Dynamic Contrast-Enhanced Perfusion MRI of High Grade Brain Gliomas Obtained with Arterial or Venous Waveform Input Function Silvano Filice, Girolamo Crisi From the Department of Medical Physics and the Department of Neuroradiology, University Hospital of Parma, Parma, Italy.
ABSTRACT BACKGROUND AND PURPOSE: The aim of this study was to evaluate the differences in dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) perfusion estimates of high-grade brain gliomas (HGG) due to the use of an input function (IF) obtained respectively from arterial (AIF) and venous (VIF) approaches by two different commercially available software applications. METHODS: This prospective study includes 20 patients with pathologically confirmed diagnosis of high-grade gliomas. The data source was processed by using two DCE dedicated commercial packages, both based on the extended Toft model, but the first customized to obtain input function from arterial measurement and the second from sagittal sinus sampling. The quantitative parametric perfusion maps estimated from the two software packages were compared by means of a region of interest (ROI) analysis. The resulting input functions from venous and arterial data were also compared. RESULTS: No significant difference has been found between the perfusion parameters obtained with the two different software packages (P-value < .05). The comparison of the VIFs and AIFs obtained by the two packages showed no statistical differences. CONCLUSIONS: Direct comparison of DCE-MRI measurements with IF generated by means of arterial or venous waveform led to no statistical difference in quantitative metrics for evaluating HGG. However, additional research involving DCE-MRI acquisition protocols and post-processing would be beneficial to further substantiate the effectiveness of venous approach as the IF method compared with arterial-based IF measurement.
Keywords: Arterial input function, venous input function, DCE-MRI, comparison, brain perfusion. Acceptance: Received February 4, 2015. Accepted for publication March 26, 2015. Correspondence: Address correspondence to Silvano Filice, MSc, Department of Medical Physics, University Hospital of Parma, via Gramsci 14, 43126, Parma, Italy. E-mail: [email protected]. Conflict of Interests: We declare that we have no conflict of interest. J Neuroimaging 2016;26:124-129. DOI: 10.1111/jon.12254
Introduction In the setting of brain tumors, magnetic resonance imaging (MRI) perfusion is being increasingly used to provide functional information about tumor vascular physiology and vessel permeability.1–3 Both dynamic susceptibility contrast (DSC) and dynamic contrast-enhanced (DCE) MRI perfusion techniques require an analysis of the time dependence of the contrast first-pass kinetic and clearance by using a pharmacokinetic model. After data acquisition, kinetic models can be applied in order to derive estimates of tissue perfusion and permeability based on the shape of lesion of interest wash-in and wash-out curves. Perfusion tracer kinetics theory has been described extensively in the literature.4–6 The tissue uptake curve is just the output response function to the input pulse of the contrast agent’s bolus that is generally referred to as the input function (IF). Many searching algorithms have been proposed to identify on the images the suitable pixels for IF measurement.7 Maps of perfusion parameters can be obtained by using many of the commercially available software applications. There is, however, a pressing need to standardize the acquisition and analysis of MRI perfusion data to allow the comparison of findings ob-
124
Copyright
tained at different sites and across different studies.8–10 In this respect, it is well known that the distortions of IF from its real shape or attenuation can affect accuracy and reproducibility of DSC-MRI and DCE-MRI perfusion outcome parameters.11–14 In most published data, the IF calculations for the analysis of MRI perfusion contrast kinetic are obtained by arterial data.15,16 Venous data have been proposed for T1-based DCE-MRI perfusion measurements17,18 but though they are more easily obtainable on time-attenuation curves within the confines of a large vessel, they reflect an output rather than an input function and thus would yield a distorted picture of contrast kinetics. Blood contrast transit time is longer and bolus sharpness would be lost to the intervening tissues. The expected differences between the pattern of DCE-MRI events seen in an arterial input function (AIF) and that seen in a vein input function (VIF) have not been documented and discussed previously. It would therefore be of interest to compare the parameters of the DCE model derived from arterial data with those derived from venous data. The primary aim of the current study was to evaluate whether differences exist in DCE-MRI parameter calculation between IF data collected from arterial and venous approaches by
◦ 2015 by the American Society of Neuroimaging C
Table 1. Patient Demographics, Clinical Data, and Pathologic Findings Variable
No. of cases Sex (M/F) Age range-mean (yr) Histology Metylation of O6 – Methylguanine-DNA methyltransferase (MGMT) promoter
WHO Grade III
4 2:2 32-48 (41) Anaplastic Astrocytoma Yes (1)
No (3)
WHO Grade IV
16 10:6 34-64 (52) Glioblastoma Yes (7)
No (9)
using two different commercially available software applications. The secondary aim was to discuss the IF problem as it applies to the time dependence and kinetics of contrast agent (CA) with the compartmental DCE-MRI model when arterial data cannot be obtained.
Materials and Methods Patients Approval for this prospective study was obtained from the local institutional review board, and signed informed consent was obtained from all patients. Inclusion criteria were predefined as follows: patients with a contrast-enhanced brain MR imaging that documents evidence of primary high-grade brain tumor (HGG). Magnetic resonance imaging studies were performed in 20 patients (Table 1) with high-grade brain tumor (16 Glioblastoma and 4 Anaplastic Astrocytoma). Histologic diagnosis was obtained in all subjects by surgical resection.
MRI Acquisition Protocol Magnetic resonance imaging and DCE-MRI examinations were performed by using a 3 T whole-body system (Discovery MR750; GE Healthcare, Milwaukee, WI) equipped with an 8-channel phased-array head coil. The lesions were initially identified using T1-weighted fluid attenuated inversion recovery sagittal images, fluid attenuated inversion recovery T2-weighted, and T2*-weighted gradient echo axial images and diffusion weighted axial images. DCE-MRI dynamic scan was acquired using 3D T1-weighted Fast Spoiled Gradient Echo sequence and setting the following sequence parameters: repetition time/echo time/flip angle (TE/TR/FA) = 1.9 ms/5.0 ms/20°, acquisition matrix = 160×160 with partial Fourier (.75), voxel size = 1.5 × 1.5 × 4 mm3 , receiver bandwidth ± 31.25 KHz (390.6 Hz/pixel), and the number of images for single dynamic phase = 24. Parallel imaging was applied in phase-encoding direction to its maximum value to increase temporal resolution with a resulting single-phase sampling time (ST) of 4.9 seconds. DCE-MRI scan consisted in a 59 dynamic phase with a total scan time (TA) of 289 seconds. The first 4 phases were used to define the baseline of signal– to-time curve (STC). Patients were injected with 0.2 R , Guermmol/kg of gadoterate meglumine (Dotarem bet USA, Bloomington, IN) using a power injector at a rate of 5 mL/second, immediately followed by a 20 mL bolus of saline injected at the same rate. After DCE-MRI scan, post-contrast high-resolution 3-dimensional T1-weighted and 2-dimensional T1-weighted axial images were acquired.
