Clinical Significance of Fractional Anisotropy Measured in Peritumoral Edema as a Biomarker of Overall Survival in Glioblastoma: Evidence Using Correspondence Analysis
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0028-3886.266284
Source of Support: None, Conflict of Interest: None
Keywords: Diffusion tensor imaging, edema, glioblastoma, magnetic resonance imaging, survival analysis
In the last years, the target in the treatment of glioblastoma (GB) has remained in maximizing the tumor resection volume; therefore, the evaluations before the surgical therapy are critical, and they provide the necessary information to plan the approach and the extension of resection. These assessments are mainly carried out through the imaging studies namely magnetic resonance imaging (MRI), which is an essential tool for identifying the size and possible extension of GB.
Worldwide protocols in the assessment of GB still included only conventional (morphologic) MRI sequences in day-to-day practices (pre and postgadolinium T1; T2, and Flair sequences). More than a decade ago, an MRI-based classification proposed a natural ordering in the severity of GB based on the involvement of the subventricular zone (SVZ) and cortex ; this classification revealed associations between the histogenetic and clinical heterogeneity in GB. Recent studies have proved that GB contacting the SVZ have lower survival, independent of common predictors of GB survival, while others have found that origins of GBMs are at sites distributed throughout the white matter and are not limited to the region of the SVZ.
The advanced MRI technique known as diffusion tensor imaging (DTI) allows the calculation of the fractional anisotropy (FA). FA is an index of the diffusion characteristics of water molecules preferentially directed along the axis of primary axonal pathways. It is a dimensionless variable with values reported between zero and one. FA values relate to the integrity of cell membranes and in this way suspect and identify changes corresponding to infiltration, invasion, and extension of different brain neoplasms. The diagnostic performance of FA has been assessed for several tumor regions (viable tumor, necrosis, peritumoral edema). Recently, FA in contrast-enhancing lesions proved a significant positive correlation with OS in GB, but no distinct influence on OS was observed in peritumoral edema region. Then, there is still missing information in understanding the FA associations with OS considering if measurements were acquired at different tumor regions. This fact might be relevant, as a recent study revealed that local and regional heterogeneity in GB plays a role in the survival stratification.
To understand GB behavior associated with OS, we need a multidimensional approach that considers variables with different dimensions involved in GB assessment. A qualitative variable represents the tumor regions produced by GB (necrosis, enhancing tumor, and edema regions), and a quantitative variable is assigned to the values of FA, an accepted imaging biomarker. The existence of these dimensions means that traditional conventional statistical analyses could not be the best approach. Among appropriate multivariate methods to evaluate this pathology, correspondence analysis (CA) is an exploratory method used for decades to study high-dimensional data. CA reduces data dimensionality to a few orthogonal factors that correspond to the most significant amount of variability in the data ; CA reveals nondetected relationships in a series of pairwise comparisons of variables. This method has been used in the analysis of clinical data (heart rate variability, health surveys, a multivariate study in ankylosing spondylitis),,; and precisely as a supplementary analysis of the overall survival and disease-free survival of patients with colorectal cancer.
Regarding this body of evidence, the goal of this study was to evaluate the association of fractional anisotropy measures at GB related tumor-regions (normal brain, peritumoral edema, enhancing ring, and necrosis) and the MRI-based location with the OS in patients with GBM. We hypothesized that FA depicts a significant association with selected intervals of OS in GB when measured within the immediate zone of peritumoral edema. We assessed this association by performing a multivariate assessment using survival and correspondence analyses with the intention to reveal a specific association using 6-month intervals of OS.
Retrospective study approved by the Institutional Review Board, followed the principles outlined in the Declaration of Helsinki and all subjects gave informed consent. The records from a cohort of 36 patients were recovered from the database of the Radioneurosurgery unit of the National Institute of Neurology and Neurosurgery (INNN). The recruitment period was from January 2012 to December 2013. Inclusion criteria were a histopathologic diagnosis of GB and availability of follow-up to three years; the data recollection was made from clinical records and digital imaging and communications in medicine (DICOM) data images.
