Simplified perfusion fraction from diffusion-weighted imaging in preoperative prediction of IDH1 mutation in WHO grade II–III gliomas: comparison with dynamic contrast-enhanced and intravoxel incoherent motion MRI
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.
Introduction
Gliomas, the most common primary intracranial neoplasms in humans, are classified as grade I–IV based on histopathological criteria. Different from grade IV, also known as glioblastoma, the outcome of grade II-III gliomas (lower-grade gliomas, LGGs) are highly variable. Published survival duration of LGGs ranged from 1 to over 15 years, reflecting molecular heterogeneity of these tumors.1, 2, 3, 4, 5 The 2016 revised fourth edition of the World Health Organization (WHO) classification of tumors of the central nervous system defines a large subset of gliomas based on molecular alterations, among which mutation of isocitrate dehydrogenase (IDH1) has shown to be the most important, for this mutation is thought to be a predictor of early steps in gliomag enesis. It has been shown that 70%–90% of LGGs- carry IDH1 mutations, and that IDH1 mutant glioma have a survival benefit associated with the maximal surgical resection, and the use of radiation and chemical therapy.6, 7, 8 Hence, assessing grade II and III gliomas by genetic alteration, which might be helpful for patient prognosis and clinical treatment, is now a common clinical practice.
The IDH1 gene plays an important role in tumor angiogenesis and vasculogenesis, which have been recognized as hallmarks of histopathological growth and progression of gliomas.9,10, 11 Therefore, preoperative assessment of tumor perfusion by MRI may give insight into the IDH1 mutation status, thus aiding in clinical decision making. Several MR perfusion techniques have been developed to evaluate the degree of tissue vascularization. Dynamic contrast-enhanced (DCE) MRI and intravo xel incoherent motion (IVIM) MRI are two common MR perfusion techniques with distinct imaging mechanisms.11, 12, 13, 14, 15
Using rapid T1-weighted imaging to measure the changes resulting from gadolinium contrast agent leakage in and out of the extracellular extravascular space, DCE MRI enables the determination of several hemodynamic parameters, including the volume transfer constant (Ktrans), the extravascular extracellular volume fraction (ve), and the vascular plasma volume fraction (vp).11, 16 Previous studies have demonstrated the clinical potential of DCE MRI in glioma grading and differential diagnosis.17, 18 However, the need for an intravenous injection of contrast agent limits its clinical application in patients with renal dysfunction or individuals who are allergic to gadolinium.
IVIM MRI is a variant of conventional diffusi on-weighted imaging (DWI) in that images at multiple b-values are required to fit the two-component mathematical model. In this model, the effect of microcirculation of blood in the capillary network (characterized by the pseudo-diffusion coefficient D*) is separated from the pure water diffusion component (characterized by the diffusion coefficient D). More than eight b-values are typically needed to fully characterize biexponential signal attenuation, thus increasing the scanning time. Some simplified models based on IVIM theory with fewer b-values have been proposed. Both the full and simplified IVIM models have shown their abilities in characterizing tumor perfusion and assessing the glioma grade.19, 20, 21
The purpose of our study, therefore, was to determine the association of the three b-value DWI-derived simplified perfusion fraction (SPF) with tumor perfusion and to compare the performance with DC E and IVIM MRI-derived parameters in the preoperative prediction of IDH1 mutation status in LGGs using surgical and histopathological findings as a standard of reference.
Patients and methods
Patient enrollment
This prospective single-center study was performed in accordance with the principle of the Declaration of Helsinki and was approved by the local ethics committee. Written informed consent was obtained from all subjects prior to study enrollment. The flowchart of the study design is demonstrated in Figure 1.
Figure 1
Flowchart of study design.
From April 2018 to March 2019, 55 patients who were suspected of primary brain tumors were prospectively enrolled in the study. All patients underwent initial MRI at the same unit and were then underwent neurosurgical resection at our hospital. Excluded from the study were 17 patients with pathological diagnosis other than LGGs, five patients without complete DCE MRI or IVIM data, and three patients due to poor image quality associated with head movement. Finally, a total of 30 patients (13 women, 17 men; average age, 44.73 years; age range, 19–78 years) with histopathologically confirmed LGGs (WHO II glioma, n = 22; WHO III glioma, n = 8) were enrolled. The descriptive statistics are shown in Table 1.