Image Processing DCE data were processed by using the two commercial packages for DCE-MRI analysis nordicICE DCE Module ver. 2.3.10 (NordicNeurolab, Bergen, Netherlands) and GenIQ version vxtl11381 installed on GE ADW4.3 (GE Healthcare, Milwaukee, WI). Both software applications are based on the twocompartment tissue kinetic model or extended Toft model and use a fully automated pixels searching algorithm for IF detection. The NordicICE module is customized to obtain IF from arterial sampling while venous data are processed by GenIQ. The NordicICE module was set to detect AIF on six different pixels of as many different brain arteries, whereas with GenIQ the VIF was measured on six separate pixels localized in the sagittal sinus. In both cases, the IF used for calculation is the mean of the six different measures. The kinetic model describes the physiological CA exchange between blood pool (BP) and the extracellular extravacular space (EES) allowing the estimates of perfusion-related parameters such as the transfer rate coefficient Ktrans , given in min−1 , representing the CA transfer rate from the BP to EES, the fractional volumes of BP and EES denoted as Ve and Vp , with values ranging between 0 and 1. To obtain the quantitative perfusion parametrics maps that are respectively the space distribution of Ktrans , Ve , and Vp , DCE dynamic series were preprocessed by an automated motion correction tool. Furthermore, fixed baseline T1 values of both tissue and blood were set to 1,250 and 1,600 ms, respectively. A blood hematocrit constant of .45 was assumed. The perfusion parametric maps were co-registered on T1-weighted contrastenhanced images as a guide to the perfusion analysis (Fig 1). The delineation of the regions of interest (ROIs) for quantitative data comparison was manually defined by a trained neuroradiologist using co-registered T1-weighted contrast-enhanced images as a reference. The ROIs were set to include perfusion abnormalities within Ktrans hot spots (Fig 1). To reproduce ROI’s position across the analysis on the two different packages for a given patient, dicom coordinates of T1-weighted contrast-enhanced images were used, and ROI’s area and shape were set identical. The values of Ktrans , Ve , and Vp were obtained from all the ROIs for every patient by both software packages, Vp higher than .2, indicating a vessel inclusion into the area of analysis, were excluded because they would lead to abnormal values of Ktrans and Ve .
Results One patient was excluded from the study because the resulting Ve was greater than 1 due to insufficient acquisition time.4 The statistical independence of the groups of measured values was tested with the Mann-Whitney U test. A P-value of less than .05 was considered to indicate a statistically significant difference. Results are summarized in Table 2 and Fig 2. No difference has been found between the perfusion parameters obtained with GenIQ or NordicICE. In order to compare the VIF and AIF (Fig 3) five different parameters were obtained from each waveform (Fig 4): the IF maximum, the width at .65 of maximum (L65), the average slope during the phases of washin (WIS) and wash-out (WOS), and the average slope during the slow wash-out phase (SWOS). The statistical independence of the five groups of values was tested with the Mann-Whitney U test. Results are summarized in Table 3. Insignificant difference has been found between all the parameters evaluated,
Filice and Crisi: The Role of Input Function in DCE-MRI Metrics
125
Fig 1. Left frontal Parietal Glioblastoma in a 52-year-old man. DCE-MRI parametric perfusion color maps overlayed onto post-contrast T1w image generated from GenIQ (upper panel) and NordicICE DCE Module (lower panel). (A, D) Transfer rate coefficient (Ktrans ) maps with ROI overimposed (red circle) to include hot spot areas. (B, E) Extravascular extracellular volume (Ve ) maps. (C, F) Plasma volume (Vp ) maps.
Table 2. Mean Value and Standard Deviation of Dynamic ContrastEnhanced MRI Perfusion Tissue Parameters for High Grade Brain Gliomas
VIF (GenIQ) AIF (NordicICE) P value
Ktrans (min−1 )
Ve
Vp
0.22 ± 0.10 0.18 ± 0.12 0.55
0.45 ± 0.21 0.38 ± 0.15 0.61
0.10 ± 0.05 0.09 ± 0.05 0.8
The values in the upper row were obtained from GenIQ package using venous input function (VIF) as input data. The values in the central raw come from NordicICE by using arterial input function (AIF). The P-value (bottom row) indicates the significance of the statistical dependence between groups of estimated values.
meaning no significant difference between AIF and VIF. In Fig 5 are shown the average VIF and AIF calculated as the mean of each patient’s IF: the two resulting curves were similar with differences inside statistical data dispersion (one standard deviation). For each patient the partial volume effect (PVE) was calculated as the ratio of the integral over time of VIF/AIF, assuming the CA leakage due to BBB disruption as negligible. The PVE ranged between .79 and 1.2 with a mean value and a standard deviation of .92 ± .14.
126
Discussion DCE-MRI is based on indicator dilution theory and requires the measurement of time course concentration of the contrast agent (CA) in the blood plasma or input function (IF).4–6 The acquisition of IF can be impaired by possible measurement errors, patient movements and much more.11–14 A poorly characterized IF can result in estimation errors of perfusion kinetic parameters. Many approaches have been proposed to derive IF, including the use of a fixed pre-modeled function19 or a standard function like, for example, the Weinmann function.20 Currently, it is believed that the measure of IF is necessary to take into account inter-patient variability related to cardiac output, cardiac rate, kidney function, arterial status, body weight, and more.21,22 Although in principle IF should be measured in arteries, some authors suggest that venous data may also be used for this purpose.17,18 In this study, we have compared DCE-MRI quantitative perfusion metrics obtained with IF data collected from arterial and venous approaches by using two different commercially available software applications. We assumed IF changes as the main source of variability in DCE-MRI quantification of vessel permeability and that other sources of error, such as software-related, would affect the results to a lesser extent. Also, fixed baseline T1 values for tissue,
Journal of Neuroimaging Vol 26 No 1 January/February 2016
Table 3. Mean Value and Standard Deviations of the Parameters Obtained from VIFs and AIFs Analysis
VIF (GenIQ) AIF (NordicICE) P-value
IF maximum (a.u.)
