Comparison of the values of MRI diffusion kurtosis imaging and diffusion tensor imaging in cerebral astrocytoma grading and their association with aquaporin-4
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0028-3886.177621
Source of Support: None, Conflict of Interest: None
Objective: To compare the value of MRI diffusion kurtosis imaging (DKI) and diffusion tensor imaging (DTI) in grading cerebral astrocytomas and to analyze the correlation of respective parameters with aquaporin-4 (AQP4) expression.
Keywords: Aquaporin-4; astrocytomas; diffusion kurtosis imaging; diffusion tensor imaging; tumor grading
Astrocytomas are the most common primary cerebral tumors. Accurate pretreatment grading of astrocytomas is important for predicting the response to treatment and the overall prognosis. Diffusion tensor imaging (DTI) has been shown to be of additional value in astrocytoma grading; however, the poor sensitivity and specificity of DTI in monitoring cellular changes related to malignant progression is a limitation.
Diffusion kurtosis imaging (DKI) is an extension of DTI., In addition to providing the conventional DTI parameters (fractional anisotropy [FA], mean diffusivity [MD]), DKI also allows for the measurement of mean kurtosis (MK), axial kurtosis (Ka), and radial kurtosis (Kr). This method facilitates the assessment of non-Gaussian diffusion patterns, which are affected by the tissue microenvironment including its component organelles, cell membranes, and water compartments. Preliminary studies have indicated the superiority of DKI over DTI in detecting microstructural changes.,,
A few studies that have evaluated the use of DKI for grading of cerebral astrocytomas ,, had a small sample size; besides, their results were not entirely consistent. The underlying mechanisms of the change in the DKI parameters are not clear. Aquaporin-4 (AQP4) has a key role in maintaining water and ion homeostasis, and it is thought to be involved in tumor malignancies, including those affecting cytoskeleton organization and cell migration. We believe that DKI could be of particular value in the assessment of astrocytoma grade and that AQP4 expression could be potentially linked to the change in DKI parameters. In this study, we compared the relative value and diagnostic efficacy of DKI and DTI parameters in the grading of astrocytomas. The correlation between DKI and AQP4 was analyzed in a larger study population than that included in the earlier studies.
The Institutional Review Board at our Medical University approved the study protocol. Written informed consent was obtained from all patients before their enrolment in the study. Sixty patients with cerebral astrocytomas diagnosed by histopathological examination of surgically resected tumor specimens were identified between January 2012 and May 2014. Tumors were graded using the World Health Organization Classification of Tumors of the Nervous System (2007) criteria. Patients who had not undergone any therapy before their magnetic resonance imaging (MRI) examination were eligible for inclusion in the study. Cystic astrocytomas and recurrent astrocytomas were excluded from the purview of this study. Patients were divided into two groups: Low-grade astrocytomas (LGAs, n = 25) [including Grades I and II] and high-grade astrocytomas (HGAs, n = 35) [including Grades III and IV].
MRI data acquisition
All examinations were performed with a 3.0 T MRI scanner (GE Signa HDxt, Waukesha of USA) using an 8-channel array coil. The scanning sequences included conventional MRI sequences (T1-weighted images [T1WIs], T2-weighted images [T2WIs], T2 fluid-attenuated inversion recovery [T2FLAIR] images, contrast-enhanced T1WI [CE-T1WI]) and DKI sequences. The parameters used for conventional MRI sequences were repetition time (TR) 195 ms and echo time (TE) 4.76 ms for gradient-echo, T1WI and CE-T1WI; TR 4000 ms and TE 98 ms for fast spin-echo T2WI; and, TR 8000 ms, TE 95 ms, and interval time (TI) 2371.8 ms for T2FLAIR. Thickness and slice interval were 5.0/1.5 mm; field of view (FOV) was 240 mm × 240 mm. Gadolinium chelate 0.1 mmol/kg body weight was used as the contrast medium.
Echo planar imaging sequence was used to obtain diffusion-weighted imaging (DWI) and DKI data. Implemented b values were 0, 1000, and 2000 mm 2/s. These were applied in 30 uniformly distributed directions. The following imaging parameters were kept constant throughout the DKI data (including DTI data) acquisition sequences: TR/TE: 6500/11 ms; FOV, 240 mm × 240 mm; matrix, 96 × 96; number of signals acquired: One; section thickness, 6 mm; slice interval, 1 mm. DKI scan time was approximately 7 min. Parameters for DWI data: TR/TE: 3000/87 ms; FOV: 240 mm × 240 mm; matrix: 128 × 128; thickness, 6 mm; slice interval, 1 mm; number of excitations (NEX) = 2; b = 0 and 1000 mm 2/s. DWI scan time was about 45 s.
