Textural features that correlate with MGMT promoter methylation in glioblastoma multiforme tumors can be detected non-invasively using S-transform based texture analysis.

Glioblastoma multiforme (GBM) is the most common, and most aggressive, type of primary brain tumor in adults.[1] GBM is difficult to treat because the tumor cells are typically resistant to chemotherapy. As a result, treatment is usually palliative, and median survival time ranges from 7.5 to 17.1 months, depending on the severity of the disease.[2] However, patients with MGMT promoter methylation tend to respond to chemotherapy and have significantly longer survival times than those with unmethylated MGMT.[3] MGMT is a DNA repair gene. When it is silenced by promoter methylation, the tissue is unable to repair DNA damage, rendering the otherwise drug resistant tumor sensitive to chemotherapy treatment. Currently, the only way to test for MGMT methylation is a molecular test, requiring a biopsy to obtain a sufficiently large tumor tissue sample. A non-invasive test could help guide clinicians to identify patients who may benefit from chemotherapy, particularly when the tissue sample is not large or well preserved enough for molecular testing.

A previous study found significant correlations between MGMT methylation status and features evident on magnetic resonance (MR) imaging.[4] Unmethylated tumors exhibited more extensive necrosis than methylated tumors (p < 0.002) and ring enhancement was significantly associated with unmethylated MGMT promoter (p = 0.006). However, in this study the location and MR appearance of GBMs were assessed visually, which makes comparing results from different groups difficult.

Intrigued by these findings, we performed a quantitative study employing a novel space-frequency texture analysis technique. Our technique was based on the polar S-transform (PST) – a method that has previously been used to identify tumors with and without 1p/19q deletion in patients with oligodendrogliomas, a tumor related to GBM [5]. However, our new technique reduces the extensive computational requirements of the PST [6]. Textural features are quantified in MR images by assessing the local spatial frequency content: strong low frequencies appear as homogenous smooth regions, while strong high frequencies are seen as heterogeneous detailed regions. Our goal is to define a non-invasive test for MGMT, based on MR image texture, which could potentially be used for the prescription of chemotherapy on an individualized basis.

Patients with newly diagnosed GBM (astrocytoma grade IV, WHO classification) treated at the Tom Baker Cancer Centre in Calgary, Alberta between January 1, 2004, and December 31, 2006, were identified through a review of pathology reports. Inclusion criteria included: age > 20 years, available preoperative MR images in the Picture Archiving, and Communication System (PACS) and available paraffin embedded tumor tissue taken from the first surgery. MGMT promoter status was assessed by methylation-specific polymerase chain reaction (MS-PCR) as previously described.[3] Exclusion criteria included: any other pathology diagnosis but GBM, lack of preoperative MR in the PACS system, or inability to determine the methylation status of the promoter of MGMT. This study received IRB approval from the Conjoint Health Research Ethics Board of the University of Calgary and Alberta Cancer Board.

T2, FLAIR and T1-weighted post-gadolinium images were evaluated (Figure 1). Since the imaging parameters varied across the patient cohort, all images were resampled to ensure a common field-of-view (FOV) of 22 cm and consistent pixel resolution of 0.859 mm. We performed a rigid registration for the three contrasts of each patient with a normalized mutual information metric using in-house registration software. The average and standard deviation of signal intensity in cerebrospinal fluid was normalized. Tumor boundaries were outlined on each slice of the T1 post-contrast image volumes using MIPAV.[7]

Figure 1: T2-weighted MR images of a methylated (left) and unmethylated (right) GBM tumor.

We evaluated every slice of each imaging volume that contained tumor, as well as extracting 16x16 ROIs from the binary mask of each slice of tumor to remove any edge effects that may interfere with the analysis, as well as to allow for comparison to the PST approach. Slices, where the tumor area was <50mm2, were excluded from the texture analysis. We calculated the average spectrum over all pixels within the tumor volume, or ROI, to obtain a single spectrum for each patient. Pixels that were not within the tumor masks were not included in the analysis. The ROI spectra were compared to those obtained using the PST to ensure consistency of the new transform.

Average tumor signal intensity on T2, FLAIR, and T1 post-contrast images, as well as tumor volumes, were measured and compared between methylated and unmethylated groups. Spectra were log-transformed prior to statistical analysis to stabilize the variation. Two-way repeated measure ANOVA tests were performed for each contrast and 95% confidence intervals (CI) were calculated on the group differences. Two-sided t-tests were used to compare patient characteristics between groups, and values < 0.05 were considered significant.

