Abstract
Objective
To investigate the effect of interoperator variability in arterial input function (AIF) definition on kinetic parameter estimates (KPEs) from dynamic contrastenhanced (DCE) MRI in patients with highgrade gliomas.
Methods
The study included 118 DCE series from 23 patients. AIFs were measured by three domain experts (DEs), and a population AIF (popAIF) was constructed from the measured AIFs. The DEAIFs, popAIF and AUCnormalized DEAIFs were used for pharmacokinetic analysis with the extended Tofts model. AIFdependence of KPEs was assessed by intraclass correlation coefficient (ICC) analysis, and the impact on relative longitudinal change in K^{trans} was assessed by Fleiss’ kappa (κ).
Results
There was a moderate to substantial agreement (ICC 0.51–0.76) between KPEs when using DEAIFs, while AUCnormalized AIFs yielded ICC 0.77–0.95 for K^{trans}, k_{ep} and v_{e} and ICC 0.70 for v_{p}. Inclusion of the popAIF did not reduce agreement. Agreement in relative longitudinal change in K^{trans} was moderate (κ = 0.591) using DEAIFs, while AUCnormalized AIFs gave substantial (κ = 0.809) agreement.
Discussion
AUCnormalized AIFs can reduce the variation in kinetic parameter results originating from operator input. The popAIF presented in this work may be applied in absence of a satisfactory measurement.
Similar content being viewed by others
Introduction
Dynamic contrastenhanced (DCE) magnetic resonance imaging (MRI) is increasingly being used to assess microvascular properties of tissue in oncology and has proven well suited to quantify brain tumor hemodynamics and damage to the blood–brain barrier. DCEMRI is a technique in which a series of T_{1}weighted images is rapidly acquired before, during and after the administration of a paramagnetic contrast agent (CA) [1]. Using pharmacokinetic (PK) modelling, these dynamic images allow for the quantification of kinetic biomarkers that are of use in the evaluation of tumors. The most commonly used PK model in highgrade glioma (HGG) diagnostics is the extended Tofts model [2], which describes the tissue by the rate constants K^{trans} and k_{ep} as well as the volume fractions v_{e} and v_{p}.
Quantitative descriptors of tumor vasculature that—in theory—can be obtained independently of equipment and technique, would be a very powerful tool in disease management and monitoring. However, DCEMRI suffers from a lack of reproducibility across different imaging sites. This is due to several potential sources of error, including inadequate temporal resolution, insufficient acquisition time frame [3], precontrast T_{1} measurement uncertainty [4, 5] and difficulty in determining the dynamic CA concentration in plasma (C_{p})—the arterial input function (AIF) [6, 7].
Different approaches exist for determining the AIF, including fully automatic algorithms [8], semiautomatic methods [9], manual selection and the use of standard models [10]. It is generally accepted that the AIF is dependent on cardiac output, blood pressure and vascular autoregulation in the regionofinterest []. The gold standard for DCEMRI imaging is thus an individual AIF from each timepoint in each patient measured in a feeding artery of the pathology/regionofinterest (ROI). The accuracy of this AIF varies by the temporal resolution, potential partial volume effects from low spatial resolution [11] and interobserver variability from manual selection.
The aim of this study was to investigate the interobserver variability, among domain experts (DEs), in AIF determination and corresponding variability in kinetic parameters obtained using the extended Tofts model in HGG patients. Further, we define a parametric form of a populationaveraged AIF from brain data according to the framework previously published [10].
Methods
Patients
Study approval was obtained from the regional medical ethics committee and patients were included only if written informed consent was signed. A total of 118 DCEMRI examinations from 23 patients (17 males, mean age 53.7 years, range 32–66 years) with histologically confirmed HGG (one grade III and 22 grade IV) were included in a prospective study of early detection of perfusion changes during radiochemotherapy [12]. Patients were imaged once before, thrice during, and up to five times for a maximum of 15 months after initiation of standard treatment regime [13]. In the present work, the initial six imaging timepoints in each patient, encompassing two postradiochemotherapy followups at two and 14 weeks, were considered for inclusion. Examinations were excluded if there were no contrastenhancing voxels (N = 2) or if DCEMRI was not successful (N = 18), resulting in a total of 118 included examinations. A surgical debulking procedure was performed in 22 patients prior to baseline imaging, whilst the remainder received biopsy.
MRI
