Diagnostic efficacy of apparent diffusion coefficient measurements in differentiation of malignant intra-axial brain tumors

Background/aim To evaluate diagnostic efficacy of the apparent diffusion coefficient measurements from tumor (ADCt) and tumor circumference hyperintensities (ADCtch) in different types of malignant intra-axial brain tumors. Materials and methods Between April 2013 and June 2017 , 125 patients (52 females (41.6%) and 73 males (58.4%); mean age: 53 years, age range: 14-81 years), who underwent diffusion-weighted imaging (DWI) with intracranial mass, were retrospectively evaluated. The mean ADCt and ADCtch values and ratios were measured. Results Of the 125 patients, 22 (17.6%) had a low-grade glioma (LGG), 55 (44%) had a high-grade glioma (HGG), 32 (25.6%) had metastasis, and 16 (12.8%) had lymphoma diagnosis. There was a statistically significant difference in LGG and HGG in terms of mean ADCt and mean ADCtch values, and ratios. ADCtch values and ratios showed a statistically significant difference in the differentiation of HGG and metastasis and in the differentiation of HGG and lymphoma. According to ROC curve analysis, a cut-off value of 1.49 × 10−3 mm2/s for the mean ADCtch value generated the best combination of 70% sensitivity and 71% specificity for differentiation of HGGs and metastasis. The mean ADCtch value had the highest statistical predictive value for differentiation of HGGs and lymphoma with a sensitivity of 78% and a specificity of 76% for the optimal cut-off value of 0.82 × 10ˉ3 mm²/s. Conclusion The mean ADCt ratio allowed reliable differentiation of LGG and high grade brain tumors, including HGGs, metastases, and lymphoma. The mean ADCtch might be a better imaging biomarker in the differentiation of HHG from metastasis and lymphoma.


Introduction
Magnetic resonance imaging (MRI) plays a critical role in the preoperative evaluation of brain tumors. However, since tumor groups generally show similar signal intensity and contrast enhancement patterns on brain MRI, advanced MRI studies are needed in terms of differentiation of tumor groups. Diffusion-weighted imaging (DWI) is used for the differentiation and grading of tumors on the basis of cellularity with the microscopic movement of water protons. The studies have reported an inverse correlation between cellularity and apparent diffusion coefficient (ADC) measurement [1].
It is a known fact that gliomas have regions of different histological grades within the tumor. Histologically, the region with the highest grade within the tumor (cellular atypia, vascularization, mitotic characteristic, and necrosis) indicates the true grade of the tumor. Therefore, knowing this region is of great importance in both predicting prognosis and planning the appropriate treatment [2].
As a result of local disruption of the blood-brainbarrier, vasogenic edema is present around most brain tumors. In addition to vasogenic edema, infiltrative cells are present in the peritumoral region of primary brain tumors. While high-grade gliomas (HGG) usually grow infiltratively and invade surrounding tissues, metastasis and lymphoma exhibit an expansive growth and cause displacement in tissues rather than invading surrounding brain tissues [1,3]. The aim of this study is to investigate the contribution of ADC values measured from the tumoral and peritumoral regions to differential diagnosis in primary and metastatic intra-axial brain tumors.

Materials and methods
This retrospective study was approved by local ethics committee (2015-50), and the requirement for patient informed consent was waived.

Study population
Between April 2013 and June 2017, MRI examinations of 183 patients, with a diagnosis of intracranial mass, were evaluated retrospectively. Patients who had undergone open surgery and stereotaxic biopsy before MRI and who received radiotherapy and/or chemotherapy were excluded from the study. Patients with hemorrhagic mass (n = 15), multiple masses (n = 13), cystic mass (n = 11), masses less than 2 cm for optimal DWI recon (n = 11) on MRI examination, and without pathological diagnosis (n = 8) were excluded from the study. The inclusion and exclusion criteria for the study are presented in the flow diagram ( Figure 1).  examination time, 4:11 min; the field of view (FOV), 280 × 83 mm). After the intravenous administration of 0.2 mg/ kg gadolinium, contrast-enhanced T1A SE sequences were obtained in the axial, coronal, and sagittal planes.

