Assessing the reproducibility of high temporal and spatial resolution dynamic contrast-enhanced magnetic resonance imaging in patients with gliomas

Temporal and spatial resolution of dynamic contrast-enhanced MR imaging (DCE-MRI) is critical to reproducibility, and the reproducibility of high-resolution (HR) DCE-MRI was evaluated. Thirty consecutive patients suspected to have brain tumors were prospectively enrolled with written informed consent. All patients underwent both HR-DCE (voxel size, 1.1 × 1.1 × 1.1 mm3; scan interval, 1.6 s) and conventional DCE (C-DCE; voxel size, 1.25 × 1.25 × 3.0 mm3; scan interval, 4.0 s) MRI. Regions of interests (ROIs) for enhancing lesions were segmented twice in each patient with glioblastoma (n = 7) to calculate DCE parameters (Ktrans, Vp, and Ve). Intraclass correlation coefficients (ICCs) of DCE parameters were obtained. In patients with gliomas (n = 25), arterial input functions (AIFs) and DCE parameters derived from T2 hyperintense lesions were obtained, and DCE parameters were compared according to WHO grades. ICCs of HR-DCE parameters were good to excellent (0.84–0.95), and ICCs of C-DCE parameters were moderate to excellent (0.66–0.96). Maximal signal intensity and wash-in slope of AIFs from HR-DCE MRI were significantly greater than those from C-DCE MRI (31.85 vs. 7.09 and 2.14 vs. 0.63; p < 0.001). Both 95th percentile Ktrans and Ve from HR-DCE and C-DCE MRI could differentiate grade 4 from grade 2 and 3 gliomas (p < 0.05). In conclusion, HR-DCE parameters generally showed better reproducibility than C-DCE parameters, and HR-DCE MRI provided better quality of AIFs.

To the best of our knowledge, no previous study has evaluated HR-DCE MR imaging in glioma patients. Thus, this study aimed to evaluate the reproducibility and validate the usage of HR-DCE MR imaging in patients with gliomas compared with C-DCE MR imaging. Table 1 shows demographics and pathologic results of study population. Examples of histograms and parametric maps of K trans and V e are displayed in Supplementary Fig. 1.
Correlations between C-DCE and HR-DCE parameters were not statistically significant (p > 0.05) in patients with glioblastoma (GBM). Both C-DCE and HR-DCE parameters were not statistically associated with pathologic results, such as Ki-67 values and O 6 -methylguanine-DNA-methyltransferase (MGMT) promoter methylation status (p > 0.05), in patients with GBM (Supplementary Table 1). AIF analyses. The 95th percentile of bolus arrival time (BAT) in C-DCE MR imaging was 20 s and that of wash-in time was 16 s. Thus, delayed bolus arrival error was defined when BAT was longer than 40 s (n = 1). In addition, delayed washout error was defined when time from maximal signal intensity (MSI) to half of MSI in wash-out slope (T wash-out ) was longer than 32 s (n = 7).
The MSI and wash-in slope (WIS) of AIFs from HR-DCE MR imaging were significantly greater than those of AIFs from C-DCE MR imaging (median, 31.85 vs. 7.09 and 2.14 vs. 0.63, respectively; p < 0.001) ( Table 3), even when suboptimal AIFs were included. After exclusion of suboptimal AIFs, HR-DCE MR imaging still demonstrated greater MSI and steeper WIS of AIFs than C-DCE MR imaging (44.60 vs. 5.87 and 3.01 vs. 0.57, respectively; p < 0.001) ( Table 3). All AIFs from 25 patients are plotted in Fig. 1.
Extended study. ICCs of C-DCE and HR-DCE MR imaging were described in Supplementary Table 2. The 95th percentile K trans and V e from C-DCE imaging were correlated with those from HR-DCE MR imaging in patients with gliomas ( Supplementary Fig. 3). Both 95th percentile K trans and 95th percentile V e derived from