Patient characteristics
Characteristic
IDH1 mutants (n = 18)
IDH1 wild-type (n = 12)
Mean age (y)a
42.8 (22–67)
47.9 (19–78)
Sex distribution (M/F)b
10/8
7/5
WHO grade
II
15
7
III
3
5
Histologic type
Astrocytoma
12
5
Oligodendroglioma
3
0
Oligoastrocytoma
0
1
Anaplastic astrocytoma
1
3
Anaplastic oligodendroglioma
1
2
Anaplastic oligoastrocytoma
1
1
* Mean (range) or count is reported; a = significant difference in age was noted between isocitrate dehydrogenase 1 (IDH1) mutant and wild-type groups (P = 0.020); b = no significant difference in sex distribution was noted between IDH1 mutant and wild-type groups (P = 0.769)
F = female; M = male
MRI acquisition protocols
MRI of all patients was performed on a 3.0-T MRI unit (Signa HDxt; GE Medical Systems, Milwaukee, WI, USA) using a standard 8-channel head coil. The advanced MRI protocol included DCE MRI and DWI with 10 b-values (0–1000 s/mm2). Conventional protocol—T1- and T2-weighted imaging with fast spin-echo sequences (T1WI, T2WI), T2 fluid-attenuated inversion recovery (FLAIR) sequence, and contrast-enhanced T1WI— were performed during the same examination.
Three-dimensional DCE MRI of head was performed after intravenous administration of a gadopentetate dimeglumine (Magnevist; Bayer Healthcare, Berlin, Germany, 0.1 mmol per kilogramof body weight) at a rate of 4 ml/s via a power injector (Spectris; Medrad, Pittsburgh, PA, USA). Precont rast scans with four dynamics were collected before gadopentetate dimeglumine was injected. The detailed parameters of the pre- and postcontrast scans were as follows: repetition time (TR)/echo time (TE), 3.3 ms/1.3 ms, flip angle, 15°; matrix, 256 × 160; field of view (FOV), 220 × 220 mm; section thickness, 2 mm; number of sections, 40; and total scanning time, 4 min.
DWI was acquired before contrast injection. Ten b-values (0, 20, 50, 80, 150, 200, 300, 500, 800, and 1000 s/mm2) were applied with a fat-suppressed single-shot echo-planar sequence in three orthogonal directions sequentially, they were averaged two times, and then trace images were generated. The other imaging parameters were: TR/TE, 3000 ms/106 ms; matrix, 192 × 192; FOV, 260 × 260 mm; section thickness/gap, 5/1 mm; number of signal averages, 2; number of sections, 15. The multi-b-value DWI was acquired at 5 min and 36 s, and if separately, 2 min and 11 s for three-b-value DWI.
MR image analysis
DCE MRI analysis
Pharmacokinetic parameters (Ktrans, ve, vp) were calculated off-line by using commercially available software (MIStar; Apollo Medical Imaging, Melbourne, VIC, Australia) according to the two-compartment Tofts model.22 Preprocessing for the perfusion data included semiautomatic selection of arterial input function (AIF). The AIF was obtained independently for every patient from the intracranial internal carotid artery. Parametric maps of Ktrans, ve, and vp were generated on a pixel-by-pixel basis.