L65 (sec)
WIS (a.u.)/sec
WOS (a.u.)/sec
SWOS (a.u.)/sec
8.2 ± 1.6 9.1 ± 1.1 0.156
15.8 ± 2.5 16.6 ± 2.9 0.494
0.81 ± 0.2 0.71 ± 0.16 0.156
−0.49 ± 0.60 −0.38 ± 0.13 0.429
−0.040 ± 0.04 −0.041 ± 0.04 0.833
The P-value in the bottom row indicates the significance of the statistical dependence between groups of estimated values. IF = input function; VIF = venous input function; AIF = arterial input function; L65 = width at 0.65 of IF maximum; WIS = IF average slope during the wash-in phase; WOS = IF average slope during the wash-out phase; SWOS = IF average slope during the slow wash-out phase.
Fig 3. Automatically extracted arterial input functions and venous input functions from 8 patients.
Fig 2. Box plot of tissue perfusion parameters transfer rate coefficient (Ktrans ), extravascular extracellular volume (Ve) and plasma volume (Vp) obtained from GenIQ (left side) and from NordicICE (right side). The box plots show the median value of each parameters (middle line) with the box representing the 95th and 75th percentile of data distribution. No significant difference has been found between the two populations of measure.
we adopted, are consistent with values reported by Larsson et al suggesting that T1 mapping pixel-wised may be unnecessary.23 We found no significant difference in the estimation of Ktrans , Ve and Vp obtained using VIF or AIF as input data. Moreover, our data are in agreement with the published literature.24–27 These findings are consistent with the results of VIFs and AIFs comparison which yielded no significant difference between the two waveforms. As principle, sampling in a venous vessel represents a true output function which should differ from input function obtained in an artery. The difference is mainly related to contrast agent (CA) dispersion due to the tissue microvasculature properties, and it is expected that the dispersion mainly
affects the amplitude and width of the bolus “first peak” during the transit from the arteries to the sagittal sinus. To explain our results, we suppose that measure-related effects,28–31 mainly T2* shortening effect, blood in-flow, sampling rate, and partial volume, may hamper the accuracy of the measure and combine in such a way that VIF and AIF waveforms resulted equivalent. Input function is characterized by a steep slope wash-in phase, where CA rapidly increases from zero to reach its maximum concentration, and then decreases giving rise to a so-called first peak. Recirculation likewise results in a second smaller peak followed by the wash-out phase, lasting several minutes, where CA slowly leaves the blood pool. The CA causes a decrease of blood T1 relaxivity, depending on its concentration, generating an increase of MR signal on T1-weighted images. The function resulting from dynamically sampling blood signal changes over time (SCT) describes the relaxivity effect of contrast transit. The IF is estimated from SCT, assuming a linear relationship between T1 change and CA concentration. This assumption is only roughly true. For high CA concentration, the T2* shortening effect of gadolinium produces a nonlinear CA dose response. Signal saturation occurs meaning that a further reduction in T1 does not result in a further increase in signal
Filice and Crisi: The Role of Input Function in DCE-MRI Metrics
127
Fig 4. Parameters obtained from a typical input function waveform.
Fig 5. Mean AIF and mean VIF. Error bar depict ± SD across the population of 20 patients.
and when the concentration becomes higher the signal starts to fall, since the T2* effects of the contrast agent dominate signal behavior. Signal saturation mainly results in an underestimation of IF first peak. The CA concentration threshold for signal saturation depends on TE. Kleppestø et al report,28 for TE of 2 ms, an error of 10% on Ktrans and Vp estimation at CA concentration of 10 mM and 7 mM, respectively. Blood in-flow effect manifests as an MR signal enhancement due to the flow of relaxed proton spin into the imaging volume and leading to an increase in signal intensity. In-flow effect is dependent on blood velocity and direction, slice location inside acquisition volume, and imaging parameters (TE, TR, and flip angle). Blood in-flow causes the overestimation of IF. To the opposite, given the large size of the superior sagittal sinus compared to the cerebral arteries, it certainly would benefit DCE-MRI to eliminate partial volume and motion artifacts by using the waveform sampled there. As known, the partial volume effects are manifested if vessel diameter is small compared with size of the voxel and 128
increase when lowering spatial resolution. In this case, a certain amount of static tissue is included in the voxel producing a lack of signal involving AIF attenuation or distortion.31 Our data showed a mean AIF attenuation due to partial volume effects about 8%. It is well known that the accuracy of DCE-MRI kinetic estimates can be affected by the choice of the temporal resolution or sampling time. At low sampling rate input function aliasing occurs. In a study by Larsson et al, the influence of sampling time was evaluated on the estimation of kinetic parameters in a HGG cohort of patients.30 The authors recommend using a temporal resolution lower than 20 seconds to minimize the uncertainty due to sampling error and ensure that tumor “hot spots” with rapid CA extravasation were not missed. In a systematic review, Heye et al analyzed 30 different DCEMRI studies concerning brain malignancy finding that the median temporal resolution adopted in the scanning protocols was 5.3 seconds.9 A 4.9 seconds sampling time as in the present study is a trade-off between the need for an appropriate
Journal of Neuroimaging Vol 26 No 1 January/February 2016
temporal resolution and control of spatial resolution, coverage and signal-to-noise ratio. Recently, Shin et al reported that DCE-MRI seems to be more powerful than DSC-MRI in differentiation of recurrences from treatment-related changes.32 They used a temporal resolution approximately of 5.5 seconds to calculate IF and Ktrans . In addition, their study was based upon commercially available software which can be easily adapted to routine practice. Our results indicate that under the specific circumstances of the described DCE-MRI acquisition technique, VIF and AIF are interchangeable in evaluating HGG. Similarly to our DCE-MRI technique and post-processing, Hamilton et al used a VIF method in order to differentiate HGG recurrence from treatment effect.18 They also concluded that IF derived from the sagittal sinus appeared to be superior to that from the middle cerebral artery. In conclusions, we have demonstrated that IF obtained by using arterial or venous waveform led to no statistical difference in DCE-MRI quantitative metrics of high-grade gliomas. However, additional research involving DCE-MRI acquisition protocols and post-processing would be beneficial to further substantiate the effectiveness of venous approach as the IF method compared with arterial-based IF measurement.