Discussion
In this prospective study, we evaluated the reproducibility and clinical applicability of HR-DCE MR imaging in glioma patients, which demonstrated superior spatial (especially z-axis) and temporal resolutions compared to C-DCE MR imaging 22 . As many previous studies mentioned, acquisition of reliable AIF is a key factor in DCE MR imaging [16][17][18][19][20][21] , and improvement in the temporal resolution of DCE MR imaging 21 could be a possible strategy to overcome the low reproducibility of this technique. Several previous studies highlighted the importance of DCE MR imaging with higher temporal resolution in the prostate or breast [23][24][25] . Indeed, accelerated temporal resolution of DCE MR imaging in our study demonstrated the improved reproducibility of DCE parameters and superior ability to detect MSI of AIF in patients with gliomas. The effect of higher spatial resolution on the reproducibility of DCE parameters seems to be less significant.
Unexpectedly, HR-DCE MR imaging-specific errors (delayed wash-out error, delayed bolus arrival error) were noted. Even after thorough review and discussion with developers of HR-DCE MR imaging, we could not determine the exact cause of those errors. We believe there might be some setting errors during MR imaging that were not identified retrospectively. Nonetheless, HR-DCE MR imaging demonstrated the superior ability to detect greater MSI and steeper WIS of AIF. According to the study by You et al 21 , DSC MR imaging-based AIF showed improvement in the reproducibility of DCE MR imaging. Although this improvement was achieved by demonstrating higher signal intensity changes caused by the T2 * effect 21 , it is also plausible that accelerated scan interval (1.6 s vs. 4.0 s) enables us to catch the MSI of AIF more precisely.
Perfusion MRI has a role in treatment response evaluation in patients who underwent antiangiogenic agents (i.e., bevacizumab) [26][27][28] , but DCE parameters are not routinely used in real clinical practice because of their low reproducibility. Because the reproducibility of imaging parameters is a critical issue in clinical practice and trials 17,29,30 , our results of improved reproducibility of HR-DCE MR imaging could increase the possibility of implementing this technique in real clinical practice.
Interestingly, it seemed that histogram of DCE parameters could be different when temporal and spatial resolutions had been changed. Differences in z-axis resolution as well as AIF might affect the shape of histogram. Further evaluation investigating the effect of spatial and temporal resolutions on histograms is needed because histogram analysis has added prognostic value in patients with GBM 31 .
According to the results for spatial resolution, ICCs were excellent in both C-DCE and HR-DCE MR imaging, even though the z-axis resolution was significantly different. This might suggest that increased spatial resolution had a less significant effect on the reproducibility of DCE parameters. Indeed, HR-DCE MR images comprise 33,984 DICOM files in each patient, and heavy data size of HR-DCE MR imaging requires excessive analysis time. Although high spatial resolution of DCE MR imaging has an advantage in superior image quality 32 , clinical usefulness of DCE MR imaging might be enhanced with lower spatial resolution, given our results showing a less significant effect of spatial resolution on the reproducibility of DCE parameters.
In terms of clinical usefulness, the 95 th percentile K trans and V e were significantly higher in grade 4 gliomas than in grade 2 and 3 gliomas. However, the superiority of HR-DCE MR imaging for tumor grading was not proven because AUCs were not different in C-DCE and HR-DCE MR imaging. These findings might be related to the small size of the study population.
There are some limitations to our study. First, our study population was rather small because of several suboptimal cases, and this might be why our study failed to prove correlation between HR-DCE parameters and   www.nature.com/scientificreports/ clinical importance of HR-DCE MR imaging in a larger population is needed. Finally, it is difficult to set a gold standard in DCE MR imaging studies. Thus, we chose C-DCE parameters as reference standards, and HR-DCE and C-DCE parameters did not always correlate with each other. Thus, DCE parameters should be interpreted with clinical context, which often failed to be demonstrated in our study, possibly due to the small study population. This also suggests the importance of clinical validation of HR-DCE MR imaging in a larger study population.
In conclusion, HR-DCE parameters showed better reproducibility than C-DCE parameters, and AIF derived from HR-DCE MR imaging exhibited higher MSI and steeper WIS in prospectively enrolled patients with gliomas. Further studies investigating clinical importance using HR-DCE MR imaging should be conducted with a larger study population.

Materials and Methods
This prospective study was approved by the institutional review board of Seoul National University Hospital, and written informed consent was obtained from all patients, and all experiments were performed in accordance with relevant guidelines and regulations. Patient selection. From October 2018 to March 2019, 30 consecutive patients with presumed primary brain tumors who underwent diagnostic MR imaging were included in our study, and the details of diagnostic MR imaging protocol were previously described by You et al 21 . During diagnostic MR imaging, C-DCE MR images were routinely obtained. In these patients with informed consent, HR-DCE MR images were additionally obtained during navigation MR imaging for surgery. Because HR-DCE MR imaging was performed with GBCAs that used for navigation MR imaging, additional administration of GBCAs did not need. Median of time interval between diagnostic and navigation MR imaging was 2 days (interquartile range, 1-3 days).
The study design was abbreviated on Fig. 3 and Supplementary Fig. 4. The final pathological diagnosis was confirmed based on the WHO 2016 classification. For the primary study, we included patients whose final pathologic diagnosis was GBM (n = 14) and one patient was finally excluded whose HR-DCE MR imaging could not be analyzed. Thus, HR-DCE MR imaging data from 13 patients were included in the primary study. Among them, 6 patients were additionally excluded from the primary analysis, because of HR-DCE specific suboptimal AIFs ( Supplementary Fig. 5). Other pathologies in 16 patients were as follows: gliosarcomas = 2, anaplastic astrocytomas = 5, diffuse astrocytomas = 2, diffuse midline glioma = 1, oligodendrogliomas = 3, metastasis = 1, lymphocytic infiltration = 1, and lymphoma = 1. For an extended study, we also included patients with other types of gliomas.