Exclusion criteria considered any brain tumor surgery, radiation therapy, or chemotherapy before inclusion in the study as well as the lack of histopathologic diagnoses and missing imaging data. Patients received the Stupp protocol ; after resection, they received chemotherapy including Temozolomide (Temodal, Schering-Plough, NJ, USA), and radiation therapy (60 Gy). All patients underwent a biopsy of a tumor with histopathologic diagnosis based on WHO criteria. A board-certified neurosurgeon, an internist (E.F.A. and C.D.M.), and a radiologist (E.R.V., 12-years of experience) reviewed the clinical records and MRI images, remaining blindfolded to other patient's history during manuscript preparation. We recorded the attendance of patients based on their scheduled appointments at the neurosurgery department every 4 months starting with the MRI diagnosis until death or up to 36 months. Thirty-six patients were included in the analysis. The group consisted of 23 (63.9%) males: 46 ± 14 y [mean and standard deviation (M, SD)], range: 16–67 y; and 13 (36.1%) females: 53 ± 13 (M, SD), range: 30–72 y.
Brain image and data acquisition
MRI images were acquired by using a 3T scanner (Signa HDxt, GE Healthcare; Magnetic Resonance Unit; Waukesha, Wisconsin, USA) with a high-resolution eight-channel head coil (Invivo, Gainesville, FL, USA). Conventional MRI sequences included axial T2-w, axial fluid-attenuated inversion recovery (FLAIR), axial spoiled gradient echo (SPGR), and axial T1-w. Contrast enhancement agent used 0.1 mmol/kg of body weight of gadopentetate dimeglumine (Magnevist; Schering, Berlin, Germany). DTI parameters include FOV of 22 × 22 mm 2, B-value of 1000 s/mm 2, 25 directions, TR = 10,000 ms, TE = 101.8 ms, slice thickness of 3 mm and a Matrix array of 112 × 112. [Figure 1] shows the appearance of conventional MRI sequences in a patient with GBM.
DTI-derived fractional anisotropy
Three tumor regions of interest (ROIs) measuring 10 × 10 mm 2 were manually outlined using a pixel-wise application of the software package FuncTool 9.4.05 (GE Healthcare; Magnetic Resonance Unit; Waukesha, Wisconsin, USA). We considered three tumor regions: necrosis, enhancing tumor, and edema; additional measurements at a symmetrical normal contralateral region (NCR) was also measured [Figure 1]. We defined peritumoral edema regions as those with high T2 and FLAIR signals but without contrast enhancement. Workstation software generated DTI-derived FA and maps of the selected ROIs by using several combinations of the terms of the diagonalized diffusion tensor (eigen values λ1, λ2, and λ3). An explanation of the FA formula was recently published.
MRI-based involvement of SVZ and cortex
We used classification of GBM based on the participation of the SVZ and cortex. Group I, the contrast-enhancing lesion (CEL) extends from the atrium SVZ to the pia. Group II, SVZ-contacting a tumor that does not involve the cortex. Group III, the CEL is invading the cortex, reaching the pia, but does not touch the SVZ and Group IV, tumors that 'respect' both the SVZ and cortex.
Since we found a limited number of patients who underwent advanced MRI (acquisition of DTI), we applied a convenience sampling method selecting only those subjects with available clinical and DICOM records that satisfy the inclusion criteria; our sample size is similar to recent studies available in the literature.
DTI-derived FA measured in three tumor regions: necrosis, enhancing tumor, and edema regions. We included three variables related to survival: the survival status; the overall survival measured in months; and an ordinal variable of several OS intervals. Several studies are coincident with reporting a median OS or progression-free survival of about 12 months in patients who received treatment involving surgery, radiotherapy, and chemotherapy.,, Knowing that only about 2% of GB will survive longer than 36 months ; we used 6-month multiples of OS and recorded up to seven intervals of OS (0–6, 7–12, 13–18, 19–24, 25–30, 31–36, or >36 months). Our last variable represented the MRI-based involvement of SVZ and cortex. A total of 252 measurements were included in the initial univariate analysis.
Overall survival was defined as the time elapsed from the MRI diagnosis to death from any cause. Survival analysis was performed using the Kaplan-Meier method and the Log-Rank test to compare the survival curves between subgroups (we compared four subgroups based on tumor location).