Imaging protocol
DWI was obtained using a spin-echo sequence in the transverse plane with the single-shot echo-planar technique. The imaging parameters were as follows: TR/ TE, 3200/94; bandwidth, 1345 Hz; matrix, 192 × 192; NEX, 3; slice thickness, 5 mm; interslice gap, 1.5 mm; examination time, 1:12 min; and FOV, 230 x 230 mm. The diffusion gradients were encoded in three orthogonal directions. Three different b values (0, 500 and 1000 s/ mm²) were used for each slice. In all patients, ADC maps were obtained using the b values of 0 and 1000 s/mm².

Image analysis
ADC values were manually measured by 2 experienced radiologists (İ.E. and İ.M.Ç. with 9 and 4 years of experience, respectively) on the workstation (Leonardo, Siemens Healthcare) independently of each other, who were unaware of the clinical knowledge of the patients and the pathological diagnosis of the masses. We selected all continuous sections that included enhancing tumor and the peritumoral region. The lowest ADC areas were determined by visual inspection. One large and 2 small, uniform, and round (50-100 mm 2 ), ROIs were drawn manually to the lowest ADC areas. The ROI with the lowest ADC was chosen from these ROIs as the minimum ADC (ADC min ). The ROI with the highest ADC was chosen from these ROIs as the maximum ADC (ADC max ). The average of ROIs was recorded as mean ADC (ADC mean ). DWI was used only in qualitative analysis with conventional MR sequences. ROIs were carefully placed in the most homogeneous solid tumor area (ADC t ) corresponding to the enhancing area, nonenhancing tumor circumference hyperintensities (ADC tch ), and contralateral normal white matter (ADC n ) on the ADC map. In nonenhancing tumors, ADC t value was measured by taking the most homogeneous parts of the tumor into account on the T1 and T2-weighted image. In all regions, mean ADC values (ADC t , ADC tch , and ADC n ) were recorded. The cystic, necrotic, or hemorrhagic parts of the lesions were not included in the circular ROI images (Figures 2-5). When there was discordance in ADC values between radiologists, consensus was reached by a collaborative decision. ADC ratios were calculated by dividing ADC values of tumoral and peritumoral regions by ADC value of contralateral normal white matter (ADC t/n and ADC tch/n ).

Statistical analysis
Statistical analyses were performed using the SPSS software, version 21 (IBM Corporation, Armonk, NY, USA). Interobserver agreement values were determined using Pearson correlation test with a 95% confidence interval: 0-0.20 inadequate agreement; 0.21-0.4 slight agreement; 0.41-0.6 moderate agreement; 0.61-0.8 substantial agreement, 0.81-1.00 almost perfect agreement, and 1.00 perfect agreements. ANOVA calculation was used to determine the statistically significant difference of the four tumor groups (LGG, HGG, and metastasis and primary cerebral lymphoma) according to their different ADC values and ratios. Significance values were tested based on a 95% confidence interval for the difference between groups. If the P value was < 0.05, differences were considered significant. The multiple comparisons Tukey's T-procedure was used for further examination of pairwise differences of significant ADC values. If the P value was < 0.05, differences were considered significant.
In the differentiation of the tumor groups, receiver operating characteristic (ROC) analysis was used to determine the threshold value of ADC with the best sensitivity and specificity combination.

