

Feasibility of tissue similarity mapbased relative cerebral blood volume assessment in the evaluation of gliomas
Correspondence Address: Source of Support: The National Natural Science Foundation of China (81171393, 31271066); a project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD), Conflict of Interest: None DOI: 10.4103/00283886.162001
Objective: To investigate the feasibility of tissue similarity map (TSM)based relative cerebral blood volume (rCBV) assessment in evaluating the hemodynamic characteristics of gliomas and in differentiating highgrade gliomas from lowgrade ones without concentration time curve (CTC). Keywords: Glioma, relative cerebral blood volume, tissue similarity maps
A highgrade glioma is highly invasive and aggressive. ^{[1]} Magnetic resonance imaging (MRI) is one of the noninvasive imaging techniques that is widely used in the early diagnosis and therapeutic evaluation of gliomas. In particular, magnetic resonance perfusionweighted imaging (PWI) is an imaging technique with a high sensitivity to the microscopic level of blood flow and allows characterization of gliomas by providing measurements of the cerebral hemodynamic status. For instance, relative cerebral blood volume (rCBV) is an important PWI measurement that can be used to evaluate histological and angiographic vascularity. ^{[2]} It has long been proven that rCBV can help in the grading of gliomas, ^{[3],[4],[5],[6]} predicting early brain tumor recurrence ^{[7],[8]} and differentiating brain tumor recurrence from posttreatment changes. ^{[9]} Conventionally, rCBV can be determined by the ratio of the areas under the concentration time curves (CTCs) of tissues. ^{[10],[11],[12],[13],[14]} However, for the conventional method, the accurate quantification of the hemodynamic characteristics of the brain is limited by uncertainties in nonlinear ΔR _{2} * effects. ^{[10],[15],[16],[17],[18]} Recently, a novel postprocessing approach named tissue similarity maps (TSMs) has been developed to provide an alternative way to estimate the rCBV depending only on the signal intensity time course without the need for CTC. ^{[10]} The TSMbased rCBV (rCBV _{TSM} ) has been proposed to examine the vascular response in lesions in subjects suffering from multiple sclerosis (MS). ^{[10]} However, the ability of rCBV _{TSM} in the differentiation of gliomas has not been tested as yet. The main purpose of this paper is to investigate the feasibility of rCBV _{TSM} in evaluating the hemodynamic characteristics of gliomas and in differentiating highgrade gliomas from the lowgrade ones without knowledge of CTC. To achieve these goals, in vivo evaluation of patients with histopathologically determined gliomas (n = 35) has been conducted in this paper. The results derived from rCBV _{TSM} maps and the corresponding conventional rCBV _{PWI} maps have been compared and analyzed.
Study population This study was approved by our institutional review board (the First Affiliated Hospital of Soochow University) and a written informed consent for the study was obtained independently from all the patients. From September 2012 to December 2013, 35 patients (male: 17±61 years; female: 18±73 years) with 35 histopathologically determined gliomas in our institute had been consecutively enrolled in this study. These gliomas comprised 23 highgrade ones (seven astrocytomas [grade III] and 16 glioblastomas [grade IV]), 12 lowgrade ones (five astrocytomas [grade I] and seven oligoastrocytomas [grade II]). The types of gliomas and their grades were determined based on the classification of the World Health Organization (WHO). MRI examination All patients underwent MR examinations on a 3.0T MRI scanner (Signa HDx, General Electric, Milwaukee, WI, USA). The MRI protocols included (1) gradientrecalled echo (GRE)based pulse sequence for localization purpose, field of view (FOV): 256 mm × 256 mm, TR/TE: 8.6 ms/4 ms, flip angle (FA) =20° , slice thickness = 7 mm, matrix size = 256 × 230. Three slices were obtained in three orthogonal directions; (2) fast spoiled gradientrecalled sequence based T1weighted imaging, FOV: 240 mm × 240 mm, TR/TE: 8/3 ms, FA = 70°, slice thickness = 1 mm, matrix size = 320 × 256; (3) Turbo spin echo sequencebased T2weighted imaging, FOV: 256 mm × 192 mm, TR/TE: 3500/102 ms, FA = 90° , slice thickness = 3 mm, matrix size = 448 × 358; (4) fluidattenuated inversion recovery (FLAIR), FOV: 256 mm × 192 mm, TR/TE: 7000/79 ms, inversion time (TI) =2500 ms, FA = 120°, slice thickness = 4 mm with 0.4 mm gap, matrix size = 256 × 192; and (5) PWI using a twodimensional echo planar imaging (EPI) pulse sequence, FOV: 256 mm × 256 mm, TR/TE = 1840/32 ms, FA = 90° , slice thickness = 5 mm, matrix size = 128 × 128. For PWI, 10 precontrast time points and 40 postcontrast time points were acquired with a temporal resolution of 2 s. GdDTPA (0.1 mmol/kg; Magnevist, Bayer Schering Pharma AG, Berlin, Germany) was used as the contrast agent and delivered into the antecubital vein at the 11 ^{th} time point using