Effects of arterial input function selection on kinetic parameters in brain dynamic contrast-enhanced MRI

Accepted Manuscript Effects of arterial input function selection on kinetic parameters in brain dynamic contrast-enhanced MRI

Vera C. Keil, Burkhard Mädler, Jürgen Gieseke, Rolf Fimmers, Elke Hattingen, Hans H. Schild, Dariusch R. Hadizadeh PII: DOI: Reference:

S0730-725X(17)30076-0 doi: 10.1016/j.mri.2017.04.006 MRI 8757

To appear in: Received date: Revised date: Accepted date:

28 December 2016 20 March 2017 20 April 2017

Please cite this article as: Vera C. Keil, Burkhard Mädler, Jürgen Gieseke, Rolf Fimmers, Elke Hattingen, Hans H. Schild, Dariusch R. Hadizadeh , Effects of arterial input function selection on kinetic parameters in brain dynamic contrast-enhanced MRI. The address for the corresponding author was captured as affiliation for all authors. Please check if appropriate. Mri(2017), doi: 10.1016/j.mri.2017.04.006

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Effects of arterial input function selection on kinetic parameters in

PT

brain dynamic contrast-enhanced MRI

NU

SC

REVISED VERSION

RI

- Original article –

Authors

Email

a

[email protected]

Dipl.-Phys. Jürgen Gieseke

b

[email protected]

MA

Dr. rer. nat. Burkhard Mädler, PhD a,b

Dr. rer. nat. Rolf Fimmers, PhD

[email protected]

c

[email protected]

Prof. Dr. med. Elke Hattingen, MD

a

Prof. Dr. med. Hans H. Schild, MD

a

[email protected]

D

Dr. med. Vera C. Keil, MD

[email protected]

CE

a

a

PT E

PD Dr. med. Dariusch R. Hadizadeh, MD

[email protected]

Department of Radiology, University Hospital Bonn, Sigmund-Freud-Strasse 25, 53127 Bonn,

Germany

Philips Healthcare, Röntgenstrasse 22, 22335 Hamburg, Germany

c

IMBIE (Statistics Department), University Hospital Bonn, Sigmund-Freud-Strasse 25, 53127 Bonn,

Germany

AC

b

Corresponding Author:

PD Dr. med. D.R. Hadizadeh Radiologische Klinik (FE MRT) Sigmund-Freud-Straße 25 53127 Bonn Tel: +49 228 287 19639 Fax: +49 228 287 15598 Email: [email protected]

1

ACCEPTED MANUSCRIPT Abstract Purpose: Kinetic parameters derived from dynamic contrast-enhanced MRI (DCE-MRI) were suggested as a possible instrument for multi-parametric lesion characterization, but have not found their way into clinical practice yet due to inconsistent results. The quantification is heavily influenced by the definition of an appropriate arterial input functions (AIF). Regarding

PT

brain tumor DCE-MRI, there are currently several co-existing methods to determine the AIF frequently including different brain vessels as sources. This study quantitatively and

RI

qualitatively analyzes the impact of AIF source selection on kinetic parameters derived from

SC

commonly selected AIF source vessels compared to a population-based AIF model. Material and Methods: 74 patients with brain lesions underwent 3D DCE-MRI. Kinetic

NU

parameters [transfer constants of contrast agent efflux and reflux Ktrans and kep, their ratio, ve, to measure extravascular-extracellular volume fraction and plasma volume fraction vp] were

MA

determined using extended Tofts model in 821 ROI from 4 AIF sources [the internal carotid artery (ICA), the closest artery to the lesion, the superior sagittal sinus (SSS), the population-

D

based Parker model]. The effect of AIF source alteration on kinetic parameters was

PT E

evaluated by tissue type selective intra-class correlation (ICC) and capacity to differentiate gliomas by WHO grade [area under the curve analysis (AUC)]. Results: Arterial AIF more often led to implausible ve >100% values (p<0.0001). AIF source

CE

alteration rendered different absolute kinetic parameters (p<0.0001), except for kep. ICC

AC

between kinetic parameters of different AIF sources and tissues were variable (0.08 – 0.87) and only consistent >0.5 between arterial AIF derived kinetic parameters. Differentiation between WHO III and II glioma was exclusively possible with vp derived from an AIF in the SSS (p=0.03; AUC 0.74). Conclusion: The AIF source has a significant impact on absolute kinetic parameters in DCE-MRI, which limits the comparability of kinetic parameters derived from different AIF sources. The effect is also tissue-dependent. The SSS appears to be the best choice for AIF source vessel selection in brain tumor DCE-MRI as it exclusively allowed for WHO grade II/III and III/IV glioma distinction (by vp) and showed the least number of implausible ve values.

