Reproducibility and Reliability of Pancreatic Pharmacokinetic Parameters Derived from Dynamic Contrast-Enhanced Magnetic Resonance Imaging

Weiwei Zhao Xi'an Hospital of Traditional Chinese Medicine Jing Yu The Fourth Military Medical University Yuyu Bi Xi'an Hospital of Traditional Chinese Medicine Yi Huan The Fourth Military Medical University Yuanqiang Zhu The Fourth Military Medical University Weiqi Zhang The Fourth Military Medical University Jianmin Zheng The Fourth Military Medical University Zhiyong Quan (  qzy1503@163.com ) The Fourth Military Medical University Yong Yang Xi'an Hospital of Traditional Chinese Medicine


Introduction
The concept of dynamic MRI after contrast-agent injection was proposed in the mid-1980s, as a way of measuring the contribution of tissue perfusion and capillary permeability to the signal changes caused by the agent [1,2]. Dynamic MRI can be used to non-invasively assess normal or diseased tissue perfusion and micro-vessel permeability by means of qualitative, semi-quantitative and quantitative methods [3][4][5]. With the inclusion of arterial input function(AIF)and pharmacokinetic models, quantitative dynamic contrast-enhanced MRI (DCE-MRI) has been demonstrated superior to either qualitative or semi-quantitative method with respect to accurate acquisition of pharmacokinetic parameters [6]. However, the differences of AIF and pharmacokinetic models will affect the reliability and reproducibility of DCE-MRI pharmacokinetic parameters [7][8][9]. Among the practicable pharmacokinetic models, Extended Tofts Linear (ETL) model as a representative of two-compartment model can be recommended for quantitative assessment of physiological and pathological features, and a series of reliable results have been obtained [9,10]. The pancreas is an important digestive organ, which can undergo a variety of neoplastic and non-neoplastic lesions [11,12]. Accurate evaluation of the pancreas using DCE-MRI is helpful in diagnosis and differential diagnosis. However, pancreas is susceptible to respiratory motion and gastrointestinal peristalsis, whether the pancreatic pharmacokinetic parameters deriving from DCE-MRI with ETL model are robust, which needs to be further explored.
Pancreas is divided into pancreatic head, body and tail, and there is a different source of the blood supply for each region [13]. Besides, pancreatic parenchyma ratio varies for different age groups due to pancreatic atrophy and fat replacement [14], and some pancreatic tumors have certain gender tendency. In DCE-MRI assessment of pancreatic lesion, adjacent non-lesion pancreatic parenchyma on the same patient or pancreas from other healthy control are generally selected as normal reference [15,16]. However, those previous studies always ignored the in uence of different pancreatic regions, age and gender on pancreatic pharmacokinetic parameters. Hence, whether the DCE-MRI pharmacokinetic parameters of the pancreas are consistent across different pancreatic regions, age and gender distributions, which is worth exploring and needing to be clari ed before the study of pancreatic disease and selection of control group.
Thus, in the present study, we evaluated the reliability and reproducibility of pancreatic pharmacokinetic parameters derived from DCE-MRI with ETL model, and provided the relationship between pancreatic DCE-MRI parameters and 3 factors (pancreatic region, gender, and age ) on the purpose of providing a reference for future study of pancreatic disease and selection of normal control.

Methods
This study was in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The study was approved by the ethics committee of the Xijing Hospital of Air Force Military Medical University. The informed consent was obtained from all participants before collecting information. Data were analyzed and interpreted by the authors. All the authors reviewed the manuscript and vouch for the accuracy and completeness of the data and for the adherence of the study to the protocol.
Subjects 66 volunteers were recruited to participate the study from May 2019 to February 2020. To be included in this study, subjects had to meet the following inclusion criteria: more than 18 years old, healthy and with normal pancreatic function, no any disease in uencing pancreas. Exclusion criteria included common exclusion criteria for MRI scans and the use of Gd-related contrast agent, subjects with atherosclerotic disease in uencing AIF, and poor DCE-MRI image quality. Poor image quality should mainly meet the criteria severe motion artifacts appeared in enhanced MRI images and thus cannot be used for further evaluation. Finally, among 66 volunteers, four were excluded due to undesirable image quality, eight were excluded due to atherosclerosis, 54 volunteers were in the nal cohort.

