Glioma grading using an optimized T1-weighted dynamic contrast-enhanced magnetic resonance imaging paradigm

Background Glioma grading is a critical procedure for selecting the most effective treatment policy. Biopsy result is the gold standard method for glioma grading, but inherent sampling errors in the biopsy procedure could lead to tumor misclassification. Aim This study evaluated grading performances of a more comprehensive collection of the physiological indices quantified using an optimized dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) paradigm for glioma grading. Methods Thirty-five patients with glioma underwent DCE-MR imaging to evaluate the grading performances of DCE-MRI-derived physiological indices. The statistical differences in the physiological indices between the different grades of gliomas were studied, and the grading performances of these parameters were evaluated using the leave-one-out cross-validation method. Results There were significant statistical differences in DCE-MRI-derived physiological indices between the different grades of gliomas. The mean rCBVs for grade II (low-grade glioma, LGG), grade III, grade IV, and high-grade (HGG) gliomas were 2.03 ± 0.78, 3.61 ± 1.64, 7.14 ± 3.19, and 5.28 ± 3.02, respectively. The mean rCBFs of 1.94 ± 0.97, 2.67 ± 0.96, 4.57 ± 1.77, and 3.57 ± 1.68 were, respectively, quantified for grade II (LGG), grade III, grade IV, and high-grade gliomas. The leave-one-out cross-validation method indicates that the grades of glioma tumors could be determined based on a specific threshold for each physiological index; for example, the optimal cutoff values for rCBF, rCBV, Ktrans, Kep, and Vp indices to distinguish between HGGs and LGGs were 2.11, 2.80, 0.025 mL/g min, 0.29 min −1 , and 0.065 mL/g, respectively. Conclusions From the results, it could be concluded that glioma grades could be determined using DCE-MRI-derived physiological indices with an acceptable agreement with histopathological results.


Background
Gliomas are the most common primary cerebral neoplasms which are categorized as highly vascularized malignant tumors [1][2][3][4].More than half of all brain tumors in patients are gliomas, approximately 53% [5].Brain tumors are classified according to their morphological, immunochemical, and molecular characteristics.In the World Health Organization (WHO) classification Page 2 of 12 Abdi Egypt J Radiol Nucl Med (2024) 55: 37 criteria, tumors are assigned a grade based on the histopathological features and immunohistochemical evaluations [6,7].Glioma grading is a critical procedure in selecting the most effective therapy policy.High-grade gliomas are usually treated by adjuvant radiation therapy and chemotherapy (after surgery) with a more aggressive treatment plan, whereas low-grade glioma would be differently treated [8].
Biopsy results are the gold standard method for glioma grading, but inherent sampling errors in the biopsy procedure could lead to tumor misclassification [9][10][11].Gliomas are typically heterogeneous.If the biopsy sites are not properly selected or the biopsy samples have been too small, a lower grade might be assigned to the tumor.These erroneously assignments lead to selecting a nonoptimal therapeutic strategy [12,13].There is an increasing interest in other complementary techniques such as imaging approaches.The magnetic resonance imaging (MRI) method is the most common imaging modality in the evaluation of brain tumors.Conventional MRI techniques have inherent limitations in evaluating the proliferation potential of the tumors [14,15].Advanced MRI methods are required to investigate the microvascular, angiogenesis, metabolism, micronecrosis, and cellularity characteristics of tumors.The bio-imaging markers can provide valuable supplementary information for glioma grading [13].Recently, several sophisticated MRI techniques have been introduced that allow assessing the metabolic and physiological characteristics of the brain tissues [14,16].
The perfusion weighted-magnetic resonance imaging (PW-MRI) method is one of the clinically most relevant procedures of functional MRI, which is used to assess microvasculatures, neovascularization, and capillary permeability of tumors.The assessment of tumor hemodynamics (including blood flow, blood volume, and vessel permeability) could give considerable insight into the angiogenic process of the tumor and provide additional pathological information for preoperative glioma grading [3,17].
Tumor neo-angiogenesis results in tortuous and leaky vessels due to the lack of muscularis propria, widened interendothelial junctions, and a discontinuous or absent basement membrane.Therefore, the permeability of tumor microvasculature would significantly increase.The permeability indices describe the predominant characteristics of tumor vessels [18].Physiological characteristics of the tumors including microvascular proliferation, aggressive cellular characteristics, and tumor-induced angiogenesis could be indirectly evaluated using perfusion indices [19].
The cerebral hemodynamic and permeability parameters are currently quantified using dynamic susceptibility contrast MRI (DSC-MRI) and dynamic contrast-enhanced MRI (DCE-MRI) approaches, respectively.Pioneer studies have shown that cerebral hemodynamic indices could be accurately quantified using an optimized DCE-MRI paradigm [20][21][22].Quantification of tumor hemodynamics (including cerebral hemodynamic and permeability indices) based on single-dose imaging would be a useful alternative approach for tumor assessments, taking into account the cost of double-dose acquisitions and patient safety issues.This study employs an optimized DCE-MRI-based paradigm to quantify cerebral hemodynamic and permeability indices in gliomas and evaluates the performance of the extracted parameters for glioma grading.DCE-MR images were obtained following the administration of gadoteric acid (Dotarem; Guerbet, Paris, France) in a dose of 0.1 mL/kg body weight.The injection was performed using an automatic injector at a rate of 2mL/ second followed by a 15 mL saline flush at the same rate.