2

ACCEPTED MANUSCRIPT

Keywords: arterial input function, DCE-MRI, vessel selection, kinetic parameters, Ktrans,

AC

CE

PT E

D

MA

NU

SC

RI

PT

glioma differentiation

3

ACCEPTED MANUSCRIPT 1. Introduction

T1-weighted dynamic contrast-enhanced MRI (DCE-MRI) is a quantitative imaging technique created to evaluate both tissue perfusion and blood-brain barrier disruption in the CNS [1]. The majority of clinically available DCE-MRI post-processing tools apply the extended Tofts

PT

model, which delivers the following kinetic parameters: contrast volume transfer constants Ktrans (efflux of contrast agent from vessel) and k ep (reflux to vessel), their ratio ve, as an

SC

RI

estimate of the extravascular-extracellular space, and the plasma volume fraction vp [2].

Although multiple studies rendered promising results concerning the non-invasive

NU

discrimination of intracranial lesions or treatment monitoring based on DCE-MRI-derived kinetic parameters, the technique is not commonly applied in routine clinical practice yet [3-

MA

11]. The complex kinetic modeling behind DCE-MRI, which allows for numerous variations in both the acquisition of MRI data as well as its post-processing in order to generate kinetic

PT E

acquisition of DCE-MRI [12-15].

D

parameters, can be identified as a pivotal reason why clinicians opt against the routine

A particularly vulnerable component and thus potential source of error during post-

CE

processing of DCE-MRI is the arterial input function (AIF), which defines the intravascular

AC

plasmatic concentration of a contrast agent over time [Cp(t)] and is used in kinetic modeling to estimate kinetic parameters in target tissue. Factors, that influence the measurement of AIF, include the sequence type, partial volume effects, the flip angle, flow compensation or the method of conversion from MR signal to concentration of contrast agent [16-20].

The AIF can be determined individually or based on population-based flow-adapted fixed models such as those presented by Weinmann or Parker [21,22]. Parker’s model was originally developed to avoid individual AIF definition in abdominal MRI [22]. In the brain, on the other hand, individual methods for determining the AIF demand manual positioning of

4

ACCEPTED MANUSCRIPT regions of interest (ROI) into an artery, the perivascular space, a reference target tissue or the superior sagittal sinus [23-27]. Recently, also fully automatized AIF selection algorithms were introduced in some centers in order to reduce miscalculations of kinetic parameters due to suboptimal ROI placements [28-31]. However, most of the currently clinically implemented post-processing tools rely on manual or computer-assisted positioning of the ROI for AIF

PT

determination in a single source vessel of choice. In recent years, the manually selected ROI for AIF determination was increasingly often placed in the superior sagittal sinus for brain

RI

tumor DCE-MRI, possibly in order to reduce partial volume effects that are noticed in smaller

SC

and tortuous arterial vessels. Effects of positioning the AIF ROI in different vessels and comparing thereby obtained kinetic parameters to those derived from a model-based

NU

approach have not yet been systematically analyzed and only a few and mainly studies regarding the prostate discussed consequences of altering the AIF source on kinetic

MA

parameters in DCE-MRI [22,26,32-35].

D

The purpose of this study was therefore to 1) assess AIF curves of different brain vessels

PT E

regarding peak contrast-agent concentration and plausibility of deduced kinetic parameters, 2) elucidate the putative qualitative and quantitative influence of the AIF source on kinetic parameters with regard to differentiability of gliomas and 3) identify the AIF source vessel,

AC

2).

CE

which delivers the most reliable kinetic data for clinical purposes based on results in 1) and

5

ACCEPTED MANUSCRIPT 2. Material and methods

2.1 Patient Cohort This study was performed after permission of the hospital ethics committee.

Informed

consent was obtained from all individual participants. DCE-MRI data of 74 patients with

PT

various intracranial lesions were analyzed (38 men, 36 women; table 1). The age range was

RI

24 – 83 years (median 58 years).

SC

2.2 Imaging and injection protocol

All patients received a whole-brain MRI at 3.0 T (Achieva TX, Philips Healthcare, Best, The

NU

Netherlands; 70 cm bore) with an 8-channel SENSE head coil. Imaging included a transverse whole-brain 3D DCE-sequence that consisted of two pre-contrast T1W dual flip

MA

angle 3D-ultrafast gradient echo (TFE) series with selective slab excitation and RF gradient spoiling for pre-contrast T1 measurement (flip angles 5°/15°), sometimes referred to as

D

reference scans, followed by a dynamic series of 50 individual scans [technical parameters:

PT E

TR 10 ms; TE 2.3 ms; flip angle 8° dynamic scan; FOV 108x220x182 mm (FHxAPxRL); acquired voxel size 1.57x1.6x3.0 mm; 36 slices; scan time 66 s for both reference scans and 5.3 min. for the dynamic sequence]. Pre-contrast T1W gradient echo series with at least two

CE

different flip angles are necessary for the assessment of T1 calibration in order to later be

AC

able to apply these corrections for kinetic parameter calculation according to the extended Tofts model [36,37].

A standard dose of gadobutrol (0.1 mmol/kg body weight; Bayer Healthcare, Leverkusen, Germany) followed by a 24 ml saline flush was automatically injected via a 18G canula placed in the right antecubital vein (Medrad Spectris Solaris EP Injection System, Bayer Healthcare, Leverkusen, Germany). Injection started during the second dynamic scan at a flow rate of 3 ml/s. The bolus typically reached the internal carotid artery (ICA) during the fourth dynamic scan and the confluent sinus one dynamic phase later (5 s).