References 1. Jalali S, Chung C, Foltz W, et al. MRI biomarkers identify the differential response of glioblastoma multiforme to anti-angiogenic therapy. Neuro Oncol 2014;16:868-79. 2. Ahn SS, Shin NY, Chang JH, et al. Prediction of methylguanine methyltransferase promoter methylation in glioblastoma using dynamic contrast-enhanced magnetic resonance and diffusion tensor imaging. Neurosurgery 2014;121:367-73. 3. O’Connor JPB, Jackson A, Parker GJM, et al. DCE-MRI biomarkers in the clinical evaluation of antiangiogenic and vascular disrupting agents. Br J Cancer 2007;96:189-95. 4. Tofts PS, Brix G, Buckley DL, et al. Estimating kinetic parameters from dynamic contrast enhanced T1-w MRI of a diffusible tracer: standardized quantities and symbols. J Magn Reson Imaging 1999;10:223-32. 5. Toft PS, Oxon BA. Modeling tracer kinetics in dynamic Gd-DTPA MR imaging. JMRI 1997;7:91-101. 6. Sourbron S, Ingrisch M, Siefert A, et al. Quantification of cerebral blood flow, cerebral blood volume, and blood–brain-barrier leakage with DCE-MRI. Magn Reson Med 2009; 62:205-17. 7. Calamante F. Arterial input function in perfusion MRI: a comprehensive review. Prog Nucl Magn Reson Spectrosc 2013;74:1-32. 8. Bernstein JM, Homer JJ, West CM. Dynamic contrast-enhanced magnetic resonance imaging biomarkers in head and neck cancer: potential to guide treatment? A systematic review. Oral Oncol 2014;50:963-70. 9. Heye AK, Culling RD, Valdes MC, et al. Assessment ´ Hernandez ´ of blood-brain barrier disruption using dynamic contrast-enhanced MRI. A systematic review. Neuroimage Clin 2014;6:262-74. 10. O’Connor JP, Jackson A, Parker GJ, et al. Dynamic contrastenhanced MRI in clinical trials of antivascular therapies. Nat Rev Clin Oncol 2012;9:167-77. 11. Buckley DL. Uncertainty in the analysis of tracer kinetics using dynamic contrast-enhanced T1-weighted MRI. Magn Reson Med 2002;47:601-6. 12. Yang C, Karczmar GS, Medved M, et al. Estimating the arterial input function using two reference tissues in dynamic contrastenhanced MRI studies: fundamental concepts and simulations. Magn Reson Med 2004;52:1110-7. 13. Lavini C. Simulating the effect of input errors on the accuracy of Tofts pharmacokinetic model parameters. Magn Reson Imaging 2015;33:222-35.
14. Van Osch MJP, van der Grond J, Bakker CJG. Partial volume effects on arterial input functions: shape and amplitude distortions and their correction. J Magn Reson Imaging 2005;22:704-9. 15. Fluckiger JU, Schabel MC, DiBella EV. Toward local arterial input functions in dynamic contrast-enhanced MRI. J Magn Reson Imaging 2010;32:924-34. 16. Calamante F. Arterial input function in perfusion MRI: A comprehensive review. Prog Nucl Magn Reson Spectrosc 2013;74:132. 17. Yuan J, Chow SKK, Zhang Q, et al. The use of dynamic tracer concentration in veins for quantitative DCE-MRI kinetic analysis in head and neck. Plos One 2013;8:e59885. 18. Hamilton JD, Lin J, Ison C, et al. Dynamic contrast-enhanced perfusion processing for neuroradiologists: model-dependent analysis may not be necessary for determining recurrent high-grade glioma versus treatment effect. AJNR Am J Neuroradiol 2015;36: 686-93. 19. Parker GJ, Roberts C, Macdonald A, et al. Experimentallyderived functional form for a population-averaged high-temporalresolution arterial input function for dynamic contrast-enhanced MRI. Magn Reson Med 2006;56:993-1000. 20. Singh A, Haris M, Rathore D, et al. Quantification of physiological and hemodynamic indices using T(1) dynamic contrastenhanced MRI in intracranial mass lesions. J Magn Reson Imaging 2007;26:871-80. 21. Port RE, Knopp MV, Brix G. Dynamic contrast-enhanced MRI using Gd-DTPA: interindividual variability of the arterial input function and consequences for the assessment of kinetics in tumors. Magn ResonMed 2001;45:1030-8. 22. Lavini C, Verhoeff JJ. Reproducibility of the gadolinium concentration measurements and of the fitting parameters of the vascular input function in the superior sagittal sinus in a patient population. Magn Reson Imaging 2010;28:1420-30. 23. Larsson C1, Kleppestø M, Grothe I, et al. T1 in high-grade glioma and the influence of different measurement strategies on parameter estimations in DCE-MRI. J Magn Reson Imaging 2015;42: 97-104. 24. Zhang N, Zhang L, Qiu B, et al. Correlation of volume transfer coefficient Ktrans with histopathologic grades of gliomas. J Magn Reson Imaging 2012;36:355-63. 25. Nguyen TB, Cron GO, Mercier JF, et al. Diagnostic accuracy of dynamic contrast-enhanced MR imaging using a phase-derived vascular input function in the preoperative grading of gliomas. AJNR 2012;33:1539-45. 26. Eliat PA, Olivie´ D, Sa¨ıkali S, et al. Can dynamic contrast-enhanced magnetic resonance imaging combined with texture analysis differentiate malignant glioneuronal tumors from other glioblastoma? Neurol Res International 2012; 195176 doi: 10.1155/2012/195176. 27. Choi HS, Kim AH, Ahn SS, et al. Glioma grading capability: comparisons among parameters from dynamic contrast-enhanced MRI and ADC value on DWI. Korean J Radiol 2013;14:48792. 28. Kleppestø M, Larsson C, Groote I, et al. T2*-correction in dynamic contrast-enhanced MRI from double-echo acquisitions. J Magn Reson Imaging 2014;39:1314-9. 29. Roberts C, Little R, Watson Y, et al. The effect of blood inflow and B(1)-field inhomogeneity on measurement of the arterial input function in axial 3D spoiled gradient echo dynamic contrast-enhanced MRI. Magn Reson Med 2011;65:108-19. 30. Larsson C, Kleppestø M, Rasmussen I, et al. Sampling requirements in DCE-MRI based analysis of high grade gliomas: simulations and clinical results. J Magn Reson Imaging 2013;37:818-29. 31. Knutsson L, Borjesson S, Larsson EM, et al. Analysis of partial volume effects on arterial input functions using gradient echo: a simulation study. Magn Reson Med 2009;61:1300-9. 32. Shin KE, Ahn KJ, Choi HS, et al. DCE and DSC MR perfusion imaging in the differentiation of recurrent tumour from treatmentrelated changes in patients with glioma. Clin Radiol 2014;69:26472.