MRI data processing and analysis
All DWIs were transferred to Advanced Workstation 4.4. DKI software (GE, USA). GE Functool 9.4.05a was used to perform the DKI analysis. The DKI data were corrected for head motion and eddy current distortions using global affine transformations. The diffusion tensor and diffusion kurtosis were calculated on a voxel-by-voxel basis and all data (b = 0, 1000, and 2000 mm 2/s) were used. They were fitted to the following Equation (1). After the kurtosis and diffusion tensor were estimated by fitting all DWIs into Equation (1), five metrics could be derived from the two tensors which included MK, Kr, Ka, FA, and MD.
Where S (n, b) is diffusion direction n and b; value b can be approximated by kurtosis and diffusion tensor; S0 is the signal intensity for b0 image; D is the MD; ni (i = 1, 2, 3) is the component of the diffusion direction vector n; Wijkl and Dij are the components of the kurtosis and diffusion tensor, respectively.
Conventional MRI sequences were used to visualize the basic features of tumors such as the cystic, necrotic and hemorrhagic components, tumoral solid parts, tumor boundaries, and range of edema, which indicated the edges of the region of interests (ROIs). With supporting workstation processing software (GE Functool 9.4.05a), ROIs were manually drawn around the solid parts of the tumors, peritumoral edema, and contralateral normal-appearing white matter (NAWM) of the two groups. Directionally encoded FA maps were inspected and ROIs that were reliably identifiable in all subjects and in which partial volume effects could be minimized were assessed. MK, Kr, Ka, MD, and FA values were measured, and the parameter values were corrected by contralateral NAWM. For example, the corrected MK value in the solid parts of the tumors = MK value in the solid parts of the tumors/MK value in the contralateral NAWM.
ROIs of the unenhanced tumors were delineated on the transverse T2WI [Figure 1], by excluding the peritumoral edema and adjacent normal tissue according to the transverse T2FLAIR image and DWI. The intensity of tumor was lower than peritumoral edema on T2FLAIR image and higher on DWI. ROIs of the enhanced parts of the tumors, which represented solid parts of the tumors, were delineated on the transverse CE-T1WI by excluding the cystic, necrotic, and hemorrhagic components and the adjacent normal tissue, which prevents the partial volume effect [Figure 2]. If the solid part of the tumor was heterogeneous and irregular, we used the DWI and apparent diffusion coefficient (ADC) to select the ROI of solid parts of the tumors, the lowest ADC value area being the ROI. ROIs of peritumoral edema were delineated on the transverse CE-T1WI according to transverse T2FLAIR image, and the ROIs of edema were delineated in the middle part of peritumoral edema by excluding adjacent normal tissue and cerebrospinal fluid (CSF), which prevented the partial volume effect and CSF contamination [Figure 1] and [Figure 2]. Two independent radiologists (YT, a Neuroradiologist with 6 years of experience; and, XCW, a Neuroradiologist with 13 years of experience) blinded to the pathological results performed image analysis and measured each value 3 times on each parameter image based on the same measurement standards, followed by calculation of the total mean ± standard deviation (SD). This method ensured the accuracy of the parameter value.
Aquaporin-4 analysis in the solid part of the tumor
Sixty patients underwent tumor resection, and the tumors were graded using hematoxylin and eosin and conventional immunohistochemical (IHC) staining at the Department of Pathology at our hospital. We chose the part of the tumor with the highest grade with the help of a pathologist. This part of the tissue, was analyzed for AQP4. Each specimen was subjected to AQP4 IHC staining (primary monoclonal rabbit anti-human antibodies [1:800]; Abcam). Two pathologists blinded to the identity of each specimen counted 100 tumor cells, numbered the positive cells, i.e., those showing brown particles in the cytoplasm or membrane, and calculated the average percentage of AQP4-positive cells of each patient in three different visual fields (×400). The total mean ± SD was calculated.
SPSS version 18.0 statistical software (IBM, USA) was used for data analyses. Inter-group differences between HGA and LGA with respect to the DKI parameters (MK, Kr, and Ka) and DTI parameters (MD and FA) values of the contralateral NAWM, solid parts of the tumors, and peritumoral edematous tissue were compared using the t-test. AQP4 expression was also compared between HGA and LGA using the t-test. P < 0.05 was considered statistically significant. Spearman correlation analysis for correlation of AQP4 expression with each parameter (MK, Kr, Ka, MD, and FA) was performed. Receiver operating characteristic (ROC) curves were generated for DKI parameter values (MK, Kr, and Ka) and DTI parameter values (MD and FA).