Fifty nine (59) patients were included in the texture analysis for MGMT status. Their median age was 59 (range: 29-82) and median Karnofsky Performance Score (KPS) was 80 (range: 50-100). About one quarter of the patients (26%) had had an initial biopsy, and 84% had a KPS >= 70 (Table 1). Thirty one patients had GBMs with MGMT promoter methylation (53%). Median imaging parameters were as follows (T2/FLAIR/T1post-contrast): TR=4160/9004/500 ms, TE=102/105/14 ms, 19 slices; median TI for FLAIR = 2400 ms. One of the FLAIR images was excluded during the study because the motion artifacts were found to interfere with calculation of the spectra.

Table 1: Comparison of patient characteristics between unmethylated (n=28) and methylated (n=31) tumors.

No significant differences in any of the textural features between methylated and unmethylated tumors were detected in the T2, FLAIR, or T1 post-contrast images when analyzing the entire tumor volumes. However, Figure 2 shows that the mean spectral power of the unmethylated tumors in the T2-weighted ROI analysis was greater than that of the methylated tumors in the spatial frequency range of 0.73-4.36 cm-1 (with the exception of 2.91 cm-1). The ROI analysis, using the PST, also identified spectral differences in the T2-weighted images for frequency components 0.73, 1.45, and 2.91 cm-1. The T2 ROI spectra were found to be significantly different, based on MGMT status in the ANOVA analysis (Table 2). There were no significant differences in signal intensity identified in any of the three contrasts or in tumor volume between the two groups (Table 3).

Figure 2: Mean group differences and 95% CI between average local T2 ROI frequency spectra of unmethylated (n=28) and methylated (n=31) tumors.

Table 2: p-values of group differences from ANOVA analysis between methylated and unmethylated spectra from full tumor volumes and 16x16 ROIs. The texture (spatial frequency distribution) from an ROI placed within the tumor boundary in the T2 weighted image was significantly different between the two groups (shaded grey).

Table 3: Average standard deviation of signal intensity and volume of methylated and unmethylated tumors along with p-values calculated using Student’s t-test.

Our results suggest that texture analysis of ROIs may be more suitable than analysis of entire brain tumor volumes. Incorporating spectra from pixels near the tumor border may result in edge effects, obscuring texture differences between tumor types. While the study in [4] found significant differences between tumor types in terms of enhancement patterns on T1 post-contrast imaging, we did not find any differences in the T1 spectra. Further work needs to be done to compare our results to those in [4].

Preliminary results suggest that textural differences between tumors with and without MGMT promoter methylation can be identified using ST-based texture analysis of MR imaging. Results were similar between our new technique and the PST. However, further work will determine whether this information can be used to accurately predict MGMT status of unknown tumors with a high sensitivity and specificity.

[1] Stupp R, Mason WP, van den Bent MJ, et al. “Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma”. N Engl J Med. 2005;352(10):987-996.

[2] Shaw EG, Seiferheld W, Scott C, et al. “Reexamining the radiation therapy oncology group (RTOG) recursive partitioning (RPA) for glioblastoma multiforme (GBM) patients.” Int J Radiat Oncol Biol Phys. 2003;57(2):S135-S136.

[3] Hegi ME, Diserens AC, Gorlia T, et al. MGMT “Gene silencing and benefit from temozolomide in glioblastoma.” N Engl J Med. 2005;352(10):997-1003.

[4] Eoli M, Menghi F, Bruzzone MG, et al. “Methylation of O6-methylguanine DNA methyltransferase and loss of heterozygosity on 19q and/or 17p are overlapping features of secondary glioblastomas with prolonged survival.” Clin Cancer Res. 2007;13(9):2606-2613.

[5] Brown R, Zlatescu M, Sijben A, et al. “The use of magnetic resonance imaging to noninvasively detect genetic signatures in oligodendroglioma”. Clin Cancer Res 14(8):2357-2362, 2008.

[6] Drabycz S and Mitchell JR. “Texture quantification of medical images using a complex space-frequency transform”. Int J CARS, DOI 10.1007/s11548-008-0219-4, 2008.

[7] McAuliffe MJ, Lalonde FM, McGarry D, et al. “Medical image processing, analysis & visualization in clinical research”. Proc. IEEE Computer-Based Medical Systems (CBMS) 381-386, 2001.