6

ACCEPTED MANUSCRIPT

2.3 Post-processing All image processing was performed offline with a dedicated software tool (Permeability Tool, Intellispace Portal 5.0; Philips, Best, The Netherlands). Transformation of MR signal intensity–time curves into contrast agent concentration-time curves [Cp(t) for plasma

PT

concentration of the AIF and Ct(t) for tissue concentration in a tissue ROI] and further deduction of DCE-MRI kinetic parameters were based on the linear least-squares fitting

RI

method introduced and described in detail by Murase [38]. For all calculations, contrast

SC

relaxivity in plasma at 3.0T was assumed to be 5.0 L/(mmol*s) for gadobutrol at 37°C [18]. Patient hematocrit levels, which are known to influence contrast relaxivity, were determined

MA

based on the extended Tofts model [2].

NU

on the day of the MRI. DCE-MRI kinetic parameters Ktrans, kep, ve and vp were assessed

2.4 AIF source definition

D

For determination of the AIF curve, a 7x7 voxel grid was positioned on transverse slices of

PT E

the target vessel on the contrast-enhanced dynamic MR images. In the literature, three vessels have been predominantly used as source vessels for AIF definition: (1) the superior sagittal sinus (SSS), (2) the terminal ICA and (3) the arterial vessel closest to the lesion (CA;

CE

fig. 1) [23,26,39-42]. In the case of arterial source vessels, the grid was placed ipsilateral to

AC

the lesion. Vessel segments with apparent inflow artifacts, as defined by hyperintensity on the initial four non-enhanced T1W dynamic scans, were not accepted for AIF grid placement, since inflow artifacts are known to influence AIF measurements [43].

The ROI, which was positioned on the target source vessel and was used for AIF determination, consisted of a 7x7 grid, such that 49 contrast concentration-time curves were generated. Each of the 49 contrast concentration-time curves represented the concentration time curve of one voxel within the ROI that included a cross-section of the target source vessel. Curves with a steep ascent and early peak of contrast agent bolus followed by steady

7

ACCEPTED MANUSCRIPT decline to near baseline were considered as optimal vascular Cp(t) curves. In contrast, multipeak or negative-peak curves were considered inappropriate and were not used. Based on the shape and amplitude of the contrast concentration-time curves, the four best eligible curves were selected for calculation of the final AIF, as AIF determination from only one grid voxel can be more prone to false measurements. The final AIF curve was determined by

PT

calculation of the mean from all selected curves on the grid. If more than four voxels showed eligible grid curves, those four with the highest amplitude were selected. Vessels with one to

RI

three eligible curves were considered as vulnerable for partial volume contamination of the

SC

AIF measurement, but were still accepted for data analysis. AIF curve definition by the Parker model (PM), on the other hand, used a population-based approach as described

NU

elsewhere [22]. AIF determination and calculation of kinetic parameters were performed for each of the three vessel-based and the PM-based calculation using injection durations of 10

MA

s, which was the calculated injection duration in this experimental setting.

D

2.5 AIF curve analyses

PT E

2.5.1 Qualitative analyses of AIF curves

Three criteria were defined to analyze the quality of AIF curves: 1. The first criterion was, whether at least four AIF-determining voxels of the 7x7 grid

CE

delivered eligible Cp(t) curves . In case of only one to three curves the patient was not

AC

excluded, but the resulting averaged final AIF curve was considered less reliable. 2. The second criterion was, whether the peak tissue concentrations of the contrast agent in the tissue ROIs (tROIs) [peak Ct(t)] were lower than the peak plasma concentration of the final AIF [Cp(t)]. When ROI-based Ct(t) was higher than Cp(t) in the final AIF, the obtained number was considered over-estimated, but neither the AIF nor the tROI were discarded under the assumption that both AIF-determining ROI and tROI positioning could not be optimized for this vessel/tissue. 3. The third criterion was, whether ve and vp measured in the tissue ROI (tROI; compare 2.6.) were below 100%.

8

ACCEPTED MANUSCRIPT

Qualitative constraints regarding the latter two points were no reason to discard the patient data, but were considered as an inherent shortcoming of the AIF-defining vessel and possible explanation regarding results of pharmacokinetic parameter calculations.

PT

2.5.2 Quantitative analysis of AIF curves Peak Cp(t) (measured in mM) of the final mean AIF curve (n=74) of each individual was

RI

compared for significant differences depending on the AIF source (ICA, CA, SSS) in order to

SC

be later able to discuss the impact of the AIF on resulting kinetic parameters. Intraclass correlation analysis was chosen to assess the comparability of the AIF curves based on the

NU

peak Cp(t) of all AIF sources (e.g. peak Cp(t),SSS vs. peak Cp(t),ICA; six possible combinations).

