activity and indices of tumor grade. Our data are consistent with this notion, and this observation may be of biologic importance, since lipid levels also substantially increased in active tumors versus inactive tumors (43).

In our study, the presence of lipids in tumors, as detected at MR spectroscopic imaging, was not correlated with Gd-DTPA enhancement at conventional MR imaging (Fig 1B). Prominent lipids were detected in both enhancing and nonenhancing parts of tumors (Fig 2). This result was in agreement with the notion that lipids represent microscopic tumor cell necrosis or membrane breakdown that may precede necrosis (23). Ex vivo analysis of brain tumor biopsy samples has also revealed a significant positive relationship between lipids and the extent of necrosis (22). According to our experience, as well as the experience of others, lipids might indicate a high-grade tumor (Fig 2) (44, 45). Moreover, lipids have been observed in contrast-enhancing areas in patients with glioblastoma multiforme after treatment; they have been attributed to radiation necrosis (6). Nevertheless, lipids may be present in viable tumor, presumably because of poor perfusion (45) and hypoxia (46), and they undergo major intensity changes during apoptosis (47). Thus, lipids may indicate still-viable tumor that tends to become necrotic, necrotic tumor, or radiation necrosis. Presumably, low-grade gliomas do not have prominent lipid signals (Fig 3).

Because the presence of lipids is consistent with necrosis or apoptosis (22, 47), we believe that the detection of lipids with MR spectroscopic imaging contributes independent additional information to that obtained with MR imaging. This information may improve the diagnostic accuracy of MR imaging in the differentiation of low- and high-grade tumors.

The concept that necrotic regions in tumors are hypointense on T1-weighted images (as seen in the posterior aspect of the tumor in Fig 4, which is dominated by prominent lipids and less prominent Cho signals) is also plausible. Whether necrotic tissue enhances with Gd-DTPA has been a topic of debate (13, 14).

Tumor MR Spectral Patterns

In our study, the different MR spectral patterns suggested that MR spectroscopic imaging can be used to distinguish at least three different tissue compartments—normal, tumor, and necrosis—whereas mixed MR spectral patterns were due to the known heterogeneity of tumors, as confirmed by the histopathologic features (Figs 2–6). These observations were in agreement with those of previous reports (5, 6, 44). MR spectral patterns with elevated Cho and lipid levels in the absence of NAA that were histologically verified to represent regions of active tumor with extensive areas of necrosis suggested that such MR spectral patterns contribute additional information that is not available with conventional MR imaging. Because glial tumors are graded according to their cellularity, proliferative activity, and degree of necrosis, Cho mapping (increased cellularity and proliferative activity), may added value to MR neuroimaging in patients with brain tumors, especially when it is combined with lipid mapping (necrosis and/or apoptosis). Indeed, contrast-enhancing regions with high lipid levels and low or no Cho levels, as shown in 11 patients with malignant or inoperable tumors, have been suggested to represent areas of high neoplastic potential intermingled with microscopic necrosis; this finding was verified at biopsy in four patients.

Gd-DTPA Enhancement and Hemodynamic and Diffusion-Weighted MR Imaging

Gd-DTPA enhancement on conventional MR images was not correlated with the rCBV (Fig 1C), and rCBV and ADC maps did not show abnormal rCBV or ADC values outside the Gd-DTPA–enhanced tumor bed (Fig 2), except during the follow-up examination of patients with progressing tumors (Fig 6). These findings were consistent with the notion that Gd-DTPA enhancement occurs in areas of blood-brain barrier breakdown and that it coincides with bright regions on hemodynamic images (Figs 2, 4). They also indicate that abnormal hemodynamics are correlated with active tumor, as previous investigators have suggested (21, 48). Others have also confirmed, histologically and angiographically, that hemodynamic images are associated with tumor vascularization (49). Nevertheless, according to other groups, maps of the rCBV cannot be used to differentiate between viable tumor and adjacent gray matter, and they cannot be used to reliably assess tumor progression (6).

In our study, varying degrees of Gd-DTPA enhancement on conventional brain tumor MR images were probably due to increased permeability of microvessels, a feature of neovascularity in human neuroglial tumors (50). This variability in enhancement limits the capability of conventional MR imaging in discriminating among types of tumors and between residual or recurrent tumor and postsurgical changes or radiation necrosis. In certain situations, it may prevent the generation of hemodynamic images, which are based on dynamic susceptibility contrast-enhanced imaging; this, in turn, is based on the assumption that the contrast agent remains within the intravascular space. This factor may complicate the results on postsurgical images or those in cases of radiation necrosis. In both situations, permeable vasculature may be present, and enhancement may or may not occur. In addition, controversial results have been reported in necrotic tumors, which may (51, 52) or may not (53, 54) enhance. Nevertheless, central necrotic regions surrounded by layers of ischemic tissue with a stabilized microcirculation and an advancing front with angiogenesis have been recognized in experimental cancers (50). This result is in agreement with our observations of more pronounced hemodynamics in the outer nonnecrotic regions of tumors, which are presumed to coincide with the advancing front of angiogenesis (Fig 2).

Our study revealed a negative correlation between Gd-DTPA enhancement and ADC and illustrated that Gd-DTPA enhancement (which typically occurs in areas of blood-brain barrier breakdown) coincides with hypointense regions on ADC images (Fig 2). However, ADC maps did not show abnormal ADC values outside the Gd-DTPA–enhanced tumor bed, as illustrated in Figure 2. This finding does not minimize the value of ADC maps, especially for monitoring pediatric brain tumors. This use was not assessed in the present study, but it has been suggested in a growing body of literature about the role of diffusion-weighted imaging of cerebral tumors (55) and by the notion that the ADC characterizes the biophysical characteristics of tissue microstructrure and microdymanics (55–57) and that it provides information (based on pathophysiologic characteristics) that differs from that obtained with contrast-enhancing imaging (58). In support of the above notion, Sugahara et al (59) reported that the minimum ADC value of high-grade gliomas is significantly higher than that of low-grade gliomas and that low ADC values were found in areas of increased cellularity. Others have suggested that the ADC may assist in the early detection of responses to anticancer therapy, because an increase in ADC values has been noted after treatment; this finding is consistent with an increase in the fraction of interstitial water due to treatment-induced cell death (60).

Finally, the analysis of the data in this study did not permit determination of whether rCBV and ADC mapping contribute independent nonrepetitive information beyond that obtained with MR imaging. Further analysis of findings from ongoing studies will reveal whether rCBV and ADC mapping can contribute to the diagnosis, grading, and monitoring of pediatric brain tumors.

Clinically Important Implications

In this study, Cho was detected outside the Gd-DTPA–enhanced tumor bed before treatment in all patients with malignant tumors. After treatment (in four patients with follow-up images) both high Cho and high rCBV values were detected outside the Gd-DTPA–enhanced tumor bed. Before treatment, these areas presumably represented areas of tumor extent. Persistence of high Cho and rCBV values in a tumor area after treatment may signify residual or recurring tumor; this information may lead to additional or alternative therapy. Lipids signify necrosis or apoptosis and, in combination with Cho, rCBV, and ADC values, add information for tumor typing. In patients with benign tumors, high Cho levels, and no lipids were present in nonenhancing areas of the tumor. Given the limited accuracy with Gd-DTPA, which is routinely used to define tumors at diagnosis and to differentiate tumor from necrosis or edema at posttreatment follow-up, the supplemental information obtained with MR spectroscopic, hemodynamic, and diffusion imaging may improve diagnostic accuracy.