DCE-MRI analysis
Motion correction of DCE-MR images was performed using the MCFLIRT function in the FMRIB Software Library (FSL; University of Oxford: http:// www.fmrip.ox.ac.uk/ fsl/).A 3 × 3 mean filter was used for data smoothing, and brain extraction was performed using a semi-automatic MATLAB-based program (ver.2008a, The MathWorks TM, Natick, Massachusetts, United States).
T1W DCE-MRI data were analyzed using in-housedeveloped perfusion software (based on MATLAB software).The permeability indices were quantified based on the modified Kety-Tofts model.The ROIs were determined by a mouse pointer-aided method.For each physiological index, the mean values of the ROIs were registered.

Semi-quantitative analysis
Semi-quantitative indices including the initial area under the curve (IAUC60(mmol/L*Sec)), the peak contrast enhancement (Peak (mmol/L)), and the slope of the timecontrast enhancement curve (Wash-in rate (mmol/L/ Sec)) are quantified using the time-contrast concentration curve.IAUC index is a robust estimation for tissue vascularization [23][24][25].
The area under the time-contrast enhancement curve from the time point of the contrast uptake to 60 s after the onset time was considered as the IAUC60.The trapezoidal method was used for the IAUC60 calculation.The peak is the absolute maximum contrast enhancement for the time-contrast concentration curve.The wash-in rate is the slope of the best-fitted line from the contrast uptake to 10 s after the onset time.The wash-in rate was determined by the sum-of-least-squares method.

Quantification of CBV and CBF indices based on the DCE-MRI data
Cerebral blood volume (CBV) and cerebral blood flow (CBF) indices could be determined based on the T1W DCE-MRI data with good agreement with the DSC-MRIderived cerebral hemodynamic indices [22].In this study, cerebral hemodynamic indices were quantified based on the DCE-MRI using a validated method [20,22,26].Cerebral blood volume (CBV) was measured using Eq.1: where C(t) and C a (t) are the arterial and tissue time-concentration curves, ρ is the brain tissue density (1.04 g/ mL), and H = (1 − Hart)/(1 − Hcap) was applied to differentiate capillary hematocrits (Hcap = 25%) from large vessel hematocrits (Har = 45%).
CBV is the blood volume of the intravascular space.The blood volume of the leakage space has been reported as a part of the CBVuncorrected.Therefore, CBVuncorrected was corrected by the removal of volume contribution of fractional leakage space (Ve) as: (1) CBF index (in mL/100gr.min)was quantified using the following equation: where R is the residual function, and F − 1{} denotes the inverse Fourier transformation.

Patients
Thirty-five patients diagnosed with glioma underwent DCE-MR imaging to assess the performance of DCE-MRI-derived physiological indices for glioma grading.Patients were selected from individuals seeking medical oncologist consultations at Erbil Teaching Hospital.Prior to their participation in the study, informed consent was obtained from all patients.The patients were scanned before any medical interventions, and their glioma grades were determined based on the biopsy results.The demographic information of the patients is summarized in Table 1.The study was approved by the local committee for medical research ethics.
Patients' data were analyzed using the mentioned algorithms and methods in the previous section.For each patient, the region of interest (ROI) was selected on the high perfusion area of the CBV map [27].The relative cerebral hemodynamic changes (rCBV and rCBF) were measured as the mean cerebral hemodynamic magnitude of tumor ROI divided by the mean value in the mirror ROI on the contralateral normal lobe.