Results
The study group consisted of a total of 125 patients (52 females and 73 males) with a mean age of 53 (14-81 years) years, which had solitary intra-axial brain tumor.
There was a statistically significant difference in all parameters (ADC values and ratios) in 4 tumor groups with the ANOVA test (P < 0.05) ( Table 2). ADC t values and ratios were higher in the LGG than in the other tumor groups. ADC tch values and ratios were lower in the HGG tumors than in the other tumor groups. The lowest ADC t values were measured in lymphoma.
In the pairwise discrimination tests conducted with the Tukey's T-procedure, there was a statistically significant difference LGG and HGG in terms of mean ADC t and mean ADC tch values and ratios. Mean ADC t ratios showed a statistically significant difference in differentiation of LGG and metastasis. ADC t values and ratios showed a statistically significant difference in the differentiation of LGG and lymphoma and in the differentiation of metastasis and lymphoma. Finally, ADC tch values and ratios showed a statistically significant difference in the differentiation of HGG and metastasis and in the differentiation of HGG and lymphoma (P < 0.05) ( Table 3).
In the pairwise discrimination tests, a ROC analysis was performed for each single factor ADC value found statistically significant between both tumor groups. The most powerful predictive marker to differentiation of LGG and HGG was the mean ADC t ratio, with a sensitivity of 72% and a specificity of 72% for the optimal cut off value 2.06 × 10ˉ³ mm²/s. The most powerful predictive marker to differentiation of LGG and metastasis was the mean ADC t ratio, with a sensitivity of 77% and a specificity of 72% for the cut off value 1.44 × 10 −3 mm²/s. The most powerful predictive marker to differentiation of LGG and lymphoma was the mean ADC t ratio, with a sensitivity of 90% and a specificity of 94% for the cut off value 1.21 × 10 −3 mm²/s. The most powerful predictive marker to differentiation of HGG and metastasis was the mean ADC tch value, with a sensitivity of 70% and a specificity of 71% for the cut off value 1.49 × 10 −3 mm²/s. The most powerful predictive marker to differentiation of HGG and lymphoma was the mean ADC tch value, with a sensitivity of 78% and a specificity of 76% for the cut off value 0.82 × 10 −3 mm²/s. Finally, the most powerful predictive marker to differentiation of metastasis and lymphoma was the mean  T1WI (b). On the diffusion-weighted imaging (c), while the tumor is hyperintense, its circumference is hypointense. The mass is hypointense on the ADC map (d). The tumor circumference is hyperintense. The ADC value, measured from the tumoral region, was 1.11 × 10 −3 mm²/s, and the ADC value, measured from the tumor circumference hyperintensities, was 2.00 × 10 −3 mm²/s (Histopathological diagnosis: lung adenocarcinoma metastasis). ADC t value, with a sensitivity of 72% and a specificity of 82% for the cut off value 1.13 × 10 −3 mm²/s ( Figure 6).

Discussion
In our study, we evaluated diagnostic performance with ADC measurements to facilitate the preoperative differentiation of malignant intra-axial brain tumors, which are usually not distinguishable with conventional MR imaging sequences. ADC t values and ratios were higher in LGGs than in other tumor groups. Our results showed that the most important parameter is the mean ADC t ratio, distinguishing LGGs from high grade brain  (d), the tumor is significantly hypointense while its circumference is hyperintense. The ADC value measured from the tumoral region was 0.48 × 10 −3 mm²/s, and the ADC value, measured from the tumor circumference hyperintensities, was 1.37 × 10 −3 mm²/s (Histopathological diagnosis: cerebral lymphoma). Table 1. Pathological diagnoses of brain tumors.