DWI analysis
DWI data were performed with a program in MATLAB (MATLAB 2017a; MathWorks, Natick, MA, USA) programming tool. Full IVIM features - the diffusion coefficient (D), pseudo-diffusion coefficient (D*), and the perfusion fraction (f)—were extracted by fitting the biexponential model using all b-values as follows:
where Shigh is signal intensity at bhigh and Slow is signal intensity at blow, respectively. As the b-value has a differential sensitivity to Brownian motion of water protons, ADC0,200 represents mixed diffusion and perfusion effects and ADC200,1000 is almost purely related to diffusion.23, 24 The b-value scheme was chosen following previous recommendations25, 26, 27 which indicated that the effects of diffusion and microcapillary perfusion are both reflected within low b-values (b < 200 s/mm2), while for higher b-values (b > 200 s/mm2), a large proportion of measured signal in each imaging voxel was caused by tissue diffusion. When a typical b-value (1000 s/ mm2) was used, the contribution of perfusion has faded away entirely. The ADC thus appears to be a sensitive index of diffusion component. On the other hand, since a b-value of 1000 s/mm2 is small enough, high image quality may be guaranteed and the kurtosis effect may be avoided. As the contribution of kurtosis is greater when b-value is beyond 1000 s/mm2.28, 29 Therefore, the relative proportion of the perfusion component in the whole diffusion pool, named SPF, can be determined as follows (20):
The regions of interest (ROIs) were drawn by two readers who have 6(M.C.) and 19(Y.Z.) years of experience in neuroradiology, respectively, and consensus was researched. Both readers were blinded to the histopathological results and other clinical data, including age and gender. Following pre vious studies30, 31, an elliptical ROI (20–340 mm2) was placed by each doctor on parametric maps of the solid tumor area as much as possible to include the portion with the minimum values of diffusion (D and ADC0,1000) and maximum values of perfusion (SPF, f, D*, Ktrans, vp, and ve). For correlation analysis between SPF and other perfusion parameters, the similar-sized ROIs used for SPF images were placed in the corresponding area of DCE images and IVIM images. T1-weighted contrast-enhanced images where contrast agent leakage in tumors was observed were used as a reference to define the ROIs on parametric maps.32, 33 The study used ADC images combined with T1-weighted, T2-weighted, and FLAIR images to determine the ROI of tumor area in nonenhancing lesion. Special care was taken to exclude necrosis, cys ts, hemorrhage, calcification, and intralesional macrovessels.
Statistical analysis
Statistical analysis was performed using commercial software (SPSS version 22, IBM Corporation, Armonk, NY, USA and MedCalc, version 11.4.2.0, MedCalc Software, Mariakerke, Belgium). The relations hip between perfusion parameters was analyzed with Spearman rank correlation. We considered correlation coefficients < 0.4, 0.4–0.6, 0.6–0.8, and > 0.8 to indicate week, moderate, strong, and very strong correlation, respectively. The unpaired t-test and Mann–Whitney U test were used to determine the difference in DWI, DCE and IVIM MRI parameters between WHO grade II and III gliomas, as well as between IDH1 mutant and wild-type gliomas, according to the data normality (Kolmogorov-Smirnov test). ROC curves were constructed to evaluate the ability to identify different IDH1 mutation statuses. Area under the curve (AUC) values of < 0.7, 0.7–0.9, and > 0.9 were considered to indicate low, medium, and high diagnostic performance, respectively. Differences between AUC values were analyzed by using the Delong method (34). Optimal thresholds were determined by maximizing the Youden index ((specificity + sensitivity) − 1). A P-value less than 0.05 was considered to indicate statistical significance.
Results
In terms of histology, 16 patients had astrocytomas, three had oligodendrogliomas, one had an oligoastrocytoma, four had anaplastic astrocytomas, three had anaplastic oligodendrogliomas, and three had anaplastic oligoastrocytomas. Intercorrelation analysis between perfusion parameters revealed a significant association for SPF and f (ρ = 0.768, P < 0.001). The study also found a moderate correlation between SPF and ve (ρ = 0.548, P = 0.002) and between SPF and Ktrans (ρ = 0.535, P = 0.002).
The statistical data of DCE MRI and DWI-derived parameters in differentiating WHO grade II and III gliomas are summarized in Table 2. Perfusion-related parameters including Ktrans, ve, vp, f, D *, and SPF were all significantly higher in WHO grade III gliomas than in WHO grade II gliomas (all P < 0.05), while ADC and D values were both significantly lower in WHO grade III gliomas (both P < 0.05).