All imaging was performed at 3.0 T (Philips Achieva, Philips Medical Systems, Best, The Netherlands), using an eightchannel head coil. DCE images were acquired from a 3D saturation recoverybased gradient echo sequence. The sequence was designed to minimize water exchange effects by employing a short saturation recovery delay (TD) [14]. Eleven slices covering the tumor volume were acquired with the following key sequence parameters: field of view (FoV): 240 × 240 mm^{2}; acquisition matrix: 120 × 120; partial Fourier: 5/8; voxel size: 1.9 × 1.9 × 4 mm^{3}; pixel bandwidth: 434 Hz; echo time (TE)/flip angle: 2.5 ms/26°; sensitivity encoding (SENSE) factor: 2.3 in the anterior–posterior phaseencoding direction. Two variants of the DCE sequence were used: one using repetition time (TR)/TD/temporal resolution: 8.2 ms/80 ms/3.4 s, 100 dynamic scans and scan time: 5:40 (variant 1) in seven patients for a total of 38 examinations, and one using TR/TD/temporal resolution: 5.1 ms/50 ms/2.1 s, 150 dynamic scans and scan time: 5:10 (variant 2) in 16 patients for a total of 80 examinations. 0.1 mmol/kg gadobutrol (Gadovist^{®}, Bayer Schering Pharma AG, Berlin, Germany) was administered after a precontrast baseline duration of five timepoints using a power injector at a rate of 3 ml/s, immediately followed by a 20 ml saline flush at the same rate. The full imaging protocol included the following structural image series: sagittal 3D fluidattenuated inversion recovery (FLAIR) images, axial 2D T_{2}weighted images and sagittal 3D T_{1}weighted gradient echo images acquired before and after CA administration. All image series were acquired before the DCE series except the CAenhanced T_{1}weighted images which were acquired directly afterwards. Tumor regionofinterest (ROI) containing contrastenhancing tissue was defined in all examinations by a radiologist (4 years of experience) using a previously described method [12, 15].
AIF extraction
AIFs were determined independently in each DCE series by three DEs (two radiologists with 1 (J. V.) and 5 (M. Kim.) years of experience and one clinician with 10 years of experience in DCEMRI (C. L.)) using a semiautomatic algorithm based on automated identification of most likely AIF voxels using Kmeans clustering from a userdefined search region, implemented in the software tool nordicICE (NordicNeuroLab, Bergen, Norway) [8]. The AIF search region was selected independently by each DE to include the large intracerebral arteries according to the tumor location and hence the available volume covered by the DCE acquisition. Since the Kmeans method is iterative with a random seed starting point, repeated AIF searches may not result in identical results. Hence, multiple AIF searches were performed for a selected search region and the AIF with the highest firstpass arterial signal peak and the highest signal tail during the washout phase was chosen. This process was repeated until the AIF was considered satisfactory by the DE. An overview of the process is given in Fig. 1. In addition, a venous output function from the large sinuses was obtained in each examination for AIF partial volume correction [3].
The AIFs were supplied as singlecolumn text files containing change in longitudinal relaxation rate \(\Delta {R}_{1}={R}_{1}\left(t\right){R}_{1}\left(0\right)\) from baseline for each timepoint t, which was converted to CA concentration by:
where r_{1} = 4.4 mM^{−1} s^{−1} is the relaxivity of gadobutrol at 3.0 T [16] and Hct is the hematocrit, set to 0.45. R_{1}(0) of arterial blood at 3.0 T was assumed to be 0.60 s^{−1} [16].
This procedure resulted in a total of 354 DEdefined AIFs from 23 patients, of which 114 AIFs from seven patients were from DCE sequence variant 1 and the remaining 240 AIFs in 16 patients were from DCE sequence variant 2.
Populationaveraged AIF
A populationaveraged AIF (popAIF) was created by fitting the mean of all measured AIFs with DCE sequence variant 2 (N = 240 AIFs) to the sum of two Gaussians and an exponential modulated by a sigmoid function, as described by Parker et al. [10]:
Here, A_{n} are scaling factors, T_{n} are centers and σ_{n} are widths of the nth Gaussian; α and β describe the amplitude and slope of the exponential function, respectively, and s and τ are the width and center of the sigmoid. Prior to calculating the mean AIF, each individual AIF was timeshifted so that the peak occurred at the same timepoint.
Pharmacokinetic modeling
The DCE time series were analyzed using the extended Tofts model according to:
where C_{t}(t) is the timedependent CA concentration in tissue, C_{p}(t) is the timedependent CA concentration in blood plasma (the AIF), K^{trans} is the flux of contrast agent from the intravascular space (IVS) to interstitium, k_{ep} is the flux of contrast agent back to the IVS, v_{e} is the ratio of K^{trans} and k_{ep}, denoting the fraction of extracellular, extravascular volume and v_{p} is the fractional plasma volume.