MA

2.6 Quantitative analyses of kinetic parameters

Based on the final mean AIF curves, tROI kinetic parameters Ktrans, kep, ve and vp were

D

calculated. In order to address possible slice- and ROI-dependent artifacts, up to 12 tROIs

PT E

were manually drawn in brain tissue of 74 patients and kinetic parameters were assessed for each of the (in total) 821 tROIs. Kinetic parameters were calculated based each of the four AIF sources (ICA, CA, SSS, PM). Each tROI was measured once per AIF source. AIF source

CE

dependent effects on absolute kinetic parameter values were assessed separately for each

AC

tissue subgroup (n=15; table 1). In order to assess whether a change of AIF source has a qualitative effect on kinetic parameters an intraclass correlation of the same kinetic parameter measured in the same tROI, but derived from different AIF sources, was performed.

2.7 Clinical impact of AIF curves In order to test whether AIF-dependent variability of kinetic parameters had a diagnostic consequence, and in order to define a criterion, from which a superiority of an AIF source may be concluded, all kinetic parameters of all AIF sources were tested with regard to their

9

ACCEPTED MANUSCRIPT capacity to differentiate gliomas by WHO grade. This is a classic application for DCE-MRI kinetic parameters [10,42]. tROIs were placed in three adjacent slices of contrast-enhancing or else in T2-hyperintense regions of the tumor. The tROI-derived kinetic parameters were consecutively averaged per patient. These averages served as the basis for parameter based glioma WHO grade differentiation. Differentiations were assessed between WHO

PT

grade IV and III as well as III and II respectively.

RI

2.8 Statistics

SC

All statistics were performed with SAS 9.4 (SAS Institute Inc., Cary, NC, USA) and SPSS 22.0 (IBM Corp., Armonk/NY, USA). All data are presented in medians with confidence

NU

intervals (CI) at a 95% level. p-values were considered significant at p<0.05.

MA

Qualitative analyses of AIF curves were performed with non-parametric McNemar tests. Quantitative analyses of AIF curves involved non-parametric Wilcoxon rank-sum tests to

D

compare peak Cp(t) values calculated with each of the AIF curves of each patient. An

PT E

additional intraclass correlation analysis determined whether quantitative differences of peak Cp(t) indicate a limited comparability of AIF sources.

CE

The influence of a variation in AIF source on kinetic parameters was examined both

AC

quantitatively and qualitatively. Quantitative analyses of kinetic parameters to assess AIF selective effects involved non-parametric Wilcoxon rank-sum tests. Data is presented stratified into five lesion types with similar appearance on MRI (table 1): (1) contrast-enhancing intra-axial lesions (blood-brain barrier disruption; IACEL), (2) contrast-enhancing extra-axial lesions (excess enhancement in primarily enhancing tissues; EACEL), (3) lesions without contrast enhancement but signal elevation on T2w images (no disruption or non-visible disruption of the blood-brain barrier on CET1W images; NEL),

10

ACCEPTED MANUSCRIPT (4) presumed central necrosis in high-grade glioma and metastases (highly fluid, but near avascular regions), and (5) normal-appearing white matter (reference tissue; NAWM).

The qualitative effect of AIF source variation on kinetic parameters was examined by

PT

intraclass correlation (ICC) of the same tROI-based pharmacokinetic parameter derived from different AIF sources. The regression equation derived from a Deming-regression for a pair

RI

of values was therefore used to rescale one of the two values before calculating the ICC due

SC

to large differences between the absolute kinetic parameter values. Analyses are presented for all 12 tissues present in this study. The ICC coefficient of all 821 tROI together was

NU

applied to assess the goodness of fit for this cohort.

MA

A Pearson correlation between kinetic parameters Ktrans and kep (derived from identical tROI) was applied to assess whether a change in the AIF source alters both parameters

D

proportionally. Both parameters are independent from each other, but their correlation

PT E

coefficient R can be hypothesized to be constant within one patient and measurement. A change in R can thus be considered as a qualitative effect of the choice of AIF source.

AC

CE

Glioma differentiation by WHO grade involved ROC curve and AUC analyses.

11

ACCEPTED MANUSCRIPT 3. Results

3.1 Qualitative analysis of AIF curves 3.1.1 AIF curve definition Selection of all 4 voxels on the 7x7 grid showing eligibly shaped Cp(t) curves in order to generate an adequate AIF curve from their arithmetic mean was less frequently possible

PT

when positioning the ROI grid for AIF determination in an arterial vessel than in the SSS

RI

(SSS: n=6/74, ICA: n=11/74, CA: n=13/74; ICA vs. CA: p=0.80; SSS vs. ICA or CA: p=0.30 and 0.12 respectively). Despite an identical definition of the shape of the curve needed to be

SC

eligible for AIF curve definition, it was noted that even adequately shaped arterial Cp(t) curves showed a more gradual return to baseline after the first peak decline than the curves

NU

defined in the SSS. The later returned closer to baseline after the sharp bolus arrival peak.

MA

Resulting final mean Cp(t) AIF curves defined in the SSS showed a closer resemblance to the population-based PM curves regarding both shape and amplitude (compare fig. 1). This

D

similarity was also observed between arterial vessel derived Cp(t) AIF curves.