Parameters derived from dynamic contrast-enhanced (DCE) MRI and diffusion-weighted imaging (DWI) between WHO grade II and III gliomas
Parameter
Grade II
Grade III
P-value
Ktrans (min−1)
0.067 ± 0.048
0.116 ± 0.064
0.013
ve
0.071 ± 0.057
0.401 ± 0.344
0.018
vp
0.036 ± 0.020
0.051 ± 0.018
0.035
ADC0,1000 (×10−3 mm2/s)
1.093 ± 0.203
0.904 ± 0.184
0.028
SPF (%)
10.78 ± 4.378
16.391 ± 5.471
0.012
D (×10−3 mm2/s)
1.194 ± 0.261
0.949 ± 0.169
0.021
D* (×10−3 mm2/s)
6.692 ± 1.564
8.618 ± 2.215
0.037
f (%)
3.315 ± 1.536
6.380 ± 3.419
0.020
* P-values are considered significant at P < 0.05.
ADC = apparent diffusion coefficient; D = diffusion coefficient; D* = pseudo-diffusion coefficient; f = perfusion fraction; Ktrans = volume transfer constant; ve = extravascular extracellular volume fraction; vp = vascular plasma volume fraction; SPF = simplified perfusion fraction
Representative cases of IDH1 mutant and wild-type LGGs are shown in Figures 2 and 3. The mean values ± standard deviations of DCE MRI and DWI-derived parameters for the IDH1 mutant and wild-type tumors in the whole LGGs group, are summarized in Table 3. Compared with IDH1 mutant LGGs, IDH1 wild-type LGGs exhibited significantly higher perfusion values, that is, Ktrans, ve, vp, f, D*, and SPF (all P < 0.05), and significantly lower diffusion values, that is, ADC and D (both P < 0.05). In the WHO grade II subgroup, vp and SPF differed significantly between IDH1 mutant and wild-type tumors (P = 0.018 and P = 0.049, respectively), whereas in the WHO grade III subgroup, only f showed a significant difference (P = 0.014).
Figure 2
Images obtained in a 44-year-old man with astrocytoma (isocitrate dehydrogenase 1 [IDH1] mutant glioma). (A) Fluid-attenuated inversion recovery (FLAIR) image shows a heterogeneous hyperintense lesion in the right frontal lobe. (B) Apparent diffusion coefficient (ADC)0,1000 map shows increased ADC value in the lesion. (C, D) Intravoxel incoherent motion (IVIM) perfusion fraction (f) and simplified perfusion fraction (SPF) maps show no increased values in the corresponding area of the hyperintense lesion as shown in (A). (E) On contrast-enhanced T1-weighted image, the lesion is non-enhancing. (F–H) Dynamic contrast-enhanced (DCE) MRI parametric maps of volume transfer constant (Ktrans), extravascular extracellular volume fraction (ve) and vascular plasma volume fraction (vp) show no increased values in the lesion. Regions of interest are marked on parametric maps.
Figure 3
Images obtained in a 72-year-old woman with astrocytoma (isocitrate dehydrogenase 1 [IDH1] wildtype glioma). (A) FLAIR shows a heterogeneous hyperintense lesion in the right hemisphere. (B) Apparent diffusion coefficient (ADC)0,1000 map shows a mixed pattern of high and intermediate ADC values in the lesion. (C, D) Intravoxel incoherent motion (IVIM) perfusion fraction (f) and simplified perfusion fraction (SPF) maps show markedly increased f and SPF values in the corresponding area of the contrast-enhanced lesion as shown in (E). (E) On contrast-enhanced T1-weighted image, the lesion is vividly enhanced. (F–H) Dynamic contrast-enhanced (DCE) MRI parametric maps of volume transfer constant (Ktrans), extravascular extracellular volume fraction (ve) and vascular plasma volume fraction (vp) show obviously increased values in the corresponding area of the contrast-enhanced lesion. Regions of interest are marked on parametric maps.