From Eq. 3, it is evident that errors in the amplitude of \({C}_{\mathrm{p}}(t)\) will directly scale to corresponding errors in K^{trans} and v_{p}. Therefore, to assess the contribution of differences in AIF peak values, the analysis was repeated using AIFs normalized to an area under the curve (AUC) equal to the mean of all measured AIFs (600 mM s). Because each patient received the same CA dose at each imaging session, the AIF AUC is expected to remain constant, and normalization should therefore reduce variability in parameter estimates due to errors in measured peak AIF amplitudes, but at the cost of loss of absolute quantitative parameter estimates. Consequently, kinetic analysis was performed with seven different AIFs in each examination: three DEsupplied, the same three AUCnormalized and the popAIF obtained by fitting the mean AIF across all examinations to Eq. 2. From each analysis, the median parameter values from each tumor ROI were extracted and used for comparison. Voxels in which parameters v_{e} or v_{p} took an unphysiological value outside the range 0–100% was excluded from median calculation for all AIFs in that examination. The PK modelling and AIF analysis were performed in MATLAB 2020a (MathWorks, Natick, MA, United States).
Statistical analysis
AIF peak and AUC, as well as estimates of parameters K^{trans}, k_{ep}, v_{e} and v_{p}, were used for statistical analysis. The median value of each parameter in each tumor ROI was used for analysis. AIF characteristics were assessed using Wilcoxon signedrank test on the AIF peak and AUC of each DEsupplied AIF against each other (e.g. DE 1 vs DE 2; four combinations). Intraclass correlation coefficient (ICC) estimates with 95% confidence intervals (CI) were calculated based on an absoluteagreement, twoway randomeffects model [17]. The ICC estimate evaluates the degree of interoperator agreement in the KPE estimates within each examination, such that identical results from all AIFs in every examination would give ICC = 1. ICCs were estimated both using the complete dataset, and including only the examinations with DCE sequence variant 1 (N = 38) to assess the performance of the popAIF. To investigate the effect of different AIFs on predicting clinical outcome, the change in median K^{trans} for each patient from the first to the sixth imaging timepoint was compared using the measured DEAIFs and the normalized variants. An increase or decrease of more than 10% was labeled as, respectively, progression or remission, and the scores were assessed with Fleiss’ kappa using linear weights. The kappa statistic was interpreted according to the Landis and Koch benchmarking scale [18]. ICC and Fleiss’ kappa statistics were performed using Stata SE 16.1 (StataCorp LLC, College Station, TX, United States) with the kappa, etc., command [19].
Results
Populationaveraged AIF
Figure 2 shows the fitted population AIF together with the mean C_{b}(t). The fitted value for each AIF parameter (Eq. 2) is given in Table 1. The fitted AIF is seen to closely follow the mean AIF, with a firstpass peak followed by a smaller secondpass peak and subsequent washout. The standard deviation of the mean AIF demonstrates a higher variation in measured AIFs during the early phase than during washout. A comparison of the popAIF and DEAIFs with resulting K^{trans} maps is shown in Fig. 3.
Statistical analysis
The AIF peaks were found to be significantly different between all domain experts (p = 0.03), while AUC was found to be different between DE 2 and 3 (p < 10^{–8}), but not between 1 and 2 (p > 0.80) or 1 and 3 (p > 0.14).
Table 2 shows ICC for median kinetic parameter values from the 118 examinations. There is a moderate to substantial agreement for all kinetic parameters when using the measured AIFs [18]. Inclusion of the popAIF yields a lower agreement estimate for k_{ep}, while the other parameters are negligibly affected. The normalized AIFs give an almost perfect (0.8–1.00) agreement for parameters K^{trans} and v_{e}, while v_{p} ‘s agreement is substantial (0.61–0.80). As expected, k_{ep} is the least affected by normalization of the AIF, as it is not sensitive to the AIF amplitude [20]. Table 3 shows the same ICC estimates calculated from only the examinations using DCE sequence variant 1 and demonstrates a similar relationship between ICC estimates with and without inclusion of the popAIF.
Figure 4 shows the relative K^{trans} change from baseline to the sixth exam for two sample patients that demonstrate different degrees of observer dependence. Panel B represents a worstcase scenario, where different DEAIFs produce opposing estimates of disease progression.
When change in median K^{trans} is segmented into remission (< 10%), status quo (− 10% ≤ and ≤ + 10%) and progression (> + 10%), Fleiss’ kappa (3 raters, 23 subjects) is calculated as 0.595 with a 95% CI of (0.363, 0.827) for the measured AIFs, suggesting a fair to moderate agreement, while the normalized counterparts yield a kappa of 0.809 with CI (0.661, 0.958), suggesting a substantial to almost perfect agreement.
Discussion
The aim of the present work was to investigate the effect of differences in userdefined AIFs on kinetic parameters estimated from DCE using the extended Tofts model on data from HGG patients. Three experienced DEs individually produced AIFs for 118 examinations of 23 patients with HGG, following a common guideline that included the use of a semiautomatic detection algorithm. Additionally, a populationaveraged AIF was fitted to a subset of the measured AIFs. A fixed T_{1}(0) both in tissue and blood was assumed in all patients, and all analyses were performed at the same facility without input from the DEs other than their supplied AIFs. Therefore, the results of the kinetic analyses in this study should not be influenced significantly by factors other than the difference in AIFs.