PT E

3.1.2 Implausible peak contrast agent concentrations and kinetic parameters The implausible and by definition illogical observation of peak Ct(t) of a tissue ROI being

CE

higher than the peak Cp(t) of the final mean AIF was significantly more often observed when selecting an arterial vessel for AIF definition than in SSS-derived AIF curves (SSS: 11/74,

AC

ICA: n=44/74, CA: n=42/74; ICA vs. CA: p=0.69; SSS vs. ICA or CA: both p<0.0001). Closely related to this finding is the registration of implausible ve or vp values (above 100%) measured in at least one tROI of a patient. ve values >100% were significantly more often observed after arterial than in venous or model-based AIF selections (SSS=3/74, ICA: n=49/74, CA: 44/74, Parker: 3/74; ICA vs. CA or PM vs. SSS: p=0.23 and 0.82 respectively; SSS or PM vs. ICA or CA: all p<0.0001). Implausibly high vp values (>100%) were extremely rare (7/821 tROI; 0.85%) and exclusively found when using arterial sources for AIF curve definition had been used (SSS: 0/74, ICA: 3/74, CA: 4/74, PM: 0/74 patients).

12

ACCEPTED MANUSCRIPT Notably, no significant differences regarding the qualitative analyses could be observed when comparing between the two arterial sources or the SSS and PM as AIF sources only.

3.2 Quantitative analysis of AIF curves Peak Cp(t) of AIF curves of 74 patients varied significantly depending on positioning of the

PT

ROI for AIF determination [range, median, 95% confidence intervals; all in mM]: SSS: 0.08 – 2.30, 0.55, 0.50 – 0.76;

RI

ICA: 0.01 – 0.65, 0.11, 0.10 – 0.14;

SC

CA: 0.01 – 0.70, 0.12, 0.11 – 0.15.

Wilcoxon rank-sum tests revealed significant (all p<0.0001) differences between the medians

NU

of peak Cp(t) of AIF curves derived from ROI in the SSS compared to those defined in the ICA or CA, but not between ICA and CA (p=0.20). The ICC goodness of fit analysis for peak

MA

Cp(t) measurements of all three individual AIF definitions confirmed a moderate to good correlation between both arterial AIF [ICC 0.50 (CI 0.31 – 0.65)], but also revealed a

D

substantial variability with marked outliers between both sources (fig. 2). The ICC of arterial

PT E

peak Cp(t) with SSS defined peak Cp(t) was very low [0.14 (CI -0.09 – 0.36) and 0.01 (CI 0.21 – 0.24) for ICA and CA respectively].

CE

3.3 AIF-source dependency of kinetic parameters

AC

3.3.1 Quantitative effects of a change of AIF source Absolute kinetic parameter values (Ktrans, kep, ve, vp) derived from different AIF sources (ICA, CA, SSS, PM) differed at a scale of sometimes more than factor 10. Table 2 provides an overview for five tissue groups separated into groups their different appearance on T2W and CET1W MRI

(tissues 1, 5, 7, 8 and 10/11, table 2). Arterial AIF rendered higher kinetic

parameter values. kep stands out as this parameter was in the same range for all four AIF sources. A change in positioning of the ROI for AIF curve definition did not alter kinetic parameters consistently: Depending on the source of AIF curve calculation and the tissue of choice, Ktrans differed significantly. When regarding contrast-enhancing tissue e.g. median

13

ACCEPTED MANUSCRIPT Ktrans about times higher in an ICA-based compared to an SSS-based AIF estimation of the parameter [575.7 (470.12 – 723.65) and 103.10 (70.75 – 123.37) respectively; in 10-3/min], but nearly ten times higher for the same variation in source of AIF when regarding NAWM [3.36 (2.39 – 6.43) and 0.38 (0.07 – 0.78) respectively; in 10-3/min].

PT

3.3.2 Qualitative aspects of a change of AIF source The detailed ICC analysis for all kinetic parameters and AIF sources with further stratification

RI

into all 12 histologically-based tissue types as defined in (table 1) reveals that the AIF source

SC

not only changes the absolute value of the resulting kinetic parameter, but indeed render kinetic parameters, which are in fact difficult to compare to each other (digital supplement for

NU

all tissues). From the tissue-stratified ICC analysis three main aspects appear as key findings. First, the ICC of a pair of AIF sources for the same kinetic parameter (e.g. K trans,SSS

MA

and Ktrans,PM) is highly variable between different tissues. Second, the ICC can be lowered by the fact that several tissues show tROI subgroups with each a higher ICC (e.g. digital

D

supplement: Ktrans,CA vs. Ktrans,ICA tissue 10), but it can also be falsely increased by outliers

PT E

(compare e.g. tissue 1 of the same pair). Third, consistently higher ICC (>0.5) could only be found in pairings of the two arterial AIF sources ICA and CA. This last aspect is corroborated by the goodness of fit analysis applying all tROI (n=821) showing ICC between 0.52 (ve) and

CE

0.87 (vp) for arterial AIF sources (table 3).