Parameters derived from dynamic contrast-enhanced (DCE) MRI and diffusion-weighted imaging (DWI) between isocitrate dehydrogenase 1 (IDH1) mutant and wild-type gliomas
Parameter
IDH1 mutant
IDH1 wild-type
P-value*
Ktrans (min−1)
0.054 ± 0.024
0.123 ± 0.073
0.007
ve
0.052 ± 0.035
0.121 ± 0.080
0.007
vp
0.032 ± 0.015
0.051 ± 0.022
0.015
ADC0,1000 (×10−3 mm2/s)
1.123 ± 0.185
0.923 ± 0.199
0.009
SPF (%)
9.572 ± 3.437
16.332 ± 4.925
< 0.001
D (×10−3 mm2/s)
1.108 ± 0.245
0.959 ± 0.146
0.047
D* (×10−3 mm2/s)
6.546 ± 1.757
8.196 ± 1.794
0.020
f (%)
3.080 ± 1.581
5.712 ± 2.924
0.005
* P-values are considered significant at P < 0.05
ADC = apparent diffusion coefficient; D = diffusion coefficient; D* = pseudo-diffusion coefficient; Ktrans = volume transfer constant; f = perfusion fraction; SPF = simplified perfusion fraction; v= e extravascular extracellular volume fraction; vp = vascular plasma volume fraction
The results of ROC curve analysis are presented in Figure 4 and Table 4. For differentiation between IDH1 mutant and wild-type LGGs, the ROC curve analysis showed that among all parameters, SPF gave the highest AUC value (0.86), followed by f (0.81) and ADC (0.80), though no significant difference in AUC values was found (P > 0.05). The optimal SPF threshold for IDH1 mutation discrimination was 14.5%, with a sensitivity and specificity of 94.4% and 75.0%, respectively.
Figure 4
Receiver operating characteristic (ROC) curves and corresponding area under the curve values for (A) diffusion-weighted imaging (DWI) parameters (simplified perfusion fraction [SPF], perfusion fraction [f], apparent diffusion coefficient [ADC]0,1000) and (B) dynamic contrast-enhanced (DCE) MRI parameters (transfer constant [Ktrans], extravascular extracellular volume fraction [ve] and vascular plasma volume fraction [vp]) in the differentiation of isocitrate dehydrogenase 1 (IDH1) mutant and wildtype gliomas. SPF showed the highest diagnostic performance with the area under the curve value of 0.86.
Diagnostic performance of parameters for differentiation between isocitrate dehydrogenase 1 (IDH1) mutant and wild-type gliomas
Parameter
AUC (95% CI)
Sensitivity (%)
Specificity (%)
Cutoff value
Ktrans (min−1)
0.773 (0.563–0.983)
77.8
75.0
> 0.062
ve
0.760 (0.569–0.951)
94.4
58.3
> 0.119
vp
0.680 (0.451–0.909)
55.6
91.7
> 0.029
ADC0,1000 (×10−3 mm2/s)
0.718 (0.531–0.904)
83.3
75.0
≤ 1.002
SPF (%)
0.861 (0.686–0.959)
94.4
75.0
> 14.500
D (×10−3 mm2/s)
0.727 (0.541–0.913)
72.2
83.3
> 1.065
D* (×10−3 mm2/s)
0.690 (0.493–0.886)
44.4
91.4
≤ 5.959
f (%)
0.810 (0.658–0.963)
72.2
83.3
> 3.617
ADC = apparent diffusion coefficient; D = diffusion coefficient; D* = pseudo-diffusion coefficient; f = perfusion fraction;*Ktrans = volume transfer constant; SPF = simplified perfusion fraction; ve = extravascular extracellular volume fraction; vp = vascular plasma volume fraction
Discussion
In this study, an analysis of DWI, DCE, and IVIM MRI was performed to evaluate the tissue diffusion and perfusion characteristics to identify histological and molecular profiles of LGGs. Our results showed that diffusion and perfusion metrics exhibited substantial differences between WHO grade II and III gliomas, as well as between IDH1 mutant and wild-type LGGs. Among all parameters, the simplified DWI-derived perfusion fraction showed higher efficacy in IDH1 mutation detection, indicating that this recently developed three-b-value DWI approach may serve as a surrogate method for LGGs molecular diagnosis.
DWI, DCE, and IVIM MRI-derived parameters showed significant differences between grade II and III gliomas. Diffusion-related parameters, including ADC and D values, were significantly lower in WHO grade III gliomas; this result is in line with those of previous studies.19, 35 It is now well established that ADC is strongly correlated with cellularity and the nuclear cytoplasmic ratio in tumor tissue36, 37, 38, both of which are important criteria in the histopathological grading of gliomas.