The results in this study show moderate to substantial agreement in kinetic parameters estimated with AIFs from different DEs (Table 2). Agreement in estimated parameters is not substantially affected when the popAIF is included. This suggests that the popAIF represents an appropriate alternative to manually measured AIFs on a perexamination basis. The AUCnormalized DEAIFs is seen to yield higher agreement in parameter estimates for K^{trans}, v_{e} and v_{p}, as expected from the known sensitivity of these parameters to the AIF peak amplitude. The parameter k_{ep} depends more on the shape of the AIF [20] and is therefore virtually unaffected by the AIF normalization. This is in agreement with previous work investigating the effects of measuring the AIF for brain DCE in different vessels [21], which found a strong correlation between lower AIF peak and overestimation of all parameters except k_{ep}.
Comparing the relative change in parameter estimates over time as an indicator for disease progression shows that the use of AUCnormalized AIFs increases the agreement between DEs. The case presented in Fig. 4b represents a worstcase scenario where longitudinal change in K^{trans} is labeled differently (progression, stable, remission) by each of the DEAIFs, while all AUCnormalized DEAIFs and the popAIF indicate progression. This demonstrates that a single DCE series can contain several clusters of voxels that produce subjectively acceptable, yet consequentially different AIFs. This operatordependency is reduced with the AUCnormalized DEAIFs and completely removed with the popAIF, at the cost of potential loss of relevant information.
The AIFs in this study were not accompanied by perexamination measurement of hematocrit, which is known to fluctuate significantly during chemotherapy treatment [22]. Hematocrit variations would lead to a true variation in AIF peak amplitude, which would not be reflected in the AUCnormalized AIF or the popAIF. However, if hematocrit values were available, this could be incorporated as an additional perpatient and perexamination scaling factor in the resulting AUCnormalized or populationbased AIFs [23].
This study included data acquired with two slightly different DCEMRI sequences, differing in TR, TE, TD and number of dynamic scans. The popAIF was derived using only data from the sequence variant with the largest number (variant 2, N = 220) of scans. To confirm that this popAIF was valid for both sequence variants, the ICC was compared including either all data from both sequences and only data from sequence variant 1. The results (Tables 2, 3) show that agreement is equally affected when including the popAIF in both datasets, indicating that the popAIF can be used on data acquired with sequence parameters deviating from the ones used when defining the popAIF.
One major challenge with the DCEMRI field is lack of standardization between sites and studies. Although white papers suggesting standards for MRI protocol and analysis exist [2], there is still a substantial variation in both acquisition and analysis methods in the published data, making comparison of quantitative parameter estimates challenging. In treatment response studies, like the one presented here, the absolute values of the kinetic parameters may be of less importance and the emphasis should be on the ability of a given method to accurately detect parameter changes during the course of disease progression and treatment response. This is in line with previous findings in a large multicenter study concerning prostate cancer [7]. Our results suggest that this can readily be achieved through a combination of standardized and fixed protocols for AIF determination and a fixed imaging protocol over time. Further, and more importantly, our results suggest that the derivation of a populationwide AIF derived from the mean of many DCEMRI examinations acquired on the same system provides a good alternative to the need for timeconsuming identifications of individual AIFs by domain experts. It is, however, important to stress that such population AIF may need to be adjusted according to sitespecific variations in MRI protocol and scanner type. Still, the population AIF obtained here, based on the same functional form as previously suggested by Parker for abdominal use [10], may form the basis for a similar popAIF for use in brain DCEMRI applications.
In conclusion, normalizing the measured AIFs to a reasonable AUC can serve to reduce the variation in kinetic parameter results that stem from operator input. Further, using a parametric population AIF for the brain as presented in this study may be applied in absence of a satisfactory measurement in kinetic parameter estimation from DCEMRI data in HGG patients.
Abbreviations
 AIF:

