Comparison of Intravoxel Incoherent Motion Diffusion-Weighted MR Imaging and Arterial Spin Labeling MR Imaging in Gliomas

Gliomas grading is important for treatment plan; we aimed to investigate the application of intravoxel incoherent motion (IVIM) diffusion-weighted imaging (DWI) in gliomas grading, by comparing with the three-dimensional pseudocontinuous arterial spin labeling (3D pCASL). 24 patients (13 high grade gliomas and 11 low grade gliomas) underwent IVIM DWI and 3D pCASL imaging before operation; maps of fast diffusion coefficient (D ∗), slow diffusion coefficient (D), fractional perfusion-related volume (f), and apparent diffusion coefficient (ADC) as well as cerebral blood flow (CBF) were calculated and then coregistered to generate the corresponding parameter values. We found CBF and D ∗ were higher in the high grade gliomas, whereas ADC, D, and f were lower (all P < 0.05). In differentiating the high from low grade gliomas, the maximum areas under the curves (AUC) of D ∗, CBF, and ADC were 0.857, 0.85, and 0.902, respectively. CBF was negatively correlated with f in tumor (r = −0.619, P = 0.001). ADC was positively correlated with D in both tumor and white matter (r = 0.887, P = 0.000 and r = 0.824, P = 0.000, resp.). There was no correlation between CBF and D ∗ in both tumor and white matter (P > 0.05). IVIM DWI showed more efficiency than 3D pCASL but less validity than conventional DWI in differentiating the high from low grade gliomas.


Introduction
Gliomas are the most common primary tumor in brain. More accurate preoperative grading of gliomas could increase the rationality of treatment plans and judge prognosis. Generally, blood supply of high grade gliomas is significantly higher than that of low grade gliomas. Perfusion imaging can assess tumor blood supply, which could provide a reference for glioma grading. The main nonenhanced method of perfusion in MRI is arterial spin labeling (ASL). In addition, there is another solution, intravoxel incoherent motion (IVIM); it also does not need contrast agent; thus it is safe, repeatable, and without radiation burden.
IVIM is the microscopic translational movement occurring in each image voxel during an MRI acquisition, which was proposed by Le Bihan et al. [1] in 1988. In the horizontal voxel size, capillary distribution can be considered as isotropic and random; therefore, the water molecule movement in the capillary also can be considered as an incoherent motion. By applying a diffusion gradient field of high values, the biexponential model can differentiate the diffusion in extracellular from that in intracellular, which are defined as fast diffusion coefficient * and slow diffusion coefficient , respectively. The perfusion fraction can be obtained simultaneously, which describes the fraction of incoherent signals which come from the vascular component [2]. The IVIM had been used for many tumors cases in recent years, such as hepatic carcinoma [3], renal cancer [4], prostate cancer [5], and salivary gland tumors [6], while less had been done for IVIM at gliomas [7][8][9]. With more and higher values, IVIM biexponential model could better demonstrate the decay curves of diffusion signal in brain gray and white matters [2,10,11]. However, in previous studies about gliomas, the maximum and number of value in DWI were relatively low and small (1,300 s/mm 2 and 14 in [7], 900 s/mm 2 and 16 in [8], and 3,500 s/mm 2 and 13 in [9]), which might affect the accuracy of the outcomes. Besides, the signal-to-noise ratio (SNR) of IVIM images in these studies was limited. The 3D pCASL is one submethod of ASL. It had been widely studied in gliomas and had reliable results with dynamic susceptibility contrast (DSC) [12][13][14]. The purpose of this study was to use the IVIM biexponential model based on DWI with higher (up to 3,500 s/mm 2 ) and more (20) values to analyze the characteristics of gliomas and to compare with the 3D pCASL findings.

Subjects.
The present study was approved by the Local Institutional Review and a written informed consent was obtained from all patients. From September 2013 to June 2014, 24 patients (nineteen men and five women, age range from 16 to 65 years, mean age 42 ± 14.24 years) confirmed as having gliomas by pathology were recruited and took MRI examinations before operation. The pathology results of all subjects were shown in Table 1.