Filice and Crisi: The Role of Input Function in DCE-MRI Metrics
129
ABSTRACT BACKGROUND AND PURPOSE: The aim of this study was to evaluate the differences in dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) perfusion estimates of high-grade brain gliomas (HGG) due to the use of an input function (IF) obtained respectively from arterial (AIF) and venous (VIF) approaches by two different commercially available software applications. METHODS: This prospective study includes 20 patients with pathologically confirmed diagnosis of high-grade gliomas. The data source was processed by using two DCE dedicated commercial packages, both based on the extended Toft model, but the first customized to obtain input function from arterial measurement and the second from sagittal sinus sampling. The quantitative parametric perfusion maps estimated from the two software packages were compared by means of a region of interest (ROI) analysis. The resulting input functions from venous and arterial data were also compared. RESULTS: No significant difference has been found between the perfusion parameters obtained with the two different software packages (P-value < .05). The comparison of the VIFs and AIFs obtained by the two packages showed no statistical differences. CONCLUSIONS: Direct comparison of DCE-MRI measurements with IF generated by means of arterial or venous waveform led to no statistical difference in quantitative metrics for evaluating HGG. However, additional research involving DCE-MRI acquisition protocols and post-processing would be beneficial to further substantiate the effectiveness of venous approach as the IF method compared with arterial-based IF measurement.
Keywords: Arterial input function, venous input function, DCE-MRI, comparison, brain perfusion. Acceptance: Received February 4, 2015. Accepted for publication March 26, 2015. Correspondence: Address correspondence to Silvano Filice, MSc, Department of Medical Physics, University Hospital of Parma, via Gramsci 14, 43126, Parma, Italy. E-mail: [email protected]. Conflict of Interests: We declare that we have no conflict of interest. J Neuroimaging 2016;26:124-129. DOI: 10.1111/jon.12254
Introduction In the setting of brain tumors, magnetic resonance imaging (MRI) perfusion is being increasingly used to provide functional information about tumor vascular physiology and vessel permeability.1–3 Both dynamic susceptibility contrast (DSC) and dynamic contrast-enhanced (DCE) MRI perfusion techniques require an analysis of the time dependence of the contrast first-pass kinetic and clearance by using a pharmacokinetic model. After data acquisition, kinetic models can be applied in order to derive estimates of tissue perfusion and permeability based on the shape of lesion of interest wash-in and wash-out curves. Perfusion tracer kinetics theory has been described extensively in the literature.4–6 The tissue uptake curve is just the output response function to the input pulse of the contrast agent’s bolus that is generally referred to as the input function (IF). Many searching algorithms have been proposed to identify on the images the suitable pixels for IF measurement.7 Maps of perfusion parameters can be obtained by using many of the commercially available software applications. There is, however, a pressing need to standardize the acquisition and analysis of MRI perfusion data to allow the comparison of findings ob-
124
Copyright
tained at different sites and across different studies.8–10 In this respect, it is well known that the distortions of IF from its real shape or attenuation can affect accuracy and reproducibility of DSC-MRI and DCE-MRI perfusion outcome parameters.11–14 In most published data, the IF calculations for the analysis of MRI perfusion contrast kinetic are obtained by arterial data.15,16 Venous data have been proposed for T1-based DCE-MRI perfusion measurements17,18 but though they are more easily obtainable on time-attenuation curves within the confines of a large vessel, they reflect an output rather than an input function and thus would yield a distorted picture of contrast kinetics. Blood contrast transit time is longer and bolus sharpness would be lost to the intervening tissues. The expected differences between the pattern of DCE-MRI events seen in an arterial input function (AIF) and that seen in a vein input function (VIF) have not been documented and discussed previously. It would therefore be of interest to compare the parameters of the DCE model derived from arterial data with those derived from venous data. The primary aim of the current study was to evaluate whether differences exist in DCE-MRI parameter calculation between IF data collected from arterial and venous approaches by
◦ 2015 by the American Society of Neuroimaging C
Table 1. Patient Demographics, Clinical Data, and Pathologic Findings Variable
No. of cases Sex (M/F) Age range-mean (yr) Histology Metylation of O6 – Methylguanine-DNA methyltransferase (MGMT) promoter
WHO Grade III
4 2:2 32-48 (41) Anaplastic Astrocytoma Yes (1)
No (3)
WHO Grade IV
16 10:6 34-64 (52) Glioblastoma Yes (7)
No (9)
using two different commercially available software applications. The secondary aim was to discuss the IF problem as it applies to the time dependence and kinetics of contrast agent (CA) with the compartmental DCE-MRI model when arterial data cannot be obtained.
Materials and Methods Patients Approval for this prospective study was obtained from the local institutional review board, and signed informed consent was obtained from all patients. Inclusion criteria were predefined as follows: patients with a contrast-enhanced brain MR imaging that documents evidence of primary high-grade brain tumor (HGG). Magnetic resonance imaging studies were performed in 20 patients (Table 1) with high-grade brain tumor (16 Glioblastoma and 4 Anaplastic Astrocytoma). Histologic diagnosis was obtained in all subjects by surgical resection.
MRI Acquisition Protocol Magnetic resonance imaging and DCE-MRI examinations were performed by using a 3 T whole-body system (Discovery MR750; GE Healthcare, Milwaukee, WI) equipped with an 8-channel phased-array head coil. The lesions were initially identified using T1-weighted fluid attenuated inversion recovery sagittal images, fluid attenuated inversion recovery T2-weighted, and T2*-weighted gradient echo axial images and diffusion weighted axial images. DCE-MRI dynamic scan was acquired using 3D T1-weighted Fast Spoiled Gradient Echo sequence and setting the following sequence parameters: repetition time/echo time/flip angle (TE/TR/FA) = 1.9 ms/5.0 ms/20°, acquisition matrix = 160×160 with partial Fourier (.75), voxel size = 1.5 × 1.5 × 4 mm3 , receiver bandwidth ± 31.25 KHz (390.6 Hz/pixel), and the number of images for single dynamic phase = 24. Parallel imaging was applied in phase-encoding direction to its maximum value to increase temporal resolution with a resulting single-phase sampling time (ST) of 4.9 seconds. DCE-MRI scan consisted in a 59 dynamic phase with a total scan time (TA) of 289 seconds. The first 4 phases were used to define the baseline of signal– to-time curve (STC). Patients were injected with 0.2 R , Guermmol/kg of gadoterate meglumine (Dotarem bet USA, Bloomington, IN) using a power injector at a rate of 5 mL/second, immediately followed by a 20 mL bolus of saline injected at the same rate. After DCE-MRI scan, post-contrast high-resolution 3-dimensional T1-weighted and 2-dimensional T1-weighted axial images were acquired.