We apply this method to evaluate the association between 7-intervals of OS and four levels of FA. The FA measurements were grouped based on quartiles (lower quartiles depicted lower FA values). CA allowed us to reduce the high-dimensional data sets into few independent factors revealing principal factorial axes and projecting the observations and variables onto a subspace of low dimensionality that accounted for the main variance in the data. Every level or category is called a profile; the CA computes the distances between profiles and represents them in a two-dimension chart. The graphical display of rows and columns shows dots in planar representations, which helped us to detect structural relationships of FA with survival groups. All analyses were carried out using the IBM ® SPSS ® Statistics (software version 22.214.171.124 IBM Corporation; Armonk, NY, USA); a P–value of < 0.05 indicated statistical significance. The study followed the guidelines of the STROBE statement.,
GB lesions distribution and FA comparison among tumor regions
MRI-based distribution showed 12 (33.3%) GB lesions in the right cerebral hemisphere; 20 (55.6%) in the left; and 4 (11.1%) lesions extending to both cerebral hemispheres. The MRI-based location of GB lesions depicted 22 lesions (61.1%) in group I (CEL extending from the atrium SVZ to the pia); one lesion (2.8%) in group II (SVZ-contacting a tumor that does not involve the cortex); twelve lesions (33.3%) pertained to group III (CEL invading the cortex, reaching the pia, but does not touch the SVZ); and one lesion (2.8%) that belonged to a group IV (tumors that 'respect' both the SVZ and cortex) [Figure 2].
Kruskal-Wallis Test revealed a statistically significant difference in FA values across selected regions χ2 (3, n = 133) = 82.432, P <.001. Median (Md) of FA values among regions showed; higher FA values measured at the contralateral normal brain regions Md = 0.351, followed by peritumoral edema Md = 0.190, enhancing ring Md = 0.116, and necrosis Md = 0.071. Pairwise comparisons using Mann-Whitney U Test revealed significant differences among all regions (even after Bonferroni's P value adjustment): NCR and edema U = 121, Z = -5.774, P <.001; NCR and enhancing tumor U = 48.5, Z = -6.688, P <.001; NCR and necrosis U <.001, Z = -6.708, P <.001; edema and enhancing tumor U = 372.5, Z = -2.962, P = 0.003; edema and necrosis U = 98, Z = -5.317, P <.001; and enhancing tumor and necrosis U = 248.5, Z = -3.299, P = 0.001. [Table 1] shows median and quartiles of FA values at selected region (non-normal distribution). [Figure 3] depics the histograms of FA values at selected regions.
The Kaplan–Meier test calculated the median OS for all patients at 19.3 months. Comparison of OS curves among subgroups I to IV (MRI-based location) using the Log-Rank analysis showed a significant difference, χ2 (3) = 8.48, P = 0.037. [Table 2] depicts the median estimate for each subgroup using MRI-based location. [Figure 4] shows the survival curves for each subset using MRI-based location; black dashed line identifies the median (50%) survival time in each group, median survival in groups II and IV was less than one year, while in groups I and III surpassed 18 months.
For each tumor region, we performed independent Kaplan-Meier analyses comparing survival curves among FA values belonging to four quartile subgroups. None of the tumor regions showed significant difference between OS curves of FA quartiles: peritumoral edema region χ2 (3) = 2.325, P = 0.508; contrast-enhancing lesion χ2 (3) = 3.348, P = 0.341; necrosis region χ2 (3) = 6.207, P = 0.102.
CA allowed us to measure the association between overall survival (sorted in 6-months subgroups, levels) and the FA values (classified in quartile-intervals subgroups, levels) for each tumor region, by creating contingency tables of few orthogonal factors that corresponded to the most significant amount of variability in the data. Only the association between FA values in the peritumoral edema region and the survival intervals showed significance χ2 (18) = 30.996, P = 0.029; the cumulative proportion of inertia (variability) explained by this model was 77.5%.
Highest scores in dimensions were assigned to the 0–6 months interval which showed FA values above Q3;higher FA values indicate respected axonal fibres. This score was followed by the 7–12 months OScategory which showed 60% of FA values below Q1, 20% between Q1–Q2, and 20% above Q3; meaning shorter survival in patients with lower FA values (lower FA values indicate axonal disruption of fibre tracts due to tumor infiltration). Interestingly, the 25–30 months OS interval depicted 100% of its values between Q2–Q3; and the >36 months OS interval depicted 66.7% of FA values above Q3; both categories meaning longer survival times were associated with FA values above the median.