The variability of the effect of a change in AIF

AC

source depending on which kinetic parameter is involved was also demonstrated by variable Pearson correlation coefficients R for Ktrans and kep depending on the AIF source (SSS: 0.66; ICA: 0.09; CA: 0.45; PM: 0.26), which indicated that ve calculation is affected by the AIF source as well.

3.3.3 Clinical impact of kinetic parameters with regard to differentiation of glioma grades The clinical consequences of an alteration of AIF sources was evaluated by the differentiability of gliomas of different WHO grades based on kinetic parameters Ktrans, kep, ve

14

ACCEPTED MANUSCRIPT and vp. These were determined based on all four AIF sources measured in the same tROI (tissues 2, 3 and 4 of table 1).

WHO grade III (n=70 tROI) and grade IV gliomas (n=85 tROI) could be reliably differentiated in all cases based on Ktrans, ve and vp (all p=0.001). The areas under the curve (AUC) for a

PT

Ktrans based differentiation of WHO grades IV and III were 0.98, 0.92, 0.98 and 0.96 for PM, SSS, ICA and CA AIF sources respectively. For ve the AUC were 0.92 (PM), 0.93 (SSS),

RI

0.90 (ICA) and 0.95 (CA) and for vp 0.99 (PM), 0.89 (SSS), 0.91 (ICA) and 0.90 (CA) in the

SC

WHO IV vs. WHO III differentiation group. Only the PM approach allowed a kep based differentiation of WHO IV and III gliomas [AUC 0.24 (PM), 0.63 (SSS), 0.65 (ICA) and 0.63

NU

(CA)] with (discordantly to the other AIF sources) significantly the exceptional constellation of higher kep in WHO III than in grade IV lesions (p=0.001; median kep,PM 425*10-3/min vs.

MA

145.32*10-3/min).

D

A differentiation between tumor grades II (n=20 tROI) and grade III was exclusively possible

PT E

based on vp when the AIF curve source was defined in the SSS (p=0.03; median vp,SSS WHO III 0.47% vs. WHO II 0.19%). The respective AUC were 0.74 (SSS), 0.53 (PM), 0.38 (ICA) and 0.31 (CA). The AUC for a Ktrans based differentiation were 0.53 (MB), 0.38 (SSS), 0.47

CE

(ICA) and 0.35 (CA) and regarding a kep based differentiation of WHO III and II gliomas the

AC

AUC were 0.50 (MB), 0.42 (SSS), 0.42 (ICA) and 0.45 (CA). The AUC for a ve based differentiation were 0.44 (PM), 0.33 (SSS), 0.37 (ICA) and 0.34 (CA).

15

ACCEPTED MANUSCRIPT 4. Discussion

This study demonstrates that the choice of the source for AIF curve definition modifies kinetic parameters on an intra- and inter-individual basis and shows tissue-dependent differences. The non-individualized Parker model rendered kinetic parameters, which resembled more closely those derived from the SSS than those derived from arterial source vessels in

PT

absolute values, but were at the level of intraclass correlation only moderately correlated to

RI

their SSS AIF derived counterparts. The two arterial AIF sources were correlated stronger on the kinetic parameter intraclass comparison level and in absolute values. Qualitative and

SC

quantitative data analysis revealed advantages of selecting the SSS over arterial vessels as the source for AIF calculation instead of an arterial AIF. Cp(t) curve shapes showed higher

NU

peak Cp, significantly fewer cases of implausible ve and vp>100% and lower absolute Ktrans

MA

and ve values when selecting the SSS as the AIF source vessel.

Tofts et al. suggested to prefer a manual arterial AIF close to the target lesion to a venous or

D

model-based AIF [44]. They argued that a spatio-temporally closer proximity to the target

PT E

tissue should render more accurate results for kinetic parameters. However, the small size of the arteries in question makes them susceptible to underestimating Cp(t) due to partial

CE

volume effects, which in consequence results in overestimation of kinetic parameters [17,26,45]. In this study, arteries serving for determination the AIF curve frequently did not

AC

fulfill the pre-defined qualitative criterion of four eligible Cp(t) curves on the selection grid necessary for determination of the final averaged AIF (ICA, CA; 11/74, 13/74 patients), most likely due to small vessel size and tortuosity. Arterial AIF eventually showed markedly lower peak Cp values than PM or SSS AIF. The low Cp values, in turn, lead to over-estimation of kinetic parameters. Possible reasons for this over-estimation are partial volume effects and inflow effects as these were observed in adjacent slices to those that served for AIF determination. The exact estimation of the role of artifacts remains yet unknown, as artifacts could not be quantitatively assessed, e.g. inflow artifacts could only be evaluated by intravascular hyperintensity on non-enhanced T1W images.