MRI protocol
Prior to MR scanning, subjects were requested to fast at least 4 hours. MR images of the pancreases were acquired on a whole body 3.0 T MR scanner (Discovery MR750, GE Medical Systems, Chicago, IL, USA) with an eight-channel phased-array Torso coil. Using variable ip angle T1 mapping, pre-contrast three-dimensional spoiled gradient recalled echo sequence series were performed with ip angles of 3°, 6°, 9° and 12°. The other imaging parameters of T1 mapping were set as follows: repetition time (TR) = 3.2 msec, echo time (TE) = 1.5 msec, slice thickness = 4 mm, matrix = 260 x 160, eld of view (FOV) = 360 x 360 mm 2 . Then, DCE-MRI scans were performed by a threedimensional fast spoiled gradient recalled echo sequence for liver acquisition with volume acceleration (LAVA) with the following parameters: TR = 3.2 msec, TE = 1.5 msec, ip angle = 12°, FOV = 360 x 360 mm 2 , matrix = 260 x 160, slice thickness = 4 mm, bandwidth = 83.33 Hz/pixel. It took 240 sec to complete the DCE-MRI scanning with 40 phases acquired and 6 sec for each phase. After three precontrast phases were obtained, 0.1 mmol/kg of Gd-DTPA (Omniscan, GE Healthcare Co., Ltd., Shanghai, China) was administrated with a venous cannula at a rate of 2 ml/sec followed by a 20-ml saline ush at the same rate.

Data Manipulation
The DCE-MRI images were post-processed by Markov random elds (MRF) 3D non-rigid registration algorithms to correct motion artifacts.
Then the images were transmitted to a workstation for quantitative analysis using DCE-MRI OK software package (Omni Kinetics, Version 2.00, GE Healthcare Co., Ltd.). The analysis process has the following steps. Firstly, the individual AIF was obtained from a region of interest (ROI) in abdominal aorta. Secondly, ROIs were manually drawn on pancreatic enhanced images on multiple slices without reaching the perimeter to avoid partial volume effect, meanwhile without inclusion of vessel and main pancreatic duct. Finally, ETL model [9,10] was used to calculate the quantitative parameters: K trans , k ep , v e and v p . The mean of each parameter in the ROIs was used for statistical analysis.
The rst observer (XXX) measured DCE-MRI pharmacokinetic parameters thrice (by a time interval of at least one week to eliminate memory effect) to evaluate intra-observer reproducibility. Then, each of the three observers (observer 1, XXX, observer 2, YYY, and observer 3, ZZZ) measured parameters once to examine inter-observer reproducibility.

Intra-and inter-observer differences in pharmacokinetic parameters
Intra-and inter-observer differences were evaluated using one-way analysis of variance (ANOVA). Intra-and inter-observer agreements of pharmacokinetic parameters were evaluated using the inter-class correlation coe cient (ICC). The agreement was de ned as good (ICC > 0.75), moderate (ICC = 0.5 -0.75), or poor (ICC < 0.5). Coe cients of variation (CoV) were computed as the proportion of the standard deviation of the mean (standard deviation/mean, expressed as percentage). For CoVs concerning the intra-observer variability, standard deviation was computed over three measurements by one observer. For CoVs describing the inter-observer variability, standard deviation was computed over each parameter obtained by all three observers.
Differences of pharmacokinetic parameters among different region, age and gender groups Shapiro Wilk test was used for the normality distribution test. If the data conformed to the normal distribution, One-way ANOVA test and Independent Two-sample t test were used to evaluate the differences of pancreatic pharmacokinetic parameters obtained by observer 1. The former test was performed to evaluate the differences of pancreatic pharmacokinetic parameters among different pancreatic regions and different age groups. And the latter was used to exam the differences of parameters between male and female groups.
All statistical analyses were performed with the SPSS software Version 19.0. P values < 0.05 were considered to indicate a statistically signi cant difference.

Results
Graphs of AIF, time-intensity curve (TIC) and images of quantitative parameters were achieved in all 54 subjects. A series of representative graphs of a volunteer were shown in Figure 1.

Intra-and inter-observer assessment for pharmacokinetic parameters
There were no statistically signi cant intra-or inter-observer differences for K trans , k ep , v e and v p ( Table 1, P all > 0.10).
Variability analysis: In both intra-and inter-observer analysis, the CoVs of K trans , k ep , v e , v p were 9.98%, 5.99%, 6.47%, 4.76% and 10.15%, 5.22%, 6.28%, 5.40%, respectively. They showed small variation (all CoVs < 10%), except for CoV of K trans in inter-observer analysis (but only 10.15%) (Figure 2). Differences of pharmacokinetic parameters among different region, age and gender groups There were no signi cant differences of K trans , k ep and v e among different pancreatic regions, the P values were all larger than 0.10.
However, v p of pancreatic head was signi cantly higher than that of pancreatic body and tail (P = 0.014, 0.043) ( Table 3).
There were no signi cant differences to K trans , k ep and v p among different age groups, the P values were all larger than 0.10. However, pancreatic v e of old group was higher than that of young and middle-aged groups (P = 0.042, 0.001) ( Table 4).
There were no signi cant differences to K trans , k ep , v e and v p between male and female groups, the P values were all larger than 0.10 ( Table   5).

