Assessment of Pre-operative Measurements of Tumor Size by MRI Methods as Survival Predictors in Wild Type IDH Glioblastoma

Objective: We evaluate the performance of three MRI methods to determine non-invasively tumor size, as overall survival (OS) and Progression Free Survival (PFS) predictors, in a cohort of wild type, IDH negative, glioblastoma patients. Investigated protocols included bidimensional (2D) diameter measurements, and three-dimensional (3D) estimations by the ellipsoid or semi-automatic segmentation methods. Methods: We investigated OS in a cohort of 44 patients diagnosed with wild type IDH glioblastoma (58.2 ± 11.4 years, 1.9/1 male/female) treated with neurosurgical resection followed by adjuvant chemo and radiotherapy. Pre-operative MRI images were evaluated to determine tumor mass area and volume, gadolinium enhancement volume, necrosis volume, and FLAIR-T2 hyper-intensity area and volume. We implemented then multivariate Cox statistical analysis to select optimal predictors for OS and PFS. Results: Median OS was 16 months (1–42 months), ranging from 9 ± 2.4 months in patients over 65 years, to 18 ± 1.6 months in younger ones. Patients with tumors carrying O6-methylguanin-DNA-methyltransferase (MGMT) methylation survived 30 ± 5.2 vs. 13 ± 2.5 months in non-methylated. Our study evidenced high and positive correlations among the results of the three methods to determine tumor size. FLAIR-T2 hyper-intensity areas (2D) and volumes (3D) were also similar as determined by the three methods. Cox proportional hazards analysis with the 2D and 3D methods indicated that OS was associated to age ≥ 65 years (HR 2.70, 2.94, and 3.16), MGMT methylation (HR 2.98, 3.07, and 2.90), and FLAIR-T2 ≥ 2,000 mm2 or ≥60 cm3 (HR 4.16, 3.93, and 3.72), respectively. Other variables including necrosis, tumor mass, necrosis/tumor ratio, and FLAIR/tumor ratio were not significantly correlated with OS. Conclusion: Our results reveal a high correlation among measurements of tumor size performed with the three methods. Pre-operative FLAIR-T2 hyperintensity area and volumes provided, independently of the measurement method, the optimal neuroimaging features predicting OS in primary glioblastoma patients, followed by age ≥ 65 years and MGMT methylation.