Notably, perfusion-related parameters, especially SPF, f, and Ktrans, showed a relatively good performance for glioma grading compared with diffusion parameters. This is most likely due to the increased perfusion feature in higher grade gliomas; Ktrans reflects the volume transfer constant of a contrast agent from the plasma space to the extravascular extracellular space.39, 40 In higher-grade gliomas, active angiogenesis and incomplete basement membrane of tumor neovasculature lead to an increment in microvascular permeability, thus a high Ktrans value. A previous study13 showed that SPF and IVIM-derived f correlated well with DCE MRI-derived Ktrans and were useful in differentiating high- from low-grade gliomas. Our results further show that f and SPF also exhibited significant differences between WHO grade II and III gliomas.
Over the last decade, studies have shown that gliomas with IDH mutation are less aggressive and more sensitive to chemotherapy, contributing to a longer overall survival.41, 42, 43, 44 Therefore, IDH plays a key role in the determination of the glioma molecular phenotype. Zhao et al.45 have shown that compared with IDH1 mutant gliomas, IDH1 wild-type gliomas are characterized by increased neoangiogenesis and a higher nuclear cytoplasmic ratio due to the infiltrative nature. Higher vascular proliferation leads to stronger perfusion effects. In this study, DWI, DCE, and IVIM MRI-derived perfusion parameters all showed significant differences between IDH1 mutant and wild-type LGGs. Elevated perfusion was observed in IDH1 wild-type LGGs, which is in agreement with several previous reports using other perfusion imaging techniques.46, 47, 48 For example, Kickingereder et al.46 and Brendle et al.48 performed dynamic susceptibility contrast and arterial spin labeling perfusion-weighted imaging on patients with LGGs, respectively, and both found significantly higher cerebral blood flow values in IDH1 wild-type LGGs. This could be explained by considering the molecular function of IDH1. Cui et al.49 and Rei s et al.50 suggested that IDH1 mutation is associated with decreased invasiveness and reduced angiogenesis via downregulation of the Wnt/β-catenin signaling pathway. Furthermore, the accumulation of 2-hydroxyglutarate, an oncometabolite produced upon IDH1 mutation, has been shown to affect hypoxia-inducible factor (HIF) levels and the HIF response and may, consequently, reduce hypoxia-induced neovascularization.51
According to our ROC curve analysis, the simplified DWI-derived perfusion fraction showed a superior diagnostic accuracy as a predictor for IDH1 mutation in LGGs compared to the full IVIM-derived f. This result suggests that the three-b-value simplified DWI approach could save substantial scanning time compared with the full IVIM approach, with no loss of diagnostic efficiency. Additionally, both simplified and full IVIM perfusion performed better than DCE MRI. These two perfusion methods represent different aspects of vasculature. IVIM measures microscopic translational motions associated with microcirculation of blood in the capillary network, while DCE MRI measures capillary leakage of gadolinium contrast agent based on pharmacokinetic modeling. When WHO grade II and III gliomas were analyzed separately, we found SPF exhibited a statistically significant difference in assessing IDH1 mutation status of WHO grade II tumors, whereas f helped assess WHO grade III tumors. However, these preliminary results must be interpreted with caution due to the small sample size. Besides perfusion, diffusion parameters like ADC0,1000 were also predictive of IDH1 mutation in LGGs, with a lower diffusion coefficient found in IDH1 wild-type tumors. Our findings are in agreement with the existing literature regarding their association.47, 52
Our study has several limitations. First, the cohort was relatively small, especially that of WHO grade III LGGs (n = 8). Therefore, we may have underestimated some associations, such as the association between perfusion-related metrics and IDH1 mutation status, in WHO grade III gliomas. A further prospective study with a larger cohort should be performed to validate our results. Second, estimation bias may occur as a result of different cutoff b-values for IVIM analysis. Therefore, the set of b-values needs to be further optimized for brain tumors. Finally, the placement of ROIs was subjective and specific to a limited area on MRI. Automatic segmentation and image analysis of the entire tumor volume may improve preoperative risk stratification.
In conclusion, DWI, DCE, and IVIM MRI can be used as quantitative perfusion methods in preoperative IDH1 mutation prediction in LGGs. Specifically, the simplified DWI-derived perfusion fraction showed a superior diagnostic performance, which holds the potential to serve as a contrast-free and time-saving alternative in the clinical setting. However, further validation in a large patient population is warranted.