Discussion
In our evaluation of pancreatic DCE-MRI pharmacokinetic parameters, we used ETL model, which is a representative of two-compartment model. The model is generally recommended for tissue and tumor characterization, and computationally faster and better repeatability than the nonlinear method [17,18]. Besides, Jesper et al has showed that linear model was more stable against time resolution reduction than nonlinear model [19]. In this study, we got pancreatic K trans , k ep , v e and v p values using ETL model. We found these pharmacokinetic parameters had an excellent reproducibility in intra-and inter-observer analysis. We also found pancreatic K trans and k ep were independent of pancreatic region, age and gender in healthy volunteers. However, v p varied with pancreatic region and v e varied with age. Our results Page 5/9 could provide a basis for further study on perfusion and permeability of diseased pancreas and selection of normal pancreas control. The choice of the normal control group is relative broadness in K trans and k ep assessment without considering the factors of pancreatic region, age and gender. However, for v e and v p , the choice of the normal control group should be prudent, because we found v p varied with pancreatic region and v e varied with age.
In our study, ICCs of K trans , k ep , v e and v p are all greater than 0.90, CoVs of these pharmacokinetic parameters are all less than 10% in intraand inter-observer analysis except for K trans (but only 10.15%) in inter-observer analysis. Our results are consistent with previous studies [20,21] in DCE-MRI assessment of tumors. However, compared to their results, the parameters that we evaluated were more comprehensively. Our ICCs of v p were higher and our CoVs of v p were lower in intra-and inter-observer analysis than Wang et al [21]. This may be explained by a great deal of efforts made by us to ensure the precision of DCE-MRI. Before scanning, we made a strict implementation of inclusion and exclusion criteria, gave our technologists MRI scan training, and gave patients respiratory training. During scans, we used a series of 3D LAVA sequences, which can markedly reduce the scanning time compared with 2D sequences, but still maintain a high signal-to-noise ratio. After scanning, we used a 3D non-rigid image registration method to correct motion artifacts as much as possible. On the other hand, we chose to draw identical ROI on the abdominal aorta to obtain AIF, which was easy to operate and thus was more stable. In addition, we chose the ETL model, which might be more suitable to give more reliable and stable results [19]. Our results are different and superior to those of other researchers [22,23], which is because they used different software to calculate DCE-MRI parameters and evaluate reproducibility. So, we should use single software to ensure very good reproducibility in a sequential study.
In our study, pancreatic v p value was found varying among different region groups, which might be caused by the difference of their blood supply sources and blood vessels of the pancreatic islets in different regions. The arterial supplies of pancreas are complex, especially in pancreatic head [13]. The ratio of the capillary surface area to the volume of the islet capillaries was different between pancreatic head and caudal portion [24]. These might be the reason why v p for pancreatic head is the highest among 3 groups. Bali MA et al [25] assessed pancreatic perfusion using DCE-MRI with and without secretin stimulation in healthy volunteers. In that study, they found distribution fraction were signi cantly different between the head and the body, tail without secretin stimulation. Distribution fraction is the volume fraction of the tissue that is accessible to the contrast agent, which corresponds to the plasma and the interstitial space. Our result showed that v p for pancreatic head is the highest among 3 groups, which validates and deepens the above study, and indicated that the changing of distribution fraction may mainly come from the plasma space difference.
In our study, pancreatic v e in old group was the highest among different age groups. Several reports demonstrated that there were signi cant correlations between pancreatic volume, parenchymal volume, fat volume, fat/parenchyma ratio, CT density and age [14,[26][27][28]. For instance, Caglar et al [14] found that pancreatic volume reached its maximum at the age of forty, and remained constant until age sixty, then decreased gradually. They also found that the CT density of the pancreas peaked at 50 years of age. Yang et al [26] found pancreatic fat fraction remained constant during the age of 20 to 40 years, but signi cantly increased during the ages of 41 to 50 and 51 to 70 years. Therefore, it is suspected that the v e value is relative to pancreatic atrophy and fat replacement since their levels increase as the age grows.
Our study has some limitations. Firstly, ethical restrictions prevent repeated injection of contrast agents to volunteers, so there is a lack of assessment of san-rescan reproducibility. Secondly, due to the in uence of the position of pancreas, respiration and the movement of surrounding organs, there are still slight artifacts and noises after registration. This is a technical problem that is di cult to avoid, but with the development of software and hardware, this condition would be improved.
In conclusion, DCE-MRI with ETL model can be applied to give a reliable, robust and reproducible quantitative assessment of pancreatic pharmacokinetic parameters noninvasively. K trans and k ep of pancreas are independent of pancreatic region, age and gender, but v p can vary with pancreatic region and v e can vary with age. Our study enriched the study of pharmacokinetics of normal pancreas and can also provide guidance for the selection of normal reference in the pharmacokinetics study of pancreatic diseases.     Note: a, b indicate there was statistically difference between the two corresponding groups. a shows P = 0.014, b shows P = 0.043. Head, pancreatic head group, Body, pancreatic body group, Tail, pancreatic tail group