16

ACCEPTED MANUSCRIPT However, it appears likely to assume that PM and SSS derived AIF suffer less frequently from the above mentioned artifacts, which may explain similarities in curve shape and absolute values for peak Cp(t) and kinetic parameters between these AIF sources. The goodness of fit ICC analysis between all AIF sources and their kinetic parameters not only revealed an anticipated comparability between both arterial AIF source approaches, but the

PT

alsouncovered a limited comparability between kinetic parameters derived from PM AIF and the SSS (Ktrans ICC=0.51). This divergence maybe explained from the fact that the AIF of the

RI

Parker model was derived from measurements in the abdominal aorta with a fixed hematocrit

SC

of 42%, which may make explain differences in bolus shape and signal intensity from our cerebral venous, hematocrit-sensitive cohort [22]. Population-based AIF models may still

NU

serve as a first step to avoid selection bias of manual AIF definition, which again will possibly be replaced in the near future by fully computerized yet individual selection of ideal voxels or

MA

clusters to determine the optimal AIF curve [28-30]. It remains beyond the scope of this work to assess in how far e.g. kinetic parameters from reference tissue based AIF models are

D

comparable to those derived from classic manually selected vascular AIF models, but based

PT E

on the fact that already arterial and venous AIF sources resulted in only weakly related kinetic parameters in this study (e.g. Ktrans ICC 0.09 between CA and SSS), it may not be too

CE

extrapolated to conclude that these new methods will differ even more.

AC

This study also showed that Ktrans was more affected by source vessel selection than kep, which surprisingly remained fairly stable between all four AIF sources and is the reason why implausible ve values above 100% were created more often in strongly contrast-enhancing tissues when arteries were chosen as source vessels. However, from the detailed ICC analyses it becomes clear that even kep from different AIF sources are not comparable to each other. This stresses a key finding of this study: the AIF source determines both quantitatively and more important qualitatively the resulting kinetic parameters, which in consequence are not truly comparable between different AIF sources. The fitting model used for signal to contrast agent concentration estimation (a linear least squares approach in this

17

ACCEPTED MANUSCRIPT study), must be mentioned as another important co-factor, whose influence however cannot be addressed as part of this publication.

The ICC between kinetic parameters of different AIF sources differed variably depending on the tissue under investigation in this study (digital supplement). These effects may possibly

PT

be explained by a tissue-dependent applicability of the extended Tofts model and also patient-dependent effects. In glioma tissue significant effects of the model selection on

RI

kinetic parameters were observed in earlier studies [46]. The model was designed to

SC

simulate contrast agent dynamics in high flow lesions with a non-negligible vp component and may not be readily be applied for assessment of healthy brain tissue, which contains a very

NU

limited plasmatic volume [2]. Notably, the ICC in contrast-enhancing brain tumor tissue (table 2, tissue 4) showed at least moderate ICC without overestimation from outlier values for all

MA

AIF sources (as in NAWM), which can be interpreted as a support of this theory.

D

In order to illuminate that the choice of AIF source limits the comparability of kinetic

PT E

parameters in brain DCE-MRI a classical example for a kinetic parameter application was chosen: WHO grade differentiation of gliomas. In this cohort, only vp derived from an SSSbased AIF curve was able to provide the clinically important differentiation between WHO II

CE

and III gliomas despite the fact that grade III and II gliomas were both in the majority

AC

consisting of non-enhancing cases in this cohort. This special cohort composition may also explain low values for e.g. ve, which are reported higher by other authors [7,8,42]. Therefore, absolute kinetic parameters from DCE-MRI have to be interpreted with care as many influential factors such as detailed histology may contribute to them.

Taken together, qualitative, quantitative, and clinical advantages are in favor of the SSS over arterial vessels and the Parker model as the most suitable source vessel for AIF determination. More importantly, this study illuminated that clinical application studies for

18

ACCEPTED MANUSCRIPT DCE-MRI suffer from a limited comparability due to different AIF source choices and that the outcome especially of clinical application studies of DCE-MRI depends on the AIF choice.

This study has several limitations. It is strictly confined to an evaluation of the consequences of source vessel choice on kinetic parameters, and may not answer the question how exactly

PT

the measured Cp(t) curves are affected by the use of other target vessels. For instance, the MR and injection protocols influence the measured shape of the AIFs as the chosen flip

RI

angles does and the dynamic temporal resolution, contrast agent and flow velocity are also

SC

known to influence AIF and thus kinetic parameters [47,48]. Second, ROI sources for AIF curves were semi-manually selected after automated grid-based Cp(t) curve visualization,

NU

which may allow for a selection bias. However, we selected ROIs for AIF curve calculation according to our above mentioned quality criteria and thereby endeavored to reduce this

MA

bias. Third, it is known from DCE-MRI of renal vessels that ROI size of the AIF influences kinetic parameters [49]. However, grid sizes and ROI sizes were kept constant during this

D

study to be able to compare the curves obtained by different source vessels under the most

PT E

appropriate circumstances. Further, the total number of tROIs that served for kinetic parameter calculation, varied from one tissue to the other. The degree of tissue-dependent variability of kinetic parameters must therefore be interpreted with care and was implemented

CE

in this study by the intraclass method chosen to balance this issue.