2.3.
Postprocessing of DWI and ASL Imaging. The DWI imaging was calculated by at two-segment monoexponential algorithm as shown in IVIM equation (1), where and * are the diffusion parameters related to molecular diffusion and to the perfusion-related diffusion, respectively, / 0 is the normalized signal attenuation, and is the perfusion fraction. Setting 200 s/mm 2 as the cutoff of the low values, the high values will generate the first, and the low values will yield the * and at the same time after removing the effects of , finally producing the , * , and maps. ADC map was also calculated by values of 0 and 1000 s/mm 2 with monoexponential equation (2). ASL imaging was analyzed by corresponding software to produce CBF maps. All processes were calculated by GE AW4.6 workstation automatically:

Image Coregistration and Regions of Interest (ROIs).
Diffusion maps and structural 3D-T1WI images and were, respectively, coregistered to CBF images using SPM8 (http://www.fil.ion.ucl.ac.uk/spm/) by applying rigid-body transformations. And then, these maps were resliced to match the CBF maps voxel-by-voxel [15]. ROIs were drawn in the regions of tumors whose blood supplies were the richest as solid component as well as normal white matter in contralateral semioval center in the CBF maps, combined with the coregistered 3D-T1WI slices to avoid cystic, hemorrhage, necrosis, and ischemic area; then the ROIs were extracted to cover the coregistered DWI maps, using MRIcroN (http://www.mccauslandcenter.sc.edu/mricro/), Figure 1. Perfusion and diffusion data from the ROIs were analyzed by Matlab (MathWorks, Natick, MA) to get median, mean, standard deviation, kurtosis, skewness, maximum, minimum, and the 90th percentile of all voxels to make a histogram analysis such as [15].

Statistical Analysis.
Normal probability plot and Shapiro-Wilk's test were used to analyze data distribution. Pairwise comparisons of ASL and IVIM parameters between HGG and LGG were calculated. Bivariate correlations of parameters of ASL and IVIM DWI as well as IVIM DWI and conventional DWI were assessed. Student's t-test and Pearson test were used for normally distributed data, and a nonparametric test (Mann-Whitney ) and Spearman test applied for data that did not fulfill the requirements for normality. Differences in all parameters between HGG and LGG gliomas were further analyzed by using receiver-operating characteristic (ROC) curves and comparing the area under the curve (AUC). The parameter with the highest AUC was chosen as the best discriminating parameter. All analyses were performed by SPSS Version 16.0 for Windows (SPSS, Chicago); a < 0.05 was considered as statistical significance.

Mean ASL and IVIM Parameters in HGG as well as LGG.
Mean CBF and * were higher in the HGG ( < 0.05), whereas mean , ADC, and tended to be lower (the former one < 0.05, the following two < 0.01). There were almost no differences between contralateral semioval center of two groups in all parameters mentioned above (all > 0.05), Table 2.

Correlation Analysis.
In tumors, CBF was negatively correlated with the ( = −0.619, = 0.001). ADC was strongly positively correlated with in both tumor and white matter ( = 0.887, = 0.000 in former and = 0.824, = 0.000 in latter). There was no correlation between CBF and * in both tumor and white matter (both > 0.05). All the correlation analyses above were done by Spearman test.

Histogram and ROC Analysis of Perfusion and Diffusion
Parameters. Histogram parameters in tumor from the ASL and IVIM DWI scans were compared between HGG and LGG. For ASL, the maximum of CBF had the lowest value (0.002) and biggest AUC (0.85) for separating HGG and LGG. For IVIM DWI, the mean of * had lowest value (0.002) and biggest AUC (0.857) for discriminating HGG and LGG, while, in conventional DWI, the median of ADC showed the lowest value (0.000) and best AUC (0.92) in all parameters mentioned above for distinguishing HGG and LGG, Table 3. Histogram analysis showed almost no significant difference of all parameters of white matter between HGG and LGG, Table 4.