Image Processing DCE data were processed by using the two commercial packages for DCE-MRI analysis nordicICE DCE Module ver. 2.3.10 (NordicNeurolab, Bergen, Netherlands) and GenIQ version vxtl11381 installed on GE ADW4.3 (GE Healthcare, Milwaukee, WI). Both software applications are based on the twocompartment tissue kinetic model or extended Toft model and use a fully automated pixels searching algorithm for IF detection. The NordicICE module is customized to obtain IF from arterial sampling while venous data are processed by GenIQ. The NordicICE module was set to detect AIF on six different pixels of as many different brain arteries, whereas with GenIQ the VIF was measured on six separate pixels localized in the sagittal sinus. In both cases, the IF used for calculation is the mean of the six different measures. The kinetic model describes the physiological CA exchange between blood pool (BP) and the extracellular extravacular space (EES) allowing the estimates of perfusion-related parameters such as the transfer rate coefficient Ktrans , given in min−1 , representing the CA transfer rate from the BP to EES, the fractional volumes of BP and EES denoted as Ve and Vp , with values ranging between 0 and 1. To obtain the quantitative perfusion parametrics maps that are respectively the space distribution of Ktrans , Ve , and Vp , DCE dynamic series were preprocessed by an automated motion correction tool. Furthermore, fixed baseline T1 values of both tissue and blood were set to 1,250 and 1,600 ms, respectively. A blood hematocrit constant of .45 was assumed. The perfusion parametric maps were co-registered on T1-weighted contrastenhanced images as a guide to the perfusion analysis (Fig 1). The delineation of the regions of interest (ROIs) for quantitative data comparison was manually defined by a trained neuroradiologist using co-registered T1-weighted contrast-enhanced images as a reference. The ROIs were set to include perfusion abnormalities within Ktrans hot spots (Fig 1). To reproduce ROI’s position across the analysis on the two different packages for a given patient, dicom coordinates of T1-weighted contrast-enhanced images were used, and ROI’s area and shape were set identical. The values of Ktrans , Ve , and Vp were obtained from all the ROIs for every patient by both software packages, Vp higher than .2, indicating a vessel inclusion into the area of analysis, were excluded because they would lead to abnormal values of Ktrans and Ve .
Results One patient was excluded from the study because the resulting Ve was greater than 1 due to insufficient acquisition time.4 The statistical independence of the groups of measured values was tested with the Mann-Whitney U test. A P-value of less than .05 was considered to indicate a statistically significant difference. Results are summarized in Table 2 and Fig 2. No difference has been found between the perfusion parameters obtained with GenIQ or NordicICE. In order to compare the VIF and AIF (Fig 3) five different parameters were obtained from each waveform (Fig 4): the IF maximum, the width at .65 of maximum (L65), the average slope during the phases of washin (WIS) and wash-out (WOS), and the average slope during the slow wash-out phase (SWOS). The statistical independence of the five groups of values was tested with the Mann-Whitney U test. Results are summarized in Table 3. Insignificant difference has been found between all the parameters evaluated,
Filice and Crisi: The Role of Input Function in DCE-MRI Metrics
125
Fig 1. Left frontal Parietal Glioblastoma in a 52-year-old man. DCE-MRI parametric perfusion color maps overlayed onto post-contrast T1w image generated from GenIQ (upper panel) and NordicICE DCE Module (lower panel). (A, D) Transfer rate coefficient (Ktrans ) maps with ROI overimposed (red circle) to include hot spot areas. (B, E) Extravascular extracellular volume (Ve ) maps. (C, F) Plasma volume (Vp ) maps.
Table 2. Mean Value and Standard Deviation of Dynamic ContrastEnhanced MRI Perfusion Tissue Parameters for High Grade Brain Gliomas
VIF (GenIQ) AIF (NordicICE) P value
Ktrans (min−1 )
Ve
Vp
0.22 ± 0.10 0.18 ± 0.12 0.55
0.45 ± 0.21 0.38 ± 0.15 0.61
0.10 ± 0.05 0.09 ± 0.05 0.8
The values in the upper row were obtained from GenIQ package using venous input function (VIF) as input data. The values in the central raw come from NordicICE by using arterial input function (AIF). The P-value (bottom row) indicates the significance of the statistical dependence between groups of estimated values.
meaning no significant difference between AIF and VIF. In Fig 5 are shown the average VIF and AIF calculated as the mean of each patient’s IF: the two resulting curves were similar with differences inside statistical data dispersion (one standard deviation). For each patient the partial volume effect (PVE) was calculated as the ratio of the integral over time of VIF/AIF, assuming the CA leakage due to BBB disruption as negligible. The PVE ranged between .79 and 1.2 with a mean value and a standard deviation of .92 ± .14.
126
Discussion DCE-MRI is based on indicator dilution theory and requires the measurement of time course concentration of the contrast agent (CA) in the blood plasma or input function (IF).4–6 The acquisition of IF can be impaired by possible measurement errors, patient movements and much more.11–14 A poorly characterized IF can result in estimation errors of perfusion kinetic parameters. Many approaches have been proposed to derive IF, including the use of a fixed pre-modeled function19 or a standard function like, for example, the Weinmann function.20 Currently, it is believed that the measure of IF is necessary to take into account inter-patient variability related to cardiac output, cardiac rate, kidney function, arterial status, body weight, and more.21,22 Although in principle IF should be measured in arteries, some authors suggest that venous data may also be used for this purpose.17,18 In this study, we have compared DCE-MRI quantitative perfusion metrics obtained with IF data collected from arterial and venous approaches by using two different commercially available software applications. We assumed IF changes as the main source of variability in DCE-MRI quantification of vessel permeability and that other sources of error, such as software-related, would affect the results to a lesser extent. Also, fixed baseline T1 values for tissue,
Journal of Neuroimaging Vol 26 No 1 January/February 2016
Table 3. Mean Value and Standard Deviations of the Parameters Obtained from VIFs and AIFs Analysis
VIF (GenIQ) AIF (NordicICE) P-value
IF maximum (a.u.)