INTRODUCTION
Glioblastomas are the most aggressive and frequent primary tumors of the central nervous system in adult populations, with an approximate incidence of 3-5/100,000 people (1,2). Overall Survival (OS) predictions around 12-18 months are normally expected, remaining poor despite surgical, and adjuvant chemo-and radio-therapy treatments (3,4). Glioblastomas are currently classified in primary or secondary subtypes, reflecting either a "de novo" origin without evidence of any preceding lesion or, a "progression from a lower grade astrocytoma, " respectively (5). Both subtypes have very different OS profiles, with longer survival estimates for the secondary. The molecular signature discriminating between primary (IDH wild type), or secondary subtypes, is the absence or presence, of the isocitrate dehydrogenase (IDH) mutation (5-7), respectively. Methylation of the O6 methylguanine DNA Methyltransferase (MGMT) promoter represents the next molecular feature in prognostic relevance, predicting a better response to alkylating agents like temozolamide (TMZ), during adjuvant chemotherapy (8,9).
Both, clinical and neuroimaging variables, have been repeatedly recommended as prognostic factors (10,11). More specifically, non-invasive measurements of the tumor size have played a controversial role in OS and Progression Free Survival (PFS) predictions of glioblastoma patients (12)(13)(14)(15)(16)(17)(18)(19). Briefly, Magnetic Resonance Imaging (MRI) measurements of twodimensional (2D) diameters of the tumor through the imaging plane were recommended initially by Macdonald et al. (20), and later incorporated by the Response Assessment Neuro-Oncology (RANO) working group (21), as criteria of response to therapy. However, the efficacy of the 2D protocol remains currently under debate, particularly, in morphologically irregular tumors (22)(23)(24). Almost in parallel, the 3D ellipsoid method, calculating the tumor volume (25,26) and using three orthogonal diameters, was implemented for different tumors, including glioblastoma. However, the irregular shape of most tumors continued to limit accurate volume determinations using this approach (24,27). More recently, advanced 3D image-processing software packages using semi-automatic segmentation algorithms have become available, providing more accurate estimations of tumor volumes, with the ability to measure irregular tumor shapes and their compartments (22,28). Nevertheless, a main question arises on whether this increased accuracy in the volume measurement results in any benefit in terms of prognosis, as compared to the other two, simpler to implement, methods (23).
Notably, OS and PFS in glioblastomas have been previously associated to various compartmental tumor volumes including: the volume of necrosis (12)(13)(14), the volume of contrastenhancing tumor (11), the volume of FLAIR-T 2 hyperintensity (13)(14)(15)(16)(17)(18), or the tumor/necrosis volume ratio (13,19). Unfortunately, results and conclusions from many of these studies remain contradictory. Discrepancies may originate, at least partially, from the diversity of volumetric techniques implemented and, in some cases, from incomplete morphological or molecular characterizations of the tumors investigated. Together, these limitations result in a plethora of uncertainties, preventing more accurate predictions (6,13). On these grounds, assessment of the influence of the different methods of presurgical tumor volume determination, their relationship with the underlying clinical and genetic biomarkers, or even with the efficacy of the neurosurgical resections, entail vital relevance to improve the accuracy of current OS and PFS predictions.
On these grounds, we aimed here to assess the influence of different methods of determination of the tumor size, tumor compartments, genetic profile, and some relevant clinical and neurosurgical variables, in the prediction of OS and PFS in primary glioblastomas. To this end, we investigated OS and PFS in a consecutive cohort of patients with primary IDH wild-type glioblastomas, selected as candidates for neurosurgical resection, followed by adjuvant chemo-and radiotherapy. Then, we compared measurements of the tumoral size and its compartments by the 2D diameter, the 3D ellipsoid, and the 3D semi-automatic (Smartbrush R software by Brainlab AG) methods, and correlated them with OS and PFS, considering also the profile of IDH and MGMT mutations, clinical variables as anatomical location or eloquence, and the extension of resection. We found high correlations among measurements of tumoral size by the three methods, with pre-operative FLAIR-T 2 hyperintensity areas and volumes providing the optimal neuroimaging biomarker to predict OS in primary glioblastoma patients, together with age ≥ 65 years and MGMT methylation.

Study Design
Our study was conducted in the period 2015-2019, with the consecutive inclusion of adult patients with newly diagnosed IDH wild-type glioblastomas (Grade IV WHO), treated in the Neurosurgery Department of the University Hospital La Paz, Madrid, Spain (https://www.comunidad.madrid/hospital/ lapaz/). The protocol was approved by the institutional ethics committee, followed the rules of the Helsinki Declaration, and complied with the STROBE checklist (29).
All consecutive adult patients with suspected high-grade glioma, candidates for surgical resection were initially included. Thereafter, 44 patients were selected, with a definitive diagnosis of glioblastoma in the absence of IDH mutation. Inclusion criteria were: i) Pre-operative MRI, at least 72 h after treatment with dexamethasone (4 mg, every 8 h), ii) Post-operative MRI, between 24 and 48 h after surgery, to assess the presence of tumor in contrast enhancement tissue (30,31) and, iii) Radiotherapy (2.0 grays Gy/day, 60 Gy total) and TMZ (75 mg/m 2 oral daily) according to Stupp (32), followed by adjuvant TMZ (150-200 mg/m 2 oral for 5 days during each 28-day cycle) with 6-12 cycles depending on the therapeutic response.
The exclusion criteria were patients who could not be offered the standard treatment (resective surgery followed by chemoradiotherapy) because of very high surgical risk, or also those who refused surgical or chemo-and/or-radiotherapy treatments.