Statistical analysis
The normality of the distribution of DCE-MRI-derived physiological indices was checked using the Shapiro-Wilk test.The Mann-Whitney U test and independent student T test analyses were used to evaluate the difference in the parameters between the different grades of gliomas.

Evaluation of DCE-MRI-derived physiological indices for glioma grading
In classification studies, cross-validation methods are used to achieve an optimal classifier.In this method, different classification structures are examined and the classification performances of these structures would be determined.Finally, the classification structure with the best classification performance is chosen.The results of these methods are not reliable when the study sample size is small.The leave-one-out cross-validation method could yield reliable results in such cases.In this study, the leave-one-out cross-validation method is used to (2) evaluate the grading performances of DCE-MRI-derived physiological parameters for glioma grading.
The grading performances of the physiological indices were evaluated for the patients with different glioma grades including grade II (LGG), grade III, grade IV, and high-grade gliomas (HGG, including grade III and grade IV gliomas).
The accuracy, sensitivity, specificity, positive prediction value (PPV), and negative prediction value (NPV) of the physiological indices for glioma grading were determined according to the biopsy results as the gold standard method.
In this study, tumor grade was determined based on the biopsy results and PW-MR imaging data.Kappa index was used to determine the agreement between these grading approaches.The magnitude of the kappa index is ranged from zero to 1.There is a better agreement between the two grading methods when the Kappa coefficient is closer to 1.The kappa coefficient is calculated using Eq. 4. where P(a) and P(e) are, respectively, the observed and expected agreements between the tumor grades determined using the imaging indices and biopsy results.P(a) and P(e) were calculated using the following equations: where TH is the number of patients that were correctly classified using the proposed method as higher-grade glioma (compared with the pathological results), TL is the number of patients that were correctly classified using the proposed method as lower-grade glioma, FH is the number of patients that were wrongly classified using the proposed method as higher-grade glioma, and FL is the number of patients that were wrongly classified using the proposed method as lower-grade glioma.
The grading performances of DCE-MRI-derived physiological indices for glioma grading were also investigated based on a unique classification score including cerebral hemodynamic (rCBV and rCBF), permeability (Ktrans, Kep, etc.), and semi-quantitative (IAUC60 and Peak) indices.In the unique classification score, the weights of the physiological indices were assumed to be the same and equal to 1.
The grading performance of the unique classification score for glioma grading was determined using the method described in Seeger et al. and Matsusue et al. studies [28,29].In this method, a grade is assigned to the tumor of each patient.If the patient is classified as a subject with a higher grade glioma, the value of 1 assigned and the value of zero would be assigned to the patient with a lower grade glioma.For each patient, the assigned values are summed.If the achieved value is greater than 3, the patient is classified as a subject with a higher grade of glioma.Otherwise, the patient's tumor would be considered a lower-grade glioma.The classification metrics of this grading system (including kappa coefficient, accuracy, sensitivity, etc.) were also determined using the biopsy results as the gold standard method.

Results
DCE-MRI data were analyzed using a valid method and the perfusion maps including cerebral hemodynamic (CBV and CBF), permeability (Ktrans, Kep, etc.), and semi-quantitative (IAUC60, Peak, etc.) indices were extracted.The exemplary maps achieved for a 57-yearold woman with glioblastoma multiforme (GBM) are shown in Fig. 1.

The statistical differences of the physiological indices between the different grades of gliomas
The Mann-Whitney U test and independent student T test were used to evaluate the statistical differences of the cerebral hemodynamic (rCBV and rCBF), permeability (Ktrans, Kep, etc.), and semi-quantitative (Peak, IAUC60, etc.) indices between the different grades of gliomas.
The levels of statistical significance for the cerebral hemodynamic (rCBV and rCBF), permeability (Ktrans, Kep, etc.), and semi-quantitative (IAUC60, Peak, etc.) indices between different grades of gliomas are listed in Table 2.