L65 (sec)
WIS (a.u.)/sec
WOS (a.u.)/sec
SWOS (a.u.)/sec
8.2 ± 1.6 9.1 ± 1.1 0.156
15.8 ± 2.5 16.6 ± 2.9 0.494
0.81 ± 0.2 0.71 ± 0.16 0.156
−0.49 ± 0.60 −0.38 ± 0.13 0.429
−0.040 ± 0.04 −0.041 ± 0.04 0.833
The P-value in the bottom row indicates the significance of the statistical dependence between groups of estimated values. IF = input function; VIF = venous input function; AIF = arterial input function; L65 = width at 0.65 of IF maximum; WIS = IF average slope during the wash-in phase; WOS = IF average slope during the wash-out phase; SWOS = IF average slope during the slow wash-out phase.
Fig 3. Automatically extracted arterial input functions and venous input functions from 8 patients.
Fig 2. Box plot of tissue perfusion parameters transfer rate coefficient (Ktrans ), extravascular extracellular volume (Ve) and plasma volume (Vp) obtained from GenIQ (left side) and from NordicICE (right side). The box plots show the median value of each parameters (middle line) with the box representing the 95th and 75th percentile of data distribution. No significant difference has been found between the two populations of measure.
we adopted, are consistent with values reported by Larsson et al suggesting that T1 mapping pixel-wised may be unnecessary.23 We found no significant difference in the estimation of Ktrans , Ve and Vp obtained using VIF or AIF as input data. Moreover, our data are in agreement with the published literature.24–27 These findings are consistent with the results of VIFs and AIFs comparison which yielded no significant difference between the two waveforms. As principle, sampling in a venous vessel represents a true output function which should differ from input function obtained in an artery. The difference is mainly related to contrast agent (CA) dispersion due to the tissue microvasculature properties, and it is expected that the dispersion mainly
affects the amplitude and width of the bolus “first peak” during the transit from the arteries to the sagittal sinus. To explain our results, we suppose that measure-related effects,28–31 mainly T2* shortening effect, blood in-flow, sampling rate, and partial volume, may hamper the accuracy of the measure and combine in such a way that VIF and AIF waveforms resulted equivalent. Input function is characterized by a steep slope wash-in phase, where CA rapidly increases from zero to reach its maximum concentration, and then decreases giving rise to a so-called first peak. Recirculation likewise results in a second smaller peak followed by the wash-out phase, lasting several minutes, where CA slowly leaves the blood pool. The CA causes a decrease of blood T1 relaxivity, depending on its concentration, generating an increase of MR signal on T1-weighted images. The function resulting from dynamically sampling blood signal changes over time (SCT) describes the relaxivity effect of contrast transit. The IF is estimated from SCT, assuming a linear relationship between T1 change and CA concentration. This assumption is only roughly true. For high CA concentration, the T2* shortening effect of gadolinium produces a nonlinear CA dose response. Signal saturation occurs meaning that a further reduction in T1 does not result in a further increase in signal
Filice and Crisi: The Role of Input Function in DCE-MRI Metrics
127
Fig 4. Parameters obtained from a typical input function waveform.
Fig 5. Mean AIF and mean VIF. Error bar depict ± SD across the population of 20 patients.
and when the concentration becomes higher the signal starts to fall, since the T2* effects of the contrast agent dominate signal behavior. Signal saturation mainly results in an underestimation of IF first peak. The CA concentration threshold for signal saturation depends on TE. Kleppestø et al report,28 for TE of 2 ms, an error of 10% on Ktrans and Vp estimation at CA concentration of 10 mM and 7 mM, respectively. Blood in-flow effect manifests as an MR signal enhancement due to the flow of relaxed proton spin into the imaging volume and leading to an increase in signal intensity. In-flow effect is dependent on blood velocity and direction, slice location inside acquisition volume, and imaging parameters (TE, TR, and flip angle). Blood in-flow causes the overestimation of IF. To the opposite, given the large size of the superior sagittal sinus compared to the cerebral arteries, it certainly would benefit DCE-MRI to eliminate partial volume and motion artifacts by using the waveform sampled there. As known, the partial volume effects are manifested if vessel diameter is small compared with size of the voxel and 128
increase when lowering spatial resolution. In this case, a certain amount of static tissue is included in the voxel producing a lack of signal involving AIF attenuation or distortion.31 Our data showed a mean AIF attenuation due to partial volume effects about 8%. It is well known that the accuracy of DCE-MRI kinetic estimates can be affected by the choice of the temporal resolution or sampling time. At low sampling rate input function aliasing occurs. In a study by Larsson et al, the influence of sampling time was evaluated on the estimation of kinetic parameters in a HGG cohort of patients.30 The authors recommend using a temporal resolution lower than 20 seconds to minimize the uncertainty due to sampling error and ensure that tumor “hot spots” with rapid CA extravasation were not missed. In a systematic review, Heye et al analyzed 30 different DCEMRI studies concerning brain malignancy finding that the median temporal resolution adopted in the scanning protocols was 5.3 seconds.9 A 4.9 seconds sampling time as in the present study is a trade-off between the need for an appropriate
Journal of Neuroimaging Vol 26 No 1 January/February 2016
temporal resolution and control of spatial resolution, coverage and signal-to-noise ratio. Recently, Shin et al reported that DCE-MRI seems to be more powerful than DSC-MRI in differentiation of recurrences from treatment-related changes.32 They used a temporal resolution approximately of 5.5 seconds to calculate IF and Ktrans . In addition, their study was based upon commercially available software which can be easily adapted to routine practice. Our results indicate that under the specific circumstances of the described DCE-MRI acquisition technique, VIF and AIF are interchangeable in evaluating HGG. Similarly to our DCE-MRI technique and post-processing, Hamilton et al used a VIF method in order to differentiate HGG recurrence from treatment effect.18 They also concluded that IF derived from the sagittal sinus appeared to be superior to that from the middle cerebral artery. In conclusions, we have demonstrated that IF obtained by using arterial or venous waveform led to no statistical difference in DCE-MRI quantitative metrics of high-grade gliomas. However, additional research involving DCE-MRI acquisition protocols and post-processing would be beneficial to further substantiate the effectiveness of venous approach as the IF method compared with arterial-based IF measurement.