Tumors n (%) Subgroups (n)
LGG   tumors, including HGGs, metastases, and lymphoma. The lowest ADC tch values and ratios were measured in HGGs. According to the ROC curve analysis, a cut-off value of 1.49 × 10 −3 mm 2 /s for the mean ADC tch value generated the best combination of sensitivity 70% and specificity 71% (P < 0.05) for differentiation of HGGs and metastasis. The mean ADC tch value had the highest statistical predictive value for differentiation of HGGs and lymphoma with a sensitivity of 78% and a specificity of 76% for the optimal cut-off value of 0.82 × 10 −3 mm²/s. The most powerful predictive value, for differentiation of metastasis and lymphoma, was the mean ADC t value, with a sensitivity of 72% and a specificity of 82% for the optimal cut-off value of 1.13 × 10 −3 mm²/s. DWI has been used to grade or differentiate among brain tumors on the basis of cellularity. The ADC measurements reflect the mobility of the free water fraction, including extracellular and intracellular water, within the tissue. Many studies have shown that calculation of ADC may help in differentiating cerebral tumors [1,4]. In addition, an inverse correlation has been found between tumor cellularity and ADC values [5,6].
Most brain tumors are surrounded by vasogenic edema that showing high signal intensity on T2WI. Vasogenic edema is the most common form of cerebral edema in brain tumors. There is a tendency for local disruption in the blood-brain barrier. Due to the increased capillary permeability and pressure, plasma fluid and protein accumulate from the vascular space to the extracellular space. Tumor circumference is pure vasogenic edema in metastatic brain tumors or lymphoma, and there are no tumor cells in tumor circumference. On the other hand, Figure 6. (a). The most powerful predictive marker to differentiate between LGG and HGG was the ADC t ratio; the area under the ROC curve (AUC) was 0.778 (670-899 with a 95% confidence interval) (b). The most powerful predictive marker to differentiation of LGG and metastasis was the mean ADC t ratio; the AUC was 0.760 (630-890 with a 95% confidence interval) (c). Th e most powerful predictive marker to differentiation of LGG and lymphoma was the mean ADC t ratio; the AUC was 0.968 (919-1017 with a 95% confidence interval) (d). The most powerful predictive marker to differentiation of HGG and metastasis was the mean ADC tch value; the AUC was 0.7767 (681-887 with a 95% confidence interval) (e). The most powerful predictive marker to differentiation of HGG and lymphoma was the mean ADC tch value; the AUC was 0.808 (704-912 with a 95% confidence interval) (f). The most powerful predictive marker to differentiation of metastasis and lymphoma was the mean ADC t value; the AUC was 0.778 (654-914 with a 95% confidence interval). the tumor circumference hyperintensities are infiltrative edema in high-grade glioma. There are infiltrative tumoral cells that infiltrate through the blood brain-barrier and may invade white matter pathways [2].
Several previous studies [7,8] reported that ADC values were not useful in preoperative grading of glioma. In our study, the mean ADC t ratio had the highest statistical predictive value in the differentiation of LGGs and HGGs, with a sensitivity of 72% and a specificity of 72% for the optimal cut off value of 2.06 × 10 −3 mm²/s. A recent study on the mean ADC of brain tumors showed that the mean ADC t values are significantly higher in LGG than in HGG. The cut off value of 1.40 × 10 −3 mm²/s can be considered an index for ADC to distinguish HGG from LGG. In addition, they emphasized that standard deviation of ADC is also substantial in differentiation of LGG and HGG [9]. Cihangiroglu MM et al.
[10] compared DWI using high b-value (b = 3000 s/mm 2 ) to standard b-value (b = 1000 s/mm 2 ) in the preoperative grading of supratentorial gliomas. They found that ADC parameters derived from DW-MRI using a high b-value allow a better differential diagnosis of gliomas, especially for differentiating grades III and IV, than those derived from DW-MRI using a standard b-value. A possible reason for this result is that the tumor boundaries were directly determined by the DWI maps with b = 3000 s/mm 2 [11].
In our study, the mean ADC t ratio had the highest statistical predictive value in the differentiation LGG and metastasis, with a sensitivity of 77% and a specificity of 72% for the cut off value of 1.44 × 10 −3 mm²/s. The mean ADC t ratio had the highest statistical predictive value in the differentiation LGG and lymphoma, with a sensitivity of 90% and a specificity of 94% for the cut off value of 1.21 × 10ˉ³ mm²/s. These results show us that tumoral cellularity is higher in metastasis and lymphoma than in LGG.