The study population comprised 39 men and 21 women (mean age: 49.6 years; range, 18–75 years). The mean ages of the patients with HGA and LGA were 55.6 ± 11.5 and 40.2 ± 9.8 years, respectively. The age difference between the two groups was statistically significant (P = 0.021). The results showed that only MK and Kr values in the contralateral NAWM were significantly lower in HGA as compared to that in the LGA (P = 0.028, P = 0.043, respectively), but the FA, MD, and Ka values in the contralateral NAWM did not differ significantly between the two groups.
Diffusion kurtosis imaging and diffusion tensor imaging parameter values of the two groups
DKI parameter values (MK, Kr, and Ka) and DTI parameter values (MD and FA) of astrocytomas corrected by contralateral NAWM are shown in [Table 1]. The MK, Ka, and Kr values in the solid parts of the tumor were significantly higher (P = 0.002, P = 0.001, and P = 0.018, respectively), and the MD value was significantly lower in HGA (P = 0.034) as compared to that in LGA. FA in the solid parts of the tumor did not differ significantly between the groups. Only the MK value in peritumoral edema was significantly higher in HGA as compared to that in LGA (P = 0.047).
Receiver operating characteristic analysis of diffusion kurtosis imaging and diffusion tensor imaging parameter values in the solid part of the tumor
ROC curve analysis of the DKI parameters (MK, Kr, and Ka) and the DTI parameters (MD and FA) in the solid parts of the tumor is shown in [Figure 3]. The optimal threshold, sensitivity, and specificity associated with each parameter are shown in [Table 2]. Ka (0.889) had the largest area under the curve (AUC), followed by MK (0.840), Kr (0.750), and MD (0.764). The AUC of Ka and MK were significantly higher than that of MD (P = 0.031, P = 0.040, respectively).
Aquaporin-4 expression analysis in the solid parts of the tumors
The AQP4 expression in the solid parts of the tumor was significantly higher in HGA (80.9 ± 11.3%) as compared to that in the LGA (25.2 ± 7.6%, respectively; P = 0.001) [Figure 4]. The results of Spearman correlation analysis for correlation between AQP4 expression and each parameter (MK, Kr, Ka, MD, and FA) pertaining to the solid parts of the tumors are shown in [Figure 5]. MK (r = 0.903; P = 0.001), Kr (r = 0.846; P = 0.003), and Ka (r = 0.817; P = 0.002) showed a positive correlation with AQP4 expression, whereas MD (r = −0.506; P = 0.030) showed an inconspicuous negative correlation with AQP4 expression. No correlation between FA and AQP4 was observed (r = −0.118; P = 0.404).
In the present study, patients with HGA were significantly older than those with LGA (P = 0.021). A distinct difference in the MK and Kr values in the contralateral NAWM was observed between the two groups (P = 0.028, P = 0.043, respectively). This could have introduced a bias because of the individual differences. These findings are similar to those reported by Van Cauter et al., and Raab et al., Individual differences were here attributed to the influence of age, tumor location, and underlying diseases, such as hypertension. Falangola et al., reported a decrease in the MK values in the frontal parts of the brain with increase in age in an elderly study population. Coutu et al., reported that the variability in the MK and Kr values was a function of the anatomical location and age of the subjects. For this reason, the effect of individual differences was factored in the analysis, by correcting the DKI and DTI parameter values using contralateral NAWM.
Our results showed that corrected MK, Ka, and Kr values in the solid parts of the tumors were significantly higher in the HGA group as compared to that in the LGA group. This was consistent with the results reported by Raab et al., and Van Cauter et al.,,, Increased values of DKI parameters associated with HGA probably reflect the higher degree of tissue complexity. Greater nuclear atypia, increased cell density, endothelial proliferation, heterogeneity (presence of hemorrhage or necrosis), and infiltrative growth were characteristic of HGAs whereas LGAs manifested larger-sized, well-differentiated cells, lower cell density, and usually a more infiltrative than nodular growth pattern, indicating fewer diffusion barriers. In this respect, HGAs are more structurally complex than LGAs, and the DKI values appear to reflect this difference.