a power injector (Medrad, Spectris MR Injection System, Pittsburgh, PA, USA) at a flow rate of 2 mL/s. Administration of contrast agent was followed by a 20mL saline flush at a flow rate of 2 mL/s. TSMbased rCBV TSM has been proposed to provide an alternative method for the estimation of the rCBV depending merely on the signal intensity time course. ^{[10]} The calculation of rCBV _{TSM} was presented as follows: First, a series of PWI are obtained at N successive time points after the delivery of contrast agent. The mean squared error (MSE) between the signal intensity of tissues of interest (i = 1, 2,…, N) and the reference tissue (i = 1, 2,…, N) over the N time points is calculated by where i denotes the index of the time point, denotes the location of the voxel of interest, t _{i} denotes the ith time point, ∆TTP denotes the difference of timetopeak (TTP) between and the reference tissue and and are the mean values of the precontrast baseline of the voxel of interest and the reference tissue, respectively. From equation (1), it is easy to see that the differences in signal baselines are removed. Second, assuming that all tissues have the same signal dependence such as local blood volume, equation (1) can be modified by introducing a linear coefficient , which gives Lastly, setting the firstorder derivative of with respect to equal to zero, we have If signals of all tissues have the same behavior, can be considered as rCBV _{TSM} . For all patients, rCBV _{TSM} maps were generated from the PWI images using equation (3) with normal white matter (WM) as reference tissue. The corresponding conventional rCBV _{PWI} maps were also determined by simply integrating the area under the tissue CTC. Image analysis All data sets were transferred from the picture archiving and communication system (PACS) workstation to a personal computer for further analysis. Postprocessing was coded with MATLAB 7.6 (Math Works, Natick, MA, USA). All regions of interest (ROIs) were determined in consensus by two radiologists (with 5 and 8 years of experience in neurology MRI, respectively) who were blinded to the histopathologic outcomes and the purpose of the study. For each of the patients, a pair of ROIs with diameters ranging from 3 to 5 mm, depending on the shape of the glioma, were selected in glioma and normal white matter (WM), respectively, in a representative slice (the representative slice is the one on which the tumor has the largest size) of rCBV _{TSM} map and corresponding rCBV _{PWI} map. For the pair of ROIs selected of both rCBV _{TSM} and rCBV _{PMI} maps, the ratio of rCBV values of ROIs selected in glioma to those in normal WM were calculated as , where rCBV _{glioma} and rCBV _{WM} denoted the rCBV values of glioma and normal WM, respectively. On the other hand, the signal to noise ratio (SNR) of ROIs in glioma and normal WM of both rCBV _{TSM} and rCBV _{PWI} maps were defined as SNR = S/SD where S and SD denoted the mean value and the standard deviation of the selected ROIs, respectively. The improvement factor of SNR, R, was calculated as , where SNR _{TSM} and SNR _{PWI} denoted the SNRs of rCBV _{TSM} and rCBV _{PWI} maps, respectively. The value of R was calculated over the 35 patients to indicate the improvement of SNR in rCBV _{TSM} compared with that in rCBV _{PWI} . Statistical analysis Statistical analyses were performed with software MedCalc (version 12.7, Mariakerke, Belgium). To examine the feasibility of rCBV _{TSM} in the evaluation of the hemodynamic characteristics of glioma, the agreement between rCBV _{TSM} and conventional rCBV _{PWI} ratios was tested. The concordance correlation coefficients (CCC) between rCBV _{TSM} and rCBV _{PWI} ratios was calculated across all patients (n = 35) as follows: ^{[19],[20]} where, denoted the covariance of the rCBV _{TSM} and rCBV _{PWI} ratios, and denoted the variance of rCBV _{TSM} and rCBV _{PWI} ratios, respectively, and denoted the mean value of rCBV _{TSM} and rCBV _{PWI} ratios, respectively. The CCC ranged from 1 to 1. The closer the CCC is to 1, the higher is the agreement between the rCBV _{TSM} ratio and rCBV _{PWI} ratio. Moreover, the rCBV _{TSM} ratio and rCBV _{PWI} ratio of all patients (n = 35) were fitted to a linear equation to evaluate the correlation between them. The MannWhitney test was used to compare both rCBV _{TSM} and rCBV _{PWI} ratios between high and lowgrade gliomas. A P value less than 0.001 was considered to be statistically significant. Meanwhile, receiver operating characteristic (ROC) curve analysis was performed to evaluate both rCBV _{TSM} and rCBV _{PWI} ratios in differentiating high and lowgrade gliomas. The cutoff values were determined by assuming equal misclassification rate. The area under the ROC curve (AUC) was also calculated to access the degree of the relation between rCBV _{TSM} and rCBV _{PWI} ratios of high and lowgrade gliomas. The closer the AUC was to 1, the stronger was the relation; and, the closer the AUC was to 0.5, the weaker was the relation.