The grading performances of the physiological indices for glioma grading
The optimal thresholds of the cerebral hemodynamic (rCBV and rCBF), permeability (Ktrans, Kep, etc.), and semi-quantitative (Peak, IAUC60, etc.) indices for glioma grading are listed in Tables 3, 4, 5.The accuracy, sensitivity, specificity, PPV, NPV, and Kappa coefficient magnitudes of the physiological indices were also listed in these tables.
The accuracy, sensitivity, specificity, PPV, NPV, and Kappa coefficient magnitudes of the unique grading score for glioma grading are listed in Table 6.

Discussion
Cellularity characteristics and tissue vascularity are two of the main factors that would be considered for glioma grading.Cerebral hemodynamic indices indirectly measure tumor-induced angiogenesis and its microvascular proliferation [19,22].Therefore, cerebral hemodynamic indices could be used as bio-imaging markers for tumor grading.In most clinical centers, especially in less developed countries, the cerebral hemodynamic and permeability parameters are currently quantified using a double-dose dynamic imaging procedure including   DSC-MRI and DCE-MRI techniques.In this study, an optimized DCE-MRI paradigm was used that could simultaneously present both cerebral hemodynamic and permeability parameters using a single dose acquisition.
To evaluate the clinical significances of the parameters quantified using this method, the achieved magnitudes for different glioma grades and their performances for glioma grading are compared with reported values.For high-grade gliomas, high CBV and CBF areas are more expected in cerebral hemodynamic maps [30,31].In our study, higher cerebral hemodynamic magnitudes were quantified for the high-grade gliomas (rCBV: 5.28 ± 3.02 vs. 2.03 ± 0.78; and rCBF: 3.57 ± 1.68 vs. 1.94 ± 0.97) that reflects higher angiogenic activity for high-grade gliomas.These parameters could indicate the metabolism and vascularity characteristics of the tumor [31].
In our study, the mean rCBVs for grade II (LGG), grade III, grade IV, and high-grade gliomas were 2.03 ± 0.78, 3.61 ± 1.64, 7.14 ± 3.19, and 5.28 ± 3.02, respectively.The mean rCBFs for grade II (LGG), grade III, grade IV, and high-grade gliomas were, respectively, quantified as 1.94 ± 0.97, 2.67 ± 0.96, 4.57 ± 1.77, and 3.57 ± 1.68.In previous studies, different rCBV and rCBF values were reported for low-and high-grade gliomas.rCBV magnitudes ranged from 0.89 to 3.94 and 2.15 to 9.84 for lowand high-grade gliomas, respectively.The reported rCBF magnitudes for low-and high-grade gliomas, respectively, ranged from 0.85 to 3.79 and 2.55 to 8.26.In Stefano et al. study [27], rCBV values were calculated in three different compartments including the contrast-enhanced area, the non-enhancing tumor, and the high perfusion area on the CBV maps for twenty-one patients with grade III and IV gliomas.The rCBV values for grades III and IV were 3.78 ± 1.70 and 7.51 ± 3.84, respectively.In Saini et al. [32] study, the mean rCBVs for grades II, III, and IV gliomas were, respectively, reported as 1.84 ± 0.57, 4.16 ± 1.49, and 6.09 ± 3.04.Our results were similar to those of these studies.
In the next step, rCBF values quantified for different glioma grades are compared with the reported magnitudes in other studies.In Falk et al. study [33], the grades II and III gliomas were investigated using DSC-and DCE-MR imagings.The mean rCBFs for grades II and III gliomas were 1.66 ± 0.99 and 2.53 ± 1.88, respectively.The glioma grading based on perfusion parameters was also evaluated by Server et al. [34].The mean rCBFs for grades II, III, and IV were, respectively, reported as 2.87 ± 2.14, Table 5 The accuracy, sensitivity, specificity, PPV, NPV, and Kappa coefficient of semi-quantitative (IAUC 60 , Peak, etc.) indices for differentiation between the different grades of gliomas The discriminative thresholds of semi-quantitative indices were also listed in the first column  5.35 ± 2.04, and 5.34 ± 1.93.There was no significant difference between grade III and IV gliomas.For the relative hemodynamic indices, there were significant differences between the different grades of gliomas.Our findings, in agreement with these studies, indicate the statistically significant differences in rCBV and rCBF values between grades II (LGG), grade III, grade IV, and high-grade gliomas (P < 0.027).