References 1. Jalali S, Chung C, Foltz W, et al. MRI biomarkers identify the differential response of glioblastoma multiforme to anti-angiogenic therapy. Neuro Oncol 2014;16:868-79. 2. Ahn SS, Shin NY, Chang JH, et al. Prediction of methylguanine methyltransferase promoter methylation in glioblastoma using dynamic contrast-enhanced magnetic resonance and diffusion tensor imaging. Neurosurgery 2014;121:367-73. 3. O’Connor JPB, Jackson A, Parker GJM, et al. DCE-MRI biomarkers in the clinical evaluation of antiangiogenic and vascular disrupting agents. Br J Cancer 2007;96:189-95. 4. Tofts PS, Brix G, Buckley DL, et al. Estimating kinetic parameters from dynamic contrast enhanced T1-w MRI of a diffusible tracer: standardized quantities and symbols. J Magn Reson Imaging 1999;10:223-32. 5. Toft PS, Oxon BA. Modeling tracer kinetics in dynamic Gd-DTPA MR imaging. JMRI 1997;7:91-101. 6. Sourbron S, Ingrisch M, Siefert A, et al. Quantification of cerebral blood flow, cerebral blood volume, and blood–brain-barrier leakage with DCE-MRI. Magn Reson Med 2009; 62:205-17. 7. Calamante F. Arterial input function in perfusion MRI: a comprehensive review. Prog Nucl Magn Reson Spectrosc 2013;74:1-32. 8. Bernstein JM, Homer JJ, West CM. Dynamic contrast-enhanced magnetic resonance imaging biomarkers in head and neck cancer: potential to guide treatment? A systematic review. Oral Oncol 2014;50:963-70. 9. Heye AK, Culling RD, Valdes MC, et al. Assessment ´ Hernandez ´ of blood-brain barrier disruption using dynamic contrast-enhanced MRI. A systematic review. Neuroimage Clin 2014;6:262-74. 10. O’Connor JP, Jackson A, Parker GJ, et al. Dynamic contrastenhanced MRI in clinical trials of antivascular therapies. Nat Rev Clin Oncol 2012;9:167-77. 11. Buckley DL. Uncertainty in the analysis of tracer kinetics using dynamic contrast-enhanced T1-weighted MRI. Magn Reson Med 2002;47:601-6. 12. Yang C, Karczmar GS, Medved M, et al. Estimating the arterial input function using two reference tissues in dynamic contrastenhanced MRI studies: fundamental concepts and simulations. Magn Reson Med 2004;52:1110-7. 13. Lavini C. Simulating the effect of input errors on the accuracy of Tofts pharmacokinetic model parameters. Magn Reson Imaging 2015;33:222-35.
14. Van Osch MJP, van der Grond J, Bakker CJG. Partial volume effects on arterial input functions: shape and amplitude distortions and their correction. J Magn Reson Imaging 2005;22:704-9. 15. Fluckiger JU, Schabel MC, DiBella EV. Toward local arterial input functions in dynamic contrast-enhanced MRI. J Magn Reson Imaging 2010;32:924-34. 16. Calamante F. Arterial input function in perfusion MRI: A comprehensive review. Prog Nucl Magn Reson Spectrosc 2013;74:132. 17. Yuan J, Chow SKK, Zhang Q, et al. The use of dynamic tracer concentration in veins for quantitative DCE-MRI kinetic analysis in head and neck. Plos One 2013;8:e59885. 18. Hamilton JD, Lin J, Ison C, et al. Dynamic contrast-enhanced perfusion processing for neuroradiologists: model-dependent analysis may not be necessary for determining recurrent high-grade glioma versus treatment effect. AJNR Am J Neuroradiol 2015;36: 686-93. 19. Parker GJ, Roberts C, Macdonald A, et al. Experimentallyderived functional form for a population-averaged high-temporalresolution arterial input function for dynamic contrast-enhanced MRI. Magn Reson Med 2006;56:993-1000. 20. Singh A, Haris M, Rathore D, et al. Quantification of physiological and hemodynamic indices using T(1) dynamic contrastenhanced MRI in intracranial mass lesions. J Magn Reson Imaging 2007;26:871-80. 21. Port RE, Knopp MV, Brix G. Dynamic contrast-enhanced MRI using Gd-DTPA: interindividual variability of the arterial input function and consequences for the assessment of kinetics in tumors. Magn ResonMed 2001;45:1030-8. 22. Lavini C, Verhoeff JJ. Reproducibility of the gadolinium concentration measurements and of the fitting parameters of the vascular input function in the superior sagittal sinus in a patient population. Magn Reson Imaging 2010;28:1420-30. 23. Larsson C1, Kleppestø M, Grothe I, et al. T1 in high-grade glioma and the influence of different measurement strategies on parameter estimations in DCE-MRI. J Magn Reson Imaging 2015;42: 97-104. 24. Zhang N, Zhang L, Qiu B, et al. Correlation of volume transfer coefficient Ktrans with histopathologic grades of gliomas. J Magn Reson Imaging 2012;36:355-63. 25. Nguyen TB, Cron GO, Mercier JF, et al. Diagnostic accuracy of dynamic contrast-enhanced MR imaging using a phase-derived vascular input function in the preoperative grading of gliomas. AJNR 2012;33:1539-45. 26. Eliat PA, Olivie´ D, Sa¨ıkali S, et al. Can dynamic contrast-enhanced magnetic resonance imaging combined with texture analysis differentiate malignant glioneuronal tumors from other glioblastoma? Neurol Res International 2012; 195176 doi: 10.1155/2012/195176. 27. Choi HS, Kim AH, Ahn SS, et al. Glioma grading capability: comparisons among parameters from dynamic contrast-enhanced MRI and ADC value on DWI. Korean J Radiol 2013;14:48792. 28. Kleppestø M, Larsson C, Groote I, et al. T2*-correction in dynamic contrast-enhanced MRI from double-echo acquisitions. J Magn Reson Imaging 2014;39:1314-9. 29. Roberts C, Little R, Watson Y, et al. The effect of blood inflow and B(1)-field inhomogeneity on measurement of the arterial input function in axial 3D spoiled gradient echo dynamic contrast-enhanced MRI. Magn Reson Med 2011;65:108-19. 30. Larsson C, Kleppestø M, Rasmussen I, et al. Sampling requirements in DCE-MRI based analysis of high grade gliomas: simulations and clinical results. J Magn Reson Imaging 2013;37:818-29. 31. Knutsson L, Borjesson S, Larsson EM, et al. Analysis of partial volume effects on arterial input functions using gradient echo: a simulation study. Magn Reson Med 2009;61:1300-9. 32. Shin KE, Ahn KJ, Choi HS, et al. DCE and DSC MR perfusion imaging in the differentiation of recurrent tumour from treatmentrelated changes in patients with glioma. Clin Radiol 2014;69:26472.
Filice and Crisi: The Role of Input Function in DCE-MRI Metrics
129