For a glioblastoma case, the representative slices of the rCBV _{TSM} map and the corresponding rCBV _{PWI} map were presented for comparison [Figure 1]a and b.
A small difference (4.29% ±2.62%) was found between the mean values of rCBV _{TSM} and rCBV _{PWI} ratios over the 35 cases. [Figure 2] showed the correlation between the rCBV _{TSM} and rCBV _{PWI} ratios in terms of a linear relationship between rCBV _{TSM} ratio (y) and rCBV _{PWI} ratio (x) [y = 0.9978x+0.0266, r = 0.9974, P < 0.05]. The CCC of rCBV _{TSM} ratios and rCBV _{PWI} ratios was about 0.9974, with 95% confidence interval (CI): (0.9948, 0.9987).
[Table 1] summarizes the range, mean value and standard deviation (SD) of the rCBV _{PWI} and rCBV _{TSM} ratios. For lowgrade gliomas, the rCBV _{PWI} ratios ranged from 0.84 to 2.89, with a mean value of 1.56 ± 0.57, and the rCBV _{TSM} ratios ranged from 0.87 to 2.81, with a mean value of 1.56 ± 0.54. For highgrade gliomas, the rCBV _{PWI} ratios ranged from 2.10 to 15.87, with a mean value of 6.13 ± 3.29, and the rCBV _{TSM} ratios ranged from 2.29 to 16.20, with a mean value of 6.15 ± 3.28. The P values of the MannWhitney test of rCBV _{TSM} ratios and rCBV _{PWI} ratios were less than 0.001.
The ROC analysis corresponding to high and lowgrade gliomas is shown in [Figure 3]. The interactive dot diagrams of rCBV _{PWI} ratios and rCBV _{TSM} ratios are shown in [Figure 4]. The cutoff value for rCBV _{PWI} ratios discrimination was 1.95, with a sensitivity of 100% and a specificity of 91.7% [Figure 4]a. The cutoff value was 2.81 with a sensitivity of 95.7% and a specificity of 100% for rCBV _{TSM} ratios [Figure 4]b.
The improvement factors of SNR of glioma and normal WM ranged from 1.00 to 2.12 and 1.01 to 1.81, respectively. The mean value R was 1.31 ± 0.24 for glioma and 1.28 ± 0.240 for normal WM. [Figure 5] demonstrated the improvement factor of SNR (R) in both glioma and normal WM for each case.