Threshold Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%) Kappa
According to the promising observations for the perfusion indices quantified using DCE-MRI data in comparison with the reported studies, in the next step, the classification performances of these parameters are evaluated.Our best rCBF and rCBV thresholds for differentiating between low-and high-grade gliomas were 2.11 (accuracy = 80%, sensitivity = 84.21%,and specificity = 75%) and 2.80 (accuracy of 82.86%, sensitivity = 84.21%,and specificity = 81.25%),respectively.In Ma et al. study [35], ASL and DSC-MR imagings were used for preoperative glioma grading.In this study, twenty-seven low-grade gliomas and twenty-three highgrade gliomas were investigated.For the ASL method, rCBF cut-off of 2.24 with sensitivity = 83.2%,specificity = 77.7%, and AUC = 0.866 was suggested as the optimal threshold.In the DSC-MRI method, the best rCBF cut-off was 1.85 (sensitivity = 91.3%,specificity = 63.9%, and AUC = 0.758).In Caulo et al. study [36], the glioma grading was performed using a multimodal MRI method.For the rCBV cut-off value of 2.59 in contrast-enhanced regions, the sensitivity and specificity values were 80% and 91%, respectively.In a similar study [37], the rCBV index was also used for glioma grading in 160 patients.For the rCBV threshold of 2.97, the sensitivity, specificity, PPV, and NPV were reported as 72.5%, 87.5%, 94.6%, and 51.5%, respectively.
Our cut-off magnitudes for DCE-MRI-derived cerebral hemodynamic indices and their grading performances to discriminate low-and high-grade gliomas were similar to the results of the reported studies.Therefore, it could be concluded that relative cerebral hemodynamic parameters quantified using the DCE-MRI method have acceptable accuracies for differentiating between lowand high-grade gliomas.
The best rCBV thresholds reported for differentiating between grade II and III gliomas and between grade III and IV gliomas were, respectively, 2.4 (sensitivity = 100% and specificity = 92%) and 3.92 (sensitivity = 81% and specificity = 58%) [37].There is good agreement between our threshold magnitudes and their grading performances with the values reported in the literature.The observed compliances between the magnitudes of perfusion indices and their classification performances for glioma grading in our study and those of the reported studies emphasize the clinical efficiencies of the DCE-MRI-derived perfusion parameters for glioma grading.Therefore, the perfusion indices quantified using DCE-MRI data could be promising and efficient classifier factors for glioma grading.
In medical imaging centers, the DCE-MRI method is routinely used to investigate the permeability characteristics of the tissues.In the second part of our study, the permeability indices were quantified for different glioma grades and the classification performances of these permeability indices were evaluated for glioma grading.Higher vascular permeability magnitudes (Ktrans: 0.37 ± 0.4 vs. 0.017 ± 0.02, and Kep: 0.55 ± 0.36 vs. 0.22 ± 0.17) were measured for the high-grade gliomas in agreement with the other studies.Higher proportions of immature blood vessels with high permeability characteristics would result due to increased angiogenesis in highgrade gliomas [32,38].Vascular endothelial permeability in brain tumors reflects valuable information about the blood-brain barrier defects, the characteristics of neoangiogenesis, and vascular morphology for tumoral tissues.Hence, the permeability indices could play an important role in tumor identification, in addition to tumor grading [28,39,40].
In Li et al. study [4], the classification performances of DCE-MRI and SWI methods were investigated for glioma grading.In this study, Ktrans and Ve parameters were evaluated for 32 patients with different grades of gliomas.The mean Ktrans values for grade II, III, and IV gliomas were 0.026 ± 0.019, 0.096 ± 0.063 min −1 , and 135 ± 0.068 min −1 , respectively.For grade II, III, and IV gliomas, the mean Ve values were, respectively, reported as 0.121 ± 0.13, 0.483 ± 0.225, and 0.525 ± 0.18.In Ludemann et al. study [41], permeability parameters were also quantified for forty-one gliomas, six meningiomas, and eight metastases.The mean Ktrans values for grade II, III, and IV gliomas were 1.584 min −1 , 1.338 min −1 , and 1.821 min −1 , respectively.For grade II, III, and IV gliomas, the mean Ve values were, respectively, reported as 18.03, 25.26, and 22.41.
There are considerable differences between the mean permeability magnitudes reported by previous studies.In these studies, different grading cut-off values were also proposed [1,4,18,[41][42][43].These huge discrepancies may be related to the differences in pulse sequences, pharmacokinetic models, dynamic scanning times, and arterial input function (AIF) selection procedures used in the studies [43,44].
There were statistically significant differences in Ktrans and Kep parameters between different grades of gliomas, which was consistent with our results.In Li et al. and Zhao et al. studies [4,42], similar to our findings, there were statistically significant differences in Ve between grade II and IV gliomas and between LGG and HGG groups.There was no significant difference between grade III and IV gliomas, which was also observed in our findings.These observations indicated that the permeability indices can complement histopathology results for tumor grading by providing valuable information about tumor vascular permeability.
In studies like that of Santarosa et al. [18], the doubledose acquisition was used to quantify the rCBV, Vp, and Ktrans parameters.But, in our study, a more comprehensive collection of the physiological indices was measured using a single-dose DCE-MRI acquisition.On the other hand, DCE-MRI is not affected by susceptibility artifacts and has a higher spatial resolution than the conventional DSC-MRI [18,45].Therefore, in addition to financial and patient safety issues, much better grading parameters could be achieved using DCE-MRI-derived physiological indices.
In our study, there were significant differences in DCE-MRI-derived physiological indices (except for Ve) between the different grades of gliomas, and the glioma grades were accurately classified using these parameters.In discriminating the glioma grades, the unique classification score had the best grading performance compared to each physiological index separately.Even though, each of the physiological indices provides useful biological and physiological information about the lesions.For glioma grading based on DCE-MRI-derived physiological indices, a unique classification score consisting of all physiological indices is proposed.The cerebral hemodynamic and permeability indices quantified using the single-dose DCE-MRI could also help to determine the lesion characteristics as a non-invasive method.