AC

Finally, it remains unclear how the post-capillary (and thus flow efferent from target tissue) location of a venous AIF may influence kinetic parameters as the contrast agent may be more dispersed in venous than in local arterial vessels [17]: Lower blood flow velocity may in this case result in in- or outflow effects that can alter the measurement of the kinetic parameters. In this study, peak Cp in the SSS differed significantly from Cp from arterial sources and consecutively the resulting kinetic parameters from AIF sources were not highly correlated, which indicates profound differences between AIF from arterial and venous sources.

19

ACCEPTED MANUSCRIPT 5. Conclusion This study shows that the choice of the source vessel for AIF determination has a significant impact on absolute kinetic parameters in brain DCE-MRI, which in consequence lose comparability and show a variable utility for clinical applications. Out of the tested sources for AIF determination (ICA, CA, SSS, Parker model), the SSS turned out to be favorable due to

PT

a higher consistency and plausibility of results: the SSS source was the only one that allowed

AC

CE

PT E

D

MA

NU

SC

RI

to distinguish low-grade gliomas from higher grade lesions by the kinetic parameter vp.

20

ACCEPTED MANUSCRIPT FIGURE LEGENDS

Fig. 1 Overview on AIF sources, curves and resulting Ktrans maps in a 49 y/o patient with glioblastoma

PT

A: Parker model AIF for intermediate bolus injection length of 10s; B: superior sagittal sinus (SSS) AIF curve; C: Internal carotid artery (ICA) AIF curve; D: distal branch of middle carotid

RI

artery as the closest artery to the tumor and respective AIF. N.B. that the Ktrans maps have the

SC

same arbitrary unit color scaling. It becomes clear that arterial AIF result in far higher Ktrans

NU

values than PM or SSS AIF.

MA

Fig. 2 Intraclass correlation of contrast agent concentration-time AIF curve peaks

Intraclass correlation plot for peak Cp(t) of individual AIF curves for all three target vessels as a

D

goodness of fit measurement (n=74 patients). Correlation of arterial peak Cp(t) is substantial

PT E

(ICC=0.5), while both arterial AIF sources do not correlate with Cp(t) measured in the SSS (ICC 0.01 and 0.14). Cp(t): plasma contrast agent concentration over time; AIF: arterial input function;

CE

ICA: internal carotid artery; CA: closest artery; SSS: superior sagittal sinus.

AC

Digital supplement: Intraclass correlation (ICC) analysis for all histological tissue subgroups and kinetic parameters differentiated by their respective AIF source The values of the kinetic parameter and their ranges were quite different for the AIF sources. The regression equation derived from a Deming-regression for a pair of values was therefore used to rescale the value plotted on the vertical axis for the plot and the calculation of the ICC. PM: population-based Parker model; ICA: internal carotid artery; CA: closest artery; SSS: superior sagittal sinus. Confidence interval at 95% level in brackets. Tissue 14: ICC missing for kep and ve for statistical reasons. For tissue legend compare page 2 of supplement.

21

ACCEPTED MANUSCRIPT TABLE TITLES (compare table file for legends)

Table 1 Overview of 821 tissue ROI used for AIF analyses in 74 patients

Table 2 AIF-dependent variability of kinetic parameters from 821 tissue ROI and four AIF sources

Table 3 Goodness of fit intraclass correlation of kinetic parameters from different AIF

AC

CE

PT E

D

MA

NU

SC

RI

PT

sources

22

ACCEPTED MANUSCRIPT References

[1] Tofts PS, Kermode AG. Measurement of the blood-brain barrier permeability and leakage space using dynamic MR imaging. 1. Fundamental concepts. Magn Reson Med 1991;17:357367. [2] Tofts PS, Brix G, Buckley DL, et al. Estimating kinetic parameters from dynamic contrast-

PT

enhanced T(1)-weighted MRI of a diffusable tracer: standardized quantities and symbols. J

RI

Magn Reson Imaging 1999;10:223-232.

[3] Roberts HC, Roberts TP, Brasch RC, et al. Quantitative measurement of microvascular

SC

permeability in human brain tumors achieved using dynamic contrast-enhanced MR imaging: correlation with histologic grade. AJNR Am J Neuroradiol 2000;21:891-899.

NU

[4] Yang S, Law M, Zagzag D, et al. Dynamic contrast-enhanced perfusion MR imaging

MA

measurements of endothelial permeability: differentiation between atypical and typical meningiomas. AJNR Am J Neuroradiol 2003;24:1554-1559. [5] Ferl GZ, Xu L, Friesenhahn M, et al. An automated method for nonparametric kinetic

PT E

Reson Med 2010;63:1366-1375.

D

analysis of clinical DCE-MRI data: application to glioblastoma treated with bevacizumab. Magn

[6] Weber MA, Henze M, Tuttenberg J, et al. Biopsy targeting gliomas: do functional imaging

CE

techniques identify similar target areas? Invest Radiol 2010;45:755-768. [7] Zhang N, Zhang L, Qiu B, et al. Correlation of volume transfer coefficient Ktrans with

AC

histopathologic grades of gliomas. J Magn Reson Imaging 2012;36:355-363. [8] Jia Z, Geng D, Xie T, et al. Quantitative analysis of neovascular permeability in glioma by dynamic contrast-enhanced MR imaging. J Clin