Tumor Size Analysis
Quantitative measurements were performed by two neurosurgeons who jointly carried out the determinations. Representative pre-operative MRI measurements of tumor size measurements are illustrated in Figure 1, implementing either; i) 2D diameters method: as the sum of the products of perpendicular diameters of all contrast-enhancing lesions in T 1 , as recommended by RANO (23) (Figure 2), ii) Orthogonal ellipsoid method: as the product of the longest perpendicular diameters in axial (a), sagittal (b), and coronal (c) sections divided by two (a × b × c/2), preferred to the formula 4/3 pi (a/2 × b/2 × c/2) for its easier and more reliable implementation (27,33) (Figure 3); and iii) Manually selecting the region of interest (ROI) using the Smartbrush tool of semi-automatic segmentation software (BrainLAB R , Munich Germany; Figure 4).

Immunohistochemistry and Genetic Assessment
Anatomopathological analysis of biopsies obtained during neurosurgery included the immunohistochemical detection of the IDH (isocitrate dehydrogenase R132H) mutation. In negative cases, with high suspicion of IDH mutation no-R132H, a genetic study was performed by pyrosequencing (34). All cases underwent analysis of the methylation status of the O 6 methylguanine-DNA methyltransferase (MGMT) promoter using the MethyLigh PCR reaction (35).

Statistics
Univariate statistical analyses included analysis of means, median and standard deviation. Bivariate analysis were     Tumor Size and Compartmental Correlation Analysis: Diameters, Ellipsoid, and Semi-automatic Methods Table 2 summarizes measurements of different tumor compartmental areas and volumes with the three methods investigated, while Figure 5 provides the correlations within them. Most of the compartmental volumes determined by the 3D ellipsoid and automatic segmentation methods were similar, revealing consistency. Similarly, using the sum of the products of major perpendicular diameters, the mean of tumor area had a high correlation with the tumor mass volume measured by the ellipsoid (r = 0.91, p < 0.001), and semiautomatic segmentation (r = 0.82, p < 0.001) methods. The mean FLAIR-T 2 area also highly correlated with the FLAIR-T 2 volume by the ellipsoid method (r = 0.95, p < 0.001), and the semi-automatic segmentation (r = 0.90, p < 0.001) methods ( Figure 5). The mean of the enhancing tumor and of the necrosis volumes were only assessed by the semi-automatic segmentation method, since the irregular morphology hampered the use of the 2D diameter method and the ellipsoid formula.

RESULTS
We found a high correlation between the means of the tumor mass volume (Figure 5E, r = 0.89, p < 0.001) and between the means of the FLAIR-T 2 volumes (Figure 5F, r = 0.91, p < 0.001) by both methods. The correlation between the means of the FLAIR-T 2 /tumor ratio were not as high as the primary measurements, between the 2D diameter and the ellipsoid method was (r = 0.69, p < 0.001), between ellipsoid and semi-automatic methods was (r = 0.84, p < 0.001) and  between 2D diameter and semi-automatic method was (r = 0.56, p < 0.001). The necrosis/tumor ratio could only be obtained by the semi-automatic method and therefore could not be correlated with other methods. Figure 6 illustrates Kaplan-Meier profiles of OS depending on the age of the patients and methylation status of the tumors. Median OS of the investigated cohort was 16 ± 9.1 months (range 1-42 months). Gender, Karnofsky status, location of the tumor, eloquence and extent of tumor resection, had no impact on OS, although tumors with a extent of resection higher than 98% showed a tendency to greater survival, almost reaching statistical significance (18 ± 2.6 vs. 14 ± 4.2 months, p = 0.497). The following features merit further comments.