Arterial input function
 AUC:

Area under the curve
 CA:

Contrast agent
 DCE:

Dynamic contrastenhanced
 DE:

Domain expert
 FLAIR:

Fluidattenuated inversion recovery
 HGG:

Highgrade glioma
 IVS:

Intravascular space
 KPE:

Kinetic parameter estimate
 MRI:

Magnetic resonance imaging
 PK:

Pharmacokinetic
 PopAIF:

Populationaveraged AIF
 ROI:

Regionofinterest
 TD:

Saturation recovery delay
 TE:

Echo time
 TR:

Repetition time
References
Larsson HBW, Stubgaard M, Frederiksen JL, Jensen M, Henriksen O, Paulson OB (1990) Quantitation of bloodbrain barrier defect by magnetic resonance imaging and gadoliniumDTPA in patients with multiple sclerosis and brain tumors. Magn Reson Med 16:117–131
Tofts PS, Brix G, Buckley DL, Evelhoch JL, Henderson E, Knopp MV, Larsson HBW, Lee TY, Mayr NA, Parker GJM, Port RE, Taylor J, Weisskoff RM (1999) Estimating kinetic parameters from dynamic contrastenhanced T1 weighted MRI of a diffusable tracer: standardized quantities and symbols. J Magn Reson Imaging 10:223–232
Larsson C, Kleppestø M, Rasmussen I, Salo R, Vardal J, Brandal P, Bjørnerud A (2013) Sampling requirements in DCEMRI based analysis of high grade gliomas: simulations and clinical results. J Magn Reson Imaging 37:818–829
Larsson C, Kleppestø M, Grothe I, Vardal J, Bjørnerud A (2015) T 1 in highgrade glioma and the influence of different measurement strategies on parameter estimations in DCEMRI. J Magn Reson Imaging 42:97–104
Heye T, Boll DT, Bashir MR, Merkle EM (2012) Impact of precontrast T1 relaxation times on DCEMRI pharmacokinetic parameters: T1 mapping versus a fixed reference value. Proc Intl Soc Mag Reson Med 20:1961
Keil VC, Mädler B, Gielen GH, Pintea B, Hiththetiya K, Gaspranova AR, Gieseke J, Simon M, Schild HH, Hadizadeh DR (2017) Intravoxel incoherent motion MRI in the brain: impact of the fitting model on perfusion fraction and lesion differentiability. J Magn Reson Imaging 46:1187–1199
Huang W, Chen Y, Fedorov A, Li XX, Jajamovich GH, Malyarenko DI, Aryal MP, Laviolette PS, Oborski MJ, O’Sullivan F, Abramson RG, Jafarikhouzani K, Afzal A, Tudorica A, Moloney B, Gupta SN, Besa C, KalpathyCramer J, Mountz JM, Laymon CM, Muzi M, Kinahan PE, Schmainda K, Cao Y, Chenevert TL, Taouli B, Yankeelov TE, Fennessy F, Li XX, Sullivan FO, Abramson RG, Jafarikhouzani K, Afzal A, Tudorica A, Moloney B, Gupta SN, Besa C, KalpathyCramer J, Mountz JM, Laymon CM, Muzi M, Kinahan PE, Schmainda K, Cao Y, Chenevert TL, Taouli B, Yankeelov TE, Fennessy F, Li XX (2016) The impact of arterial input function determination variations on prostate dynamic contrastenhanced magnetic resonance imaging pharmacokinetic modeling: a multicenter data analysis challenge. Tomography 5:56–66
Mouridsen K, Christensen S, Gyldensted L, Østergaard L (2006) Automatic selection of arterial input function using cluster analysis. Magn Reson Med 55:524–531
Bjørnerud A, Emblem KE (2010) A fully automated method for quantitative cerebral hemodynamic analysis using DSC–MRI. J Cereb Blood Flow Metab 30:1066–1078
Parker GJM, Roberts C, Macdonald A, Buonaccorsi GA, Cheung S, Buckley DL, Jackson A, Watson Y, Davies K, Jayson GC (2006) Experimentallyderived functional form for a populationaveraged hightemporalresolution arterial input function for dynamic contrastenhanced MRI. Magn Reson Med 56:993–1000
Chen J, Smith M, Frayne R (2005) The impact of partialvolume effects in dynamic susceptibility contrast magnetic resonance perfusion. J Magn Reson Imaging 22:390–399