There was no statistically significant difference between LGGs and metastasis, between LGG and lymphoma, and between metastasis and lymphoma in terms of peritumoral ADC values and ratios (P > 0.05). These values show us that there is pure vasogenic edema in the tumor circumference hyperintensities of LGGs, metastasis, and lymphoma.
The previous studies reported that ADC t values are not useful in differentiation of HGG and metastasis [12][13][14]. However, Krabbe et al. and Lee EJ et al. reported that ADC t value was higher in cerebral metastasis than in HGGs [6,15]. They stated that the high value in metastases was due to the high content of intracellular or extracellular water in metastases compared to gliomas. In our study, there was no statistically significant difference between HGG and metastasis in terms of ADC t values and ratios.
HGG typically shows an infiltrative growth pattern with invasion of the surrounding brain tissues. However, brain metastasis reveals an expansive growth pattern and displaces the surrounding brain tissues [15]. Some researchers hypothesized that ADC tch values could be used to distinguish infiltrative edema of the glioma from metastatic vasogenic edema [1,15]. But, some conflicting results were reported in these studies. Pierre Lemercier et al. used the gradient of ADC values in the tumor circumference hyperintensities to differentiate HGG from metastases. ADC measurements were made in areas near, an intermediate distance from, and far from the coreenhancing tumor (G1, G2, and G3). They found a gradient of ADC in the peritumoral edema of glioblastoma associated with peritumoral glial alterations [3]. While, Eun Ja Lee et al. reported that the mean minimum ADC values and mean ADC ratios in the tumor circumference of glioblastomas were significantly higher than those in metastases [15], Ionut Caravan et al. reported that the mean ADC min values in the tumoral circumference of HGGs were significantly lower than those in brain metastases [16]. These conflicting results revealed that assessment of the tumor circumference hyperintensities with DWI could be challenging. Because, some high-grade tumor, such as primary (de-novo) glioblastoma, show minimal microscopic tumor infiltration in the tumor circumference hyperintensities. A meta-analysis study reported that DWI showed a moderate diagnostic performance for differentiation of HGG from brain metastasis [17]. In our study, ADC tch values and ratios were lower in HGG than metastases, and there was a statistically significant difference (P < 0.05).
Authors of several studies [9,[18][19][20] have reported that DWI allows differentiation of lymphoma and HGG on the basis of ADC measurements. ADC t and ADC tch values and ratios for patients with lymphoma were substantially lower than those for patients with HGG. Although lymphomas had lower ADC t values compared to HGGs in our study, there was no statistically significant difference (P > 0.05). However, a cut-off value of 0.82 × 10 −3 mm 2 /s for the mean ADC tch value generated the best combination of sensitivity 78% and specificity 76% (P < 0.05) for differentiation of HGGs and lymphoma. According to several authors, the differences in ADC between patients with lymphoma and those with metastasis were not statistically significant, although the ADC for patients with lymphoma was lower than that for patients with metastasis [2,14]. The lower ADC t values in patients with lymphoma may have been related to the high degree of cellularity of these tumors leading to more restricted diffusion compared with that of patients with other malignant brain tumors [19][20][21]. In our study, the most powerful predictive marker was the mean ADC t value in differentiation of metastasis and lymphoma.
In a variety of advanced noninvasive MRI techniques such as spectroscopy, perfusion imaging, and diffusion tensor imaging, we only evaluated the usefulness of ADC measurements in the differentiation of intra-axial brain tumors because DWI is the cost-effective and least timeconsuming technique available in most hospitals. Data processing is fast and relatively easy [15].
Our study has several limitations. Firstly, the accuracy of statistical data varies due to the numerical inhomogeneity of the current patient groups. Secondly, the ROI placement method is subjective and may differ among researchers. Thirdly, the geometric resolution of the ADC maps is lower than conventional MRI. Therefore, some difficulties are encountered in ROI placement, especially in small-sized tumors or ring-enhancing tumors.
In conclusion, DWI with ADC measurements contribute significantly to conventional MRI in grading glial tumors and differentiating HGG from metastasis and lymphoma. Detecting peritumoral invasion and grading glial tumors, in the preoperative period, prevent unnecessary stereotaxic biopsy, help the surgeon in preoperative planning and may contribute to decrease in mortality and morbidity.