MR imaging parameters.
Glioma study MR imaging studies were performed by using 3.0 T MR imaging units (Verio or Skyra, Siemens Healthineers, Erlangen, Germany; Ingenia, Philips Heathcare, Best, Netherlands) with a 32-channel head coil, and HR-DCE MR imaging was performed by using either Verio or Skyra (Siemens Healthineers, Erlangen, Germany).
Glioma study MR imaging protocols included pre-and postcontrast three dimensional magnetization-prepared rapid acquisition with gradient echo ( Image analysis for primary study. All image analyses were performed using commercially available software, Nordic ICE (NordicNeuroLab, Bergen, Norway), by a single radiologist (5 years of experience in neuroradiology) under the supervision of an expert neuroradiologist (18 years of experience in neuroradiology).
For C-DCE MR image analyses, parametric maps of K trans , V p , and V e were generated based on extended Tofts model 33 , using C-DCE MR image digital imaging and communications in medicine (DICOM) files. AIF search box was drawn at the level of the middle cerebral artery transverse segment 34 , and AIF was generated as a mean of five different AIFs from 5 automatically selected pixels. The second AIF was generated by using an AIF search box to draw subsequent axial images in the same manner to evaluate the reproducibility of AIF. After coregistration between DCE parametric maps and postcontrast T1WI images 21 www.nature.com/scientificreports/ on a pixel-by-pixel basis. HR-DCE MR image analyses were performed in the same manner using postcontrast T1WI from navigation MR imaging and HR-DCE MR image DICOM files. After a month, AIF selections in two consecutive image planes were repeated to evaluate intraobserver reproducibility. Previously designated ROIs for enhancing lesions were reused to exclude the effect of ROI selection and assess interobserver reproducibility primarily arising from z-axis and temporal resolutions. DCE parametric maps and AIFs obtained from these measurements were used to evaluate the intraobserver reproducibility and AIF analysis.
In addition, to evaluate the reproducibility originating from spatial resolution, DCE parametric maps using population-based AIF were generated to reduce the effect of AIF selection, and a second ROI for measurable enhancing foci was drawn for each patient. Using two individual ROIs and DCE parametric maps derived from population-based AIF, intraobserver reproducibility related to in-plane resolution was evaluated. www.nature.com/scientificreports/ AIF analysis. As in a previous study 21 , five parameters from AIF were evaluated: (a) BAT, (b) time to peak (TTP), (c) baseline signal intensity (BSI), (d) MSI, and (e) WIS. AIF analyses were performed both with and without suboptimal AIFs. Virtual AIFs from the median value were plotted with 25th (Q1) and 75th percentile (Q3) values 21 .
In cases in which AIFs were significantly different from others, AIFs were only used for AIF analysis, while DCE parameters derived from those AIFs were not included for DCE parameter comparison and correlation with clinical parameters. In our study, two types of suboptimal AIFs were detected in HR-DCE MR imaging. To identify these findings, T wash-out was calculated to define delayed wash-out error: T wash-out > two times more than 95 th percentile value of wash-in time (time from BAT to TTP) using C-DCE MR images ( Supplementary Fig. 5). Similarly, delayed bolus arrival error was defined when BAT was significantly delayed (two times more than the 95 th percentile value of BAT using C-DCE MR images) ( Supplementary Fig. 5).
Extended study. For extended study, T2WI images instead of postcontrast T1WI images were used as structural images. Using T2WI images, T2 hypersignal intense areas and contrast-enhancing lesions on postcontrast T1WI were included for ROI designation. DCE parametric map generation and coregistration were performed in the same manner as previously described. DCE parameters as differentiators for tumor grading were evaluated 21 in these patients. When there was a technical problem in coregistration of DCE and structural images, DCE parameter comparison or correlation with clinical parameters were not performed (Supplementary Fig. 4).
Statistical analysis. ICC and Bland-Altman plotting were performed to evaluate the intraobserver reproducibility of DCE MR imaging according to spatial and temporal resolution. For ICC interpretation, ICC values were considered (a) poor (ICC < 0.50), (b) moderate (ICC, 0.50-0.75), (c) good (ICC, 0.75-0.90), and (d) excellent (ICC > 0.90) 35 . Linear correlation coefficient was evaluated between mean pharmacokinetic parameters derived from C-DCE MR imaging and those derived from HR-DCE MR imaging. Relationship between DCE parameters and pathologic results such as Ki-67 or MGMT promoter methylation status were compared with linear and logistic regression analyses.
Normality of the parameters was assessed using Kolmogorov-Smirnov test, and subsequently, comparisons of AIFs or DCE parameters were performed by paired t test or Wilcoxon test, as appropriate. Unpaired parametric and nonparametric data were compared by independent samples t test or Mann-Whitney test, as appropriate. Comparison of area of receiver operating characteristic curve (AUROC) was performed between grade 4 tumors and others.
The statistical analysis was performed using statistical software (MedCalc version 15.2, Ostend, Belgium), and p-values less than 0.05 were considered statistically significant. www.nature.com/scientificreports/