Larsson C, Groote I, Vardal J, Kleppestø M, Odland A, Brandal P, DueTønnessen P, Holme SS, Hope TR, Meling TR, Fosse E, Emblem KE, Bjørnerud A (2020) Prediction of survival and progression in glioblastoma patients using temporal perfusion changes during radiochemotherapy. Magn Reson Imaging 68:106–112
Stupp R, Hegi ME, Gilbert MR, Chakravarti A (2007) Chemoradiotherapy in malignant glioma: standard of care and future directions. J Clin Oncol 25:4127–4136
Larsson HBW, Rosenbaum S, FritzHansen T (2001) Quantification of the effect of water exchange in dynamic contrast MRI perfusion measurements in the brain and heart. Magn Reson Med 46:272–281
Odland A, Server A, Saxhaug C, Breivik B, Groote R, Vardal J, Larsson C, Bjørnerud A (2015) Volumetric glioma quantification: comparison of manual and semiautomatic tumor segmentation for the quantification of tumor growth. Acta radiol 56:1396–1403
Shen Y, Goerner FL, Snyder C, Morelli JN, Hao D, Hu D, Li X, Runge VM (2015) T1 relaxivities of gadoliniumbased magnetic resonance. Invest Radiol 50:330–338
Koo TK, Li MY (2016) A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med 15:155–163
Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33:159
Klein D (2018) Implementing a general framework for assessing interrater agreement in stata. Stata J 18:871–901
Li X, Cai Y, Moloney B, Chen Y, Huang W, Woods M, Coakley FV, Rooney WD, Garzotto MG, Springer CS (2016) Relative sensitivities of DCEMRI pharmacokinetic parameters to arterial input function (AIF) scaling. J Magn Reson 269:104–112
Keil VC, Mädler B, Gieseke J, Fimmers R, Hattingen E, Schild HH, Hadizadeh DR (2017) Effects of arterial input function selection on kinetic parameters in brain dynamic contrastenhanced MRI. Magn Reson Imaging 40:83–90
Groopman JE, Itri LM (1999) Chemotherapyinduced anemia in adults: incidence and treatment. J Natl Cancer Inst. https://doi.org/10.1093/jnci/91.19.1616
Sahoo P, Gupta PK, Patir R, Vaishya S, Saha I, Gupta RK (2016) Comparison of actual with default hematocrit value in dynamic contrast enhanced MR perfusion quantification in grading of human glioma. Magn Reson Imaging 34:1071–1077
Funding
Open access funding provided by University of Oslo (incl Oslo University Hospital).
Author information
Authors and Affiliations
Contributions
AB: study conception and design, analysis and interpretation of data, critical revision. IRG: study conception and design, drafting of manuscript, critical revision. MK: acquisition of data, critical revision. MK: study conception and design, analysis and interpretation of data, drafting of manuscript, critical revision. CL: study conception and design, analysis and interpretation of data, drafting of manuscript, critical revision. JV: acquisition of data, critical revision.
Corresponding author
Ethics declarations
Conflict of interest
Author Atle Bjørnerud is a paid consultant of NordicNeuroLab AS. No other conflict of interest was reported from any of the authors. No funding was received for the writing of this study.
Ethical standards
Informed consent from all participants was acquired before imaging according to the ethical regional committee and the Helsinki declaration from 1964.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Kleppestø, M., Bjørnerud, A., Groote, I.R. et al. Operator dependency of arterial input function in dynamic contrastenhanced MRI. Magn Reson Mater Phy 35, 105–112 (2022). https://doi.org/10.1007/s1033402100926z
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s1033402100926z