Limitations
This study had some limitations.First, this is a singlecenter study and participants were primarily local residents.These limitations restrict the generalizability of the results for other countries.External validation of the proposed paradigm merits future investigations using bigger and multicenter databases.
Second, the DCE-MRI-derived cerebral hemodynamic and permeability indices of different glioma grades and the grading performance of these physiological indices were only evaluated for glioma grading and could not be readily translated to other tumors.

Conclusions
In this study, the grading performances of DCE-MRIderived physiological indices were evaluated for glioma grading.A more comprehensive collection of the physiological parameters including the cerebral hemodynamic and permeability indices could be achieved using a single-dose DCE-MRI approach.The cerebral hemodynamic and permeability indices quantified using the DCE-MRI could determine glioma grades with an acceptable agreement with histopathological results.For glioma grading based on DCE-MRIderived physiological indices, a unique classification score consisting of all physiological indices is proposed.The cerebral hemodynamic and permeability indices quantified using DCE-MRI could also help to determine the lesion characteristics as a non-invasive method.

Fig. 1
Fig.1Exemplary maps achieved for a 57-year-old woman with right temporal glioblastoma multiforme (GBM).Axial post-contrast T1-weighted image shows an enhancing lesion in the right temporal lobe.Axial T2-weighted image shows an ill-defined mass with higher signal intensity than the normal brain tissue.The cerebral hemodynamic (CBV and CBF), semi-quantitative (IAUC 60 , Wash-in rate, and Peak), and permeability (Ktrans, Vb, and Ve except for Kep) maps have shown an enhancing lesion in the right temporal lobe.There is a well-discriminated border around the tumor

Table 1
Demographic and clinical characteristics of the studied patients

Table 3
The accuracy, sensitivity, specificity, PPV, NPV, and Kappa coefficient of cerebral hemodynamic (rCBV and rCBF) indices for differentiation between the different grades of gliomas

Table 4 The
accuracy, sensitivity, specificity, PPV, NPV, and Kappa coefficient of permeability (Ktrans,Kep, etc.)indices for differentiation between the different grades of gliomasThe discriminative thresholds of permeability indices were also listed in the first column

Table 6
The accuracy, sensitivity, specificity, PPV, NPV, and Kappa coefficient of the unique classification score (including cerebral hemodynamic, permeability, and semi-quantitative indices) for differentiation between the different grades of gliomas