Survival Analysis
Age. Elderly were associated with a higher mortality in the bivariate analysis, being 18 ± 1.6 months in patients < 65 years, and 9 ± 2.4 months in patients ≥ 65 years (Figure 6A, HR 2.76, p = 0.008).
MGMT methylation was associated with longer OS, 30 ± 5.2 months in methylated vs. 13 ± 2.5 months in non-methylated ( Figure 6B Hyperintensity in FLAIR-T 2 involving the corpus callosum had a significant impact in OS, 10 ± 1.08 months when there was involvement of the corpus callosum vs. 19 ± 3.7 months when there was no involvement of the corpus callosum ( Figure 6C, HR = 3.22, p = 0.006).
Gadolinium enhancing tumor, necrosis, and the tumor mass volumes calculated by the 3D ellipsoid and 3D semi-automatic methods did not have any impact on OS and PFS, and the tumor mass area, calculated with the 2D diameter method, was not associated with OS or PFS either. Figure 7 and Tables 3-5, summarizes Kaplan-Meier tests of OS and PFS with pre-operative FLAIR-T 2 volume measurements by the 3D ellipsoid, and semiautomatic segmentation methods, respectively. FLAIR-T 2 volume ≥ 60 cm 3 by the semi-automatic method was associated with shorter OS, with median survival times of 13 ± 1.8 m (≥60 cm 3 ) vs. 24 ± 6.5 m (< 60 cm 3 ), respectively ( Figure 7B, HR = 2.84, p = 0.024). A similar impact was observed with the 3D ellipsoid method, with the median survival ranging from 15 ± 1.8 (≥60 cm 3 ) to 24 ± 5.8 m (<60 cm 3 ), (Figure 7A, HR = 2.35, p = 0.051), reaching closely statistical significance. The FLAIR-T 2 volume ≥ 60 cm 3 had also impact in progression-free survival by both 3D methods (Figures 7C,D); FLAIR-T 2 area ≥ 2,000 mm 2 as determined by the 2D diameter method was also associated with shorter OS (HR = 3.06, p = 0.041).
Necrosis/Tumor ratio had no impact on survival (p = 0.798). The FLAIR-T 2 /tumor ratio by the semi-automatic method had a negative impact on survival (p = 0.008) without detecting such an association with the 3D ellipsoid and 2D diameter methods.

Multivariate Analysis
Finally, we implemented a multivariate analysis (Cox Regression analysis) strategy to identify the independent predictors of OS among the demographic, imaging, and genetic variables that were significant in the bivariate analysis. We carried out the analysis individualizing the results for each method ( Tables 3-5). The following variables were involved; age ≥ 65 years, MGMT methylation status, FLAIR-T 2 abnormality ≥ 60 cm 3 or ≥ 2,000 mm 3 and FLAIR-T 2 /tumor ratio ≥ 5 for all methods.  The variable Hyperintensity in FLAIR-T 2 involving the corpus callosum was excluded because did not provide more significance to the multivariate models due to the close association with the variable FLAIR-T 2 volume > 60 cm 3 (Fisher's exact test p = 0.009). We found that independence of the variables FLAIR-T 2 > 60 cm 3 , age > 65 years, as well as the MGMT methylation status, remained significant with a similar impact on OS than in the bivariate analysis.