Discussion
There were many kinds of water molecule movements in brain tissue, including intracellular, intercellular, and transmembrane molecular diffusion, as well as microcirculation of blood in the capillary network. It has been confirmed that when applying a diffusion gradient field of high values, signal attenuation of water molecule diffusion in brain tissue is following the way of multiexponential models, which (multiexponential models) can better reflect the actual diffusion information of brain parenchyma than monoexponential model [16]. Compared to the multiexponential model, the biexponential model might be oversimplified, but in regard to the evaluation of two or more different proton diffusion pools in voxel, the biexponential model is more feasible [2,17].
Federau et al. [18] found that IVIM can monitor the change in cerebral blood flow caused by the expansion or contraction induced by CO 2 and O 2 . Wirestam et al. [19] indicated that and CBF derived from IVIM agree reasonably with conventional CBV and CBF get from DSC, respectively. The studies mentioned above revealed that it was feasible to evaluate the perfusion of brain tissue by IVIM. Once more, the applications of IVIM in varieties of pathological changes in brain tissue, such as stroke, gliomas, metastasis,  Figure 3: Male, aged 33 years with astrocytoma glioma in the right parietal (WHO II). The mass was hyperintensive with a fuzzy edge in axial T2-weighted image (a) and without any gadolinium enhancement (b). It shows slight perfusion in the CBF map (c) but fails to distinguish the boundaries of the tumor from normal brain substance. The TBF and CBF of contralateral semioval center were 80.72 and 27.53 mL/100 g/min, respectively. The * map (d) and map (e) well demonstrate the borderlines of the tumor as an enhancement image. The map also shows an inhomogeneous intrinsic hyperintensity in the tumor. The map (f) fails to delineate internal details of the tumor but better describes the scope of peritumoral edema. The ADC map (g) resembled map. The * , , and value of tumor and contralateral healthy semioval center were * , 2.73, 2.30 × 10 −3 mm 2 /s; , 0.74, 0.41 × 10 −3 mm 2 /s; ADC, 0.868, 0.524 × 10 −3 mm 2 /s; , 34.65%, 0.28, respectively. The logarithmic plot of signal intensity decay as function of with corresponding biexponential fit (h) of the tumor (arrow and solid circles) and contralateral semioval center (arrow head and hollow circles). and encephalitis, also show the potential advantages of IVIM in discovering the abnormal brain tissue [7][8][9].
Compared to previously recent published works in gliomas [7][8][9], in this research more and greater values were used, as more values in segment of low values can get more acute perfusion-related diffusion, while higher values can better eliminate the perfusion-related diffusion; thus it can in turn generate a more realistic molecular diffusion coefficient value and perfusion-related diffusion [2,10,11]. A histogram analysis of data also allows for a deeper analysis of the perfusion or diffusion distribution in the tumor that exceeds that of mean value.
For example, mean values of and ADC had closed AUC in identification of LGG from HGG; however, the median value of ADC showed the lower value and better AUC than the former two.
Like CBF, * of HGG was significantly higher than that of LGG in this study; mean of * was the most efficient in differentiating the HGG from LGG in three parameters of IVIM DWI. As * values could indirectly reflect tumor tissue perfusion, this result met the perfusion characteristics of gliomas, although * does not have a direct correlation with CBF in this study. Compared to the results obtained by Bisdas et al. [7], the * value in this study was relatively small, probably due to the fact that the higher values were used. The larger value, the more signal attenuation, which may affect the results as well as different fitting algorithms that can also be seen in [9]. In the * map, it was found that the * value for the tumors varied widely, perhaps because of the greater heterogeneity of tumor tissue, as reported by Bennett et al. [20].
In this study, the value of high grade gliomas was lower than that of low grade gliomas, which was not consistent with [7]. The correlation analysis also found that the value of the tumor was negatively correlated with TBF. As the value is the volume fraction of the rapid diffusion component in voxel, it could also reflect the perfusion of tumor tissue. So the value should be higher in high grade gliomas and positively correlated with CBF theoretically. In addition to the variation among selected tumor cases and structural characteristics of tumors, the different choices of values and parameters of DWI scheme might also be the reasons that led to the opposite result, according to Guiu et al. [21] and Lemke [22]. Time of echo (TE) is an independent factor of ; the longer the TE, the more signal attenuation at low values, and the greater value. Although Bisdas et al. [7] reported that the value was higher in high grade gliomas than that in low grade gliomas, their results also showed that the average value of white matter was higher than that of low grade gliomas and the similar results also can be seen in [9], so the impact coming from TE on value of different brain tissues still needs further study. We also found the values of lateral ventricles and edema around the tumors are large, which means the value is easy polluted by cerebrospinal fluid and edema. For these reasons, before setting a proper TE and removing the effects of cerebrospinal fluid and edema, an accurate value of tumor may not be acquired. Nonetheless, the map shows a high contrast in the scope and details of the tumor and may have extensive application.