[Table 3] presents the percentages of participation of each level after matching FA quartiles and OS intervals. The superior part of the table shows the profiles (levels) calculated by rows (horizontal direction); the inferior portion of the chart depicts the profiles calculated by columns (vertical direction). [Table 4] shows an overview of the scores calculated for each dimension and the contributions to mass and inertia by each of the four FA levels measured in the edema region. [Table 5] depicts the overview of the scores calculated for each dimension, and the contributions to mass and inertia assigned to each of the seven OS intervals associated with the edema region.
[Figure 5] depicts the two-dimensions biplot for the classification of FA values in subgroups (based on quartiles), and the overall survival using subsets of 6-months intervals. The dashed ellipses represent the levels between variables with shorter distances from each other, meaning stronger correlations.
In this study we were able to prove the hypothesis posed at the beginning of this article; it provides additional evidence of the significant association between the peritumoral edema measurement of FA with the OS of patients with GB. Our study represents an incremental advance over previously published work by improving the understanding of the association of FA at several tumor locations with the OS of GB patients.
The clinical relevance of our study has several components. Firstly, we compared OS among four subgroups of GB using an MRI-based classification: Group I: CEL extending from the atrium SVZ to the pia. Group II: SVZ-contacting a tumor that does not involve the cortex, Group III: CEL invading the cortex, reaching the pia, but does not touch the SVZ, and Group IV: tumors that “respect” both the SVZ and cortex; and found a significant difference among groups. GB patients from group I presented a longer median survival time. These findings might seem contrarian to the MRI-based classification of GB, based on their involvement in the SVZ and cortex, where GB in the zone I were described more likely to be multifocal at the time of initial diagnosis and more likely to have a recurrent tumor at locations distant from the initial lesion; however, we did not have enough patients in each subgroup hence we cannot conclude this finding. Secondly, we examined the FA of each tumor region (necrosis, enhancing tumor, and edema regions) grouped by FA-values intervals using quartiles (Q1, Q2, Q3) with no significant difference found among ranges and tumor regions. Lastly, we present a graphical display of the association between FA measures and groups of global survival by using correspondence analysis; this assessment was significant for the edema region: lower FA values indicate axonal disruption of fibre tracts due to tumor infiltration and were associated with lower survival times. Several studies have focused on the evaluation of multiparametric MRI data for GBM grading and characterization.,,, Other studies correlated DTI metrics with survival analysis;,, but, none of them to our knowledge associated FA values of selected tumor regions with intervals of OS using CA.
Our findings reveal cumulative evidence about the usefulness of FA as an OS biomarker. A lower survival (7–12 months interval) was associated with FA values below 25% (Q1). On the other hand, maximum observed OS (more than 36 months) was associated directly with FA values overlapping with those of contralateral normal tissue, [Figure 2]; meaning that higher FA values indicate respected axonal fibres. Peritumoral edema in MRI images is relevant, considering the study by Schoenegger et al. They found that when peritumoral edema was classified as minor (<1 cm), and major (>1 cm), patients with the presence of significant peritumoral edema had significantly shorter overall survival compared to patients with minor edema; which reveals its role as an independent prognostic factor. Also, a higher degree of necrosis and vascular endothelial growth factor (VEGF) expression correlates with distant edema. Our group previously proved that FA depicts a better diagnostic performance in the immediate zone of edema  Its mechanism could be the decrement in value related to an increase in extracellular space secondary to neuronal and fibre tract destruction ; or to a reduction in extracellular space secondary to tumor infiltration. The presence of edema in stereotactic radiosurgery planning is also a primary concern because the target volume definition covers all peritumoral edema plus a 1–2 cm margin. Recent studies have associated the edema regions with high choline peaks indicating the possibility of microscopic higher tumor grade infiltration in this area. A recent study showed that nonenhancing regions had the highest content of viable tumor cells when compared with enhancing and necrotic regions. We believe these regions with an infiltrative tumor and high cellularity, affect FA estimation and survival and should be considered in resection planning. Then, it is realistic to suppose that target volumes often consist of a mixture of tumor cells and nontumoral tissue. Furthermore, an increased risk of the development of incremental edema after Stereotactic Radiosurgery (SRS) associates with pretreatment peritumoral edema, in many cases causing a seizure, aggravating headache, or hemiparesis, probably affecting the activities of daily living adversely.