Angiogenesis has been considered a key element in the pathophysiology of tumor growth. ^{[21]} After growing beyond a diameter of 12 mm, passive diffusion is no longer sufficient to support further growth of malignant cells. At this stage, neovascularization becomes necessary for a high rate of tumor growth; ^{[22]} therefore, microvascular density in gliomas can be taken as a prognostic indicator of histological tumor grade. ^{[23]} PWI is a noninvasive imaging technique of assessing the hemodynamic characteristics of gliomas and can be used as an indirect marker of angiogenesis. Measurements of rCBV on PWI have long been recognized to help in the diagnosis of brain tumors. For instance, Law et al. have reported that rCBV is helpful in identifying lowgrade gliomas that tend to become malignant. ^{[24]} In another work, they have proposed that rCBV is the best predictor for grading of gliomas. ^{[25]} Sugahara et al. have reported that rCBV is valuable in the differentiation of recurrent brain tumors and nonneoplastic tissues that have also been enhanced by using a contrast agent. ^{[9]} Fuss et al. have demonstrated the ability of rCBV in prediction of the risk of tumor recurrence after fractionated stereotactic radiotherapy. ^{[7]} Despite its usefulness, the rCBV estimated by the conventional method suffers from the uncertainties of the measurement of the nonlinear ΔR _{2} * effects. ^{[10],[15],[16],[17],[18]} The nonlinear ΔR _{2} * effect is a factor that has an impact on the estimation of perfusion parameters. For instance, Calamente et al. have proposed that the commonly assumed linear relationship between contrast agent concentration and ΔR _{2} * may cause large errors in characterizing the perfusion state because of the existence of nonlinear ΔR2 _{*} effects. ^{[15]} In all, these uncertainties in measurement of nonlinear ΔR _{2} * effects will jeopardize the accuracy of rCBV estimated by the conventional method. The TSMbased rCBV _{TSM} implemented in this paper depends only on the signal intensity curve rather than the CTC. These advantages make it more reliable to provide an accurate estimation of the hemodynamic characteristics of gliomas. The results of our preliminary experiments demonstrate that the rCBV _{TSM} is consistent with the conventional rCBV _{PWI} , which implies that the rCBV _{TSM} assessment is feasible in evaluating the hemodynamic characteristics of gliomas. Generally speaking, the distribution of the rCBV _{TSM} map is close to the corresponding conventional rCBV _{PWI} map [Figure 1]. The small difference in percentage between the rCBV _{PWI} ratio and the corresponding rCBV _{TSM} ratio indicates that rCBV _{TSM} is "accurate," with the conventional rCBV _{PWI} map as a criterion. The statistical analysis also confirms this point [Figure 2]. The linear fitting and the high correlation coefficient reflect a high correlation between the rCBV _{TSM} ratio and the corresponding rCBV _{PWI} ratio. The high CCC between the rCBV _{TSM} ratio and the rCBV _{PWI} ratios again suggests a high agreement between the two. Besides this, the rCBV _{TSM} is feasible in discriminating highgrade gliomas from lowgrade ones. As shown in [Table 1], the results of the MannWhitney test suggest that the rCBV _{TSM} ratio can discriminate between high and lowgrade gliomas at a significant level (P < 0.001) as well as the rCBV _{PWI} ratio can. As shown in [Figure 3], the mean rCBV ratios were able to clearly distinguish the high grade gliomas from the low grades ones. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve of the rCBV _{TSM} ratio (AUC = 0.996) was larger than the rCBV _{PWI} ratio (AUC = 0.993), which implies that the parameter derived from the TSMbased rCBV has an advantage in differentiating between high and lowgrade gliomas over that derived from the conventional rCBV. This advantage in differentiating gliomas might be attributed to the improved SNR of rCBV _{TSM} (which is up to twofolds) compared with that of rCBV _{PWI} . The main drawback of the study is that the number of patients (n = 35) recruited in this study is limited. Further clinical evaluation on a larger patient population is warranted. Based on the current investigation, further cooperation with clinicians to directly assess the reliability of TSMbased rCBV in the evaluation of histological and pathological characteristics of gliomas may have several clinical applications. Furthermore, although the current study focuses on gliomas, actually TSMbased rCBV may also be implemented in the diagnosis of other kinds of brain tumors such as a primary cerebral lymphoma ^{[26]} and metastases ^{[27]} or other cerebral diseases such as arteriovenous malformations and migraine. ^{[28]}
In conclusion, the preliminary results of this study suggest that the TSMbased rCBV is feasible in the evaluation of the hemodynamic characteristic of gliomas, and the differentiation of high and lowgrade gliomas depends only on the signal intensity curve rather than CTC. Moreover, compared with the conventional rCBV map, the TSMbased rCBV map possesses a closer relationship to the grades of gliomas and a higher SNR, which allows potentially more accurate diagnosis. Further studies with a larger patient population are warranted to explore the full potential of TSMbased rCBV in a clinical setting.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5]
[Table 1]