The infiltrative nature of HGAs is a widely acknowledged characteristic, as is the limitation of the conventional MRI in delineating the ultrastructural morphology of astrocytomas on a microscopic scale. Only the MK value in peritumoral edema was found to be significantly higher in the HGA group as compared to that in the LGA group (P = 0.047). This particular finding is not consistent with the results reported by Raab et al., and Van Cauter et al.,, We postulate that microscopic tumor infiltration may increase MK in the peritumoral edema for some HGAs because they have more infiltrative tumor cells than LGAs. This aspect merits further investigation.
Jensen et al., reported that DKI parameters were more sensitive to changes in the tissue microstructure as compared to the conventional diffusion methods and that DKI allows for capturing of more comprehensive information than the latter. Kamagata et al., examined the brains of patients with Parkinson's disease and found that DKI parameters were more sensitive to changes in the white matter fibers when compared with the DTI parameters. Bar-Shir et al., reported a high sensitivity of MK in discerning myelin deficiency, indicating that pathological processes can be better tracked by means of non-Gaussian diffusion studies than by DTI analysis. Raab et al., showed that Grade II and Grade III gliomas differed significantly in MK values but not in FA or MD values. These findings further attest to the higher sensitivity of DKI in detecting microstructural changes.
In the present study, Ka (0.889) had the highest AUC, followed by MK (0.840), Kr (0.750), and MD (0.764). AUC of Ka and MK were significantly greater than that of MD (P = 0.031, P = 0.040, respectively), suggesting that DKI allowed for a more accurate grading of cerebral astrocytomas as compared to that allowed by DTI.
AQP4 is the most important member of the AQP family in the central nervous system (CNS) and is known to play a key role in the maintenance of the ion and water homeostasis. Various studies have reported significantly increased AQP4 expression in HGAs as compared to that in LGAs and normal brain tissue, suggesting its involvement in tumor malignancies, particularly those affecting cytoskeleton organization and cell migration.,, Consistent with these findings, we observed a significantly higher expression of AQP4 in the solid parts of the HGA (80.9 ± 11.3%) than LGA (25.2 ± 7.6%) (P = 0.001). Further, MK (r = 0.903; P = 0.001), Kr (r = 0.846; P = 0.003), and Ka (r = 0.817; P = 0.002) showed a positive correlation with AQP4 expression, whereas MD (r = −0.506; P = 0.030) had a slight negative correlation with AQP4 expression.
AQP4 is known to facilitate water transport into and out of the brain in CNS disorders. In brain tumors, there is an increased rate of water inflow and outflow from the brain, resulting in increased intracranial pressure from excess water accumulation in the intracranial compartments. Recent studies have also shown AQP4 to be redistributed across the entire surface of the cells in HGAs, but not in LGAs., The redistribution of AQP4 in glioblastoma cells is thought to be a reaction to the vasogenic edema that is induced by the breakdown of the blood–brain barrier, in order to facilitate reabsorption of excess fluid., We postulate that with increase in AQP4 in HGA, the astrocyte plasma membrane water permeability, and water diffusion from the blood into the CNS will also increase. The consequent increase in intra- and extra-cellular volume of water possibly results in an increase in the non-Gaussian diffusion pattern. Thus, DKI parameters appear to reflect the level of AQP4 expression in astrocytomas.
The present study has several limitations. The average age in the low-grade group was significantly lower than that in the high-grade group, which may have had influenced the parameter values. This was taken into account by performing a correction using contralateral NAWM. Second, the retrospective case selection, and the subjectivity in identifying the ROIs could also have had a bearing on the estimation of parameters. These are preliminary findings that will need to be validated in future studies. Further, peritumoral edematous tissue was not biopsied for histological examination at the time of surgery. Finally, the relatively small sample size was another limitation in our study. Further studies with a larger sample of patients with brain astrocytoma are planned.
DKI allowed for a more accurate grading of cerebral astrocytomas as compared to DTI, and DKI parameters may reflect the level of AQP4 expression in astrocytomas.
The authors thank Dandan Zheng, PhD at GE Healthcare, Beijing, China, for the technical support.
Financial support and sponsorship
This study was supported by grants from the National Natural Science Foundation of China (81471652) to Hui Zhang, the Science and Technology Innovation Fund of Shanxi Medical University to Yan Tan (01201304), the Youth Innovation Fund of First Hospital of Shanxi Medical University to Yan Tan (YC1426), the Scientific and Technological Project of Shanxi Province Health Department to Yan Tan (201201071), and the Basic Research Project of Shanxi Province to Xiao-Chun Wang (2015011092).
Conflicts of interest
There are no conflicts of interest.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5]
[Table 1], [Table 2]