DISCUSSION
Our pilot study investigates the prognostic value of three different methods of pre-surgical tumor size measurement by MRI, combined with relevant clinical and genetic information, as OS and PFS predictors. Briefly, we selected a homogeneous cohort of patients harboring primary glioblastomas (IDH wild type), who were candidates for resective surgery and adjuvant chemoradiotherapy. Tumor sizes and compartments were determined pre-operatively using the 2D diameter method recommended by the RANO (23,24), the 3D ellipsoid formula (25,26) and the semi-automatic 3D segmentation (33) methods. Clinical and genetic variables, including gender, age, expression of IDH mutation, and MGMT status, were investigated additionally. In general, our results confirm the correlation of elderly patients with shorter OS, with the cut-off point in terms of survival prediction in 65 years, as well as the methylated status of the MGMT promoter with longer OS, two well-known variables normally associated with OS and PFS (6,11). In addition, incomplete neurosurgical resections, considered as important negative survival markers (4,(38)(39)(40), depicted in the present study, only a trend to shorter OS without statistical significance, a finding probably related to the relatively small sample of the present patient cohort.
As far as pre-operative measurement of tumor size is concerned, the three methods employed in this study presented a high correlation (41,42). Despite the semi-automatic segmentation method is currently considered the most accurate, because of its high adaptability to irregular morphologies, and the possibility to determine, separately, the different compartmental volumes (27,43), its advantages in terms of prognosis over the 2D diameter and the 3D ellipsoid methods, did not reach statistical significance, even in measurements of tumor size in pre-operative images. This may be caused, at least in part, by the relatively homogeneous tumor morphology in the pre-operative period. However, post-surgical images show limits and contours much more difficult to be silhouetted, and consequently, the semiautomatic method is the only one that can provide insight of the actual volume of the post-surgical residual tumor volume.
Among the areas, volumes and ratios investigated, including necrosis, gadolinium enhancement in T 1 , tumor mass, necrosis/tumor, and FLAIR-T 2 /tumor ratios, only FLAIR-T 2 hyper-intensity had an important impact in OS and PFS, with the three methods of tumor size measurement used. Hyper-intensity in FLAIR-T 2 surrounding tumor mass includes a mixture of peritumoral reactive edema and tumor cells with variable density, as has been previously characterized (44)(45)(46). Present results confirm that, the higher the volume of the FLAIR-T 2 image, the worse the prognosis is. However, there is some controversy in this respect, since some studies have shown correlations between FLAIR-T 2 hyper-intensities with longer OS (13-18, 40, 47), while others have not been able to prove it (12,13,48,49). Many factors, including the non-exclusion of cystic cavities, the influence of corticosteroid treatment, and the inclusion of both, IDH-mutated, and IDH wild-type tumors, may further underlie these discrepancies.
Additionally, some studies have reported associations with survival of the necrosis volume, enhancement tumor volume, and the tumor/necrosis ratio (13,18,48). Our study could not confirm these, either in either bivariate or multivariate analyses. However, we observed a trend indicating that the volume of FLAIR-T 2 hyper-intensity, tumor mass, and necrosis, were greater in the tumors with non-methylated MGMT promoter (50). The lower survival of this glioblastoma subgroup is clearly described by its greater resistance to alkylating chemotherapeutic agents, but it could also involve a higher rate of tumor growth, with higher volumes being observed in all measurements when diagnosing the disease, and before any treatment prescribed. Future studies with a larger sample size could contribute to clarify this finding.
Summarizing, our study shows that the size of FLAIR-T 2 hyper-intensity, as measured by the three methods, presents a similar significant impact in the OS and PFS under both, bivariate and multivariate analyses. The larger the FLAIR-T 2 hyper-intensity volume, the shorter OS and PFS predictions. On these basis, surgical resection of the largest part of the FLAIR-T 2 hyperintensity tissue might determine a better prognosis, in agreement with previous studies (40,51,52). Additionally we show, that the 3D semi-automatic method, more complex and time consuming than the other alternatives (22,53), ends up providing similar pre-operative tumor volumetry results, without any preponderance in terms of survival. Consequently, our study supports the use of the 2D diameter and 3D ellipsoid methods, to estimate FLAIR-T 2 hyper-intensity areas and volumes, revealed here as parameters with an important impact in OS and PFS predictions.

Limitations
Acknowledged limitations of this pilot study relate to the single center implementation and the reduced size of the patient cohort. Thus, increasing the number of patients examined and extension to multicenter assays would constitute immediate priorities to validate the present observations.

CONCLUSIONS
The present study reveals that pre-operative tumor size estimations in primary glioblastoma, as determined by the 2D diameter and 3D ellipsoid methods, provide similar results to the 3D semi-automatic segmentation method. The three methods reveal that FLAIR-T 2 hyperintensity areas and volumes, provide independent factors closely associated with OS and PFS predictions. The larger the FLAIR-T 2 hyper-intensity volume, the worse OS and PFS prediction. In addition, we emphasize the negative impact of age ≥ 65 years and non-methylation of the MGMT promoter in survival predictions.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Ethics Commitee University Hospital La Paz, Madrid. The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
AP performed most of the surgeries, collected and integrated the patient data, applied the different volumetric methods to the image database, provided the univariate and multivariate statistical analyses, and wrote the first draft of the manuscript. CV collected and integrated the patient and imaging data, implemented the MRI volumetric methods, and contributed to the first draft of the manuscript. AB, CU, and AR contributed in the radiologic assessment of pre-and post-operative MRI and validated the volumetric measurements. MG performed many of the surgeries, validated demographic and histopathological assessments, and contributed to the final version of the manuscript. SC and JR integrated all clinical, neurosurgical, and imaging information and accomplished the writing of the final version, with all authors commenting. All authors contributed to the article and approved the submitted version.

FUNDING
This work was supported in part by grants PI 2017/00361 from Instituto de Salud Carlos III to JR, MG, and AP and by grant B2017/BMD3688 from the Community of Madrid to JR, MG, and SC.