Compared to LGG, the value tended to be lower in HGG in this study. As mainly reflects the Brownian motion of water molecules in the organization, which is more limited by smaller intercellular gaps, it is in line with the characteristic of gliomas, that is, the higher the grade, the greater the cell density. This is also consistent with the result of IVIM research in gliomas [9] and prostate cancer [5]. Compared to [7][8][9], the efficacy of in identifying HGG from LGG was lower than that of ADC derived from only two values (0 s/mm 2 and 1,000 s/mm 2 ) in this study, but there is a strong correlation between and ADC in both tumor and white matter, and map was comparable with ADC map in the exhibition of the scope and detail of tumor, whether could replace the ADC is still worthy of further study.
Through the measurement process of the cerebral cortex, it was found that the cerebrospinal fluid in sulcus had a larger impact on the * and value of cerebral cortex, which is consistent with the literatures [7,8]. This is because there are two kinds of water molecules movements in cerebrospinal fluid, that is, free diffusion and liquidity. Small blood vessels in sulcus also affect the results of the cortex in IVIM or ASL images. AS IVIM and ASL were based on mathematical models, IVIM results were influenced by the distribution of values and parameters of DWI [11,22]. The positions of subjects and PLD will affect the results of ASL [23,24]. And the average age of patients with HGG was greater than that of patients with LGG in this study. All these aforementioned factors will affect the results.
Currently, perfusion methods of MRI were mainly DSC and ASL. DSC was based on precise understanding of the arterial input function and a lot of assumptions. Thus, it was difficult to measure and quantify [25]. When applied to the brain, its low SNR, partial volume effects, susceptibility artifacts, and other factors outside the brain such as heart rate, cardiac output, and arterial stenosis would affect the results [26,27]. When DSC was applied to a brain tumor, there were also deviations caused by leaking contrast agent and vascular contamination. Patients with radiotherapy-induced vasculitis could not accept the bolus-injection pressure and the rate of high-pressure syringe. Contrast agent also may cause nephrogenic systemic fibrosis in patients with renal insufficiency. The mechanism of ASL uses the protons of freely diffused water molecules in the artery as an endogenous tracer, making subtraction between the before and after marking images, thus obtaining subtraction images only with perfusion information. In the absence of impacts of exogenous contrast agents, the physical and chemical characteristics of the blood are unchanged, while influence of bloodbrain barrier damage is also avoided. For some perspective, ASL could better reflect the hemodynamic parameters of the issue itself and has less susceptibility to any artifacts [12][13][14]. It may be more valuable in those diseases, which could lead to damage of blood-brain barrier, such as malignant gliomas. However, the SNR of ASL is low, and as described above, results of ASL are sensitive to the subjects' postures and instrument parameters, and so forth. Like ASL, IVIM does not need a contrast agent. It is based on the characteristics of the tissue itself. It stimulates and reads the information in voxels successively, and it is mainly sensitive to the increase of incoherent motion in voxel. Therefore, it is not directly affected by the anterior cerebral vascular (e.g., stenosis of internal carotid artery or vertebral artery), or the cardiac output and subjects' postures. As Henkelman [28] pointed out, classical perfusion measures the pattern of delivery, while IVIM measures the traffic flow in the direction of encoding gradient. The link between IVIM and classical perfusion had been discussed in detail [29]. For these reasons, IVIM could be a supplemental perfusion method to conventional perfusion sequences, but BioMed Research International 9 the influence of DWI parameters on the accuracy of IVIM parameters and cerebrospinal fluid contamination still needs further study.

Conclusion
The IVIM DWI shows efficacy in differentiating the low grade from high grade gliomas, * was better than CBF calculated by 3D pCASL, perfusion-related parameter was negatively correlated with the CBF, and map could show more clear scope of tumor. Though the efficacy of in identifying HGG from LGG was lower than ADC derived from conventional DWI, there was a strong correlation between them. Limitation of this study was that the sample size was small, and the subjects' ages in high grade gliomas group were considerably higher than those in low grade gliomas group. In summary, IVIM MR imaging could get the perfusion and diffusion information of gliomas simultaneously and might enable a noninvasive method in differentiating grade LGG from HGG gliomas.