A marked heterogeneity characterizes GB. Tumor cells frequently contain multiple areas of variable histologic features, so that sampling error in a biopsy may mean that the degree of malignancy seen by the neuropathologist may not reflect the degree of malignancy present elsewhere in a tumor, resulting in significant under-rating of some lesions. Also, even when all radiologically visible portions of a tumor have been excised, molecular methods have also been unable to demonstrate the tumor margin, thus further neoplastic growth can (and usually does) occur in the adjacent brain tissue, leading from microscopic residual to gross recurrence. As a consequence, surgery usually reduces only a tumor; this information is relevant as recent evidence has proved that gross total resection (surgical margin status) significantly correlates with overall survival in patients with GB., GB should be considered as a whole-brain disease with radiotherapy and chemotherapy following surgery.
We acknowledge some limitations in this study. To minimize bias we selected a homogeneous sample of patients from the same hospital, study protocol, and with information collected using the same information criteria for all the subjects. In addition, we removed incomplete records; however, the pathologic diagnosis was limited from a tissue sample and literature shows that lesions for which the risks of biopsy are high cannot be accurately diagnosed and graded. Although FA has demonstrated its reliability as an imaging biomarker in other brain pathologies such as ischemia  and Alzheimer's disease ; the general acceptance of DTI-derived biomarkers by the medical community is still limited. FA does not even reach the anatomic validity of the myeloarchitectonic studies yet, and cannot currently differentiate between individual axons or synaptic connections. An extensive discussion about the mechanism of FA changes in GB, and reasons for the relationship between FA and OS, and its morphological basis are beyond the scope of this study. However, we can mention that mapping FA values across tumor region is relevant because at present we cannot separate glioblastomas based on their extent of invasion; gliomas preferentially spread along white matter tracts (which affects FA), in contrast with metastases which tend to spread along vascular planes. MRI perfusion and spectroscopy  are becoming routinely used methods to locate high-malignancy regions in GB that should be biopsied. Due to time constraints in the acquisition of MRI protocol, most of our patients could not have these sequences included; this fact points out that the development of specific and sensitive biomarkers in GB remains a critical unmet need. We did not add other variables associated with survival (progression-free survival, recurrence pattern) as they were beyond the scope of this study. This study is part of a research line focused on the study of imaging biomarkers and surgical treatments in GB. CA is still not as familiar to researchers in clinical specialities as are other multivariate statistical analysis; it offers advantages over different methods such as principal component analysis by enabling the joint projection of observations and variables to the same low-dimensional factorial subspace. Unlike principal component analysis, that group variable, the CA groups the variable categories (levels); there are no assumptions like normality or a linear relationship. CA simplifies complex data and provides a detailed description of information that can be used for metric as well as contingency tables; in CA, each cell corresponds to the number of occurrences associating the corresponding line and column.
In conclusion, there is a significant association between the peritumoral edema measurement of FA with intervals of OS in patients with GB. We believe the information provided may help radiologist and bioimaging experts to explain the FA clinical applicability to clinicians [neurosurgeons, neurologists, oncologists, psychiatrists, and paediatricians]. Since FA has been accepted as an imaging biomarker in other brain pathologies; we expected in the short term that quantitative analysis using several DTI-derived tensor metrics might be recognized as a routine assessment in GB for screening, triaging and evaluating treatment response.
Eduardo Flores-Alvarez, Coral Durand-Muñoz, and Filiberto Cortes-Hernandez were short-term research fellows at the Radio neurosurgery unit of the National Institute of Neurology and Neurosurgery in 2015.
Ernesto Roldan-Valadez carried out the statistics analyses for the preparation of this manuscript.
Ernesto Roldan-Valadez countersigned as 'guarantor' for the entire study.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5]
[Table 1], [Table 2], [Table 3], [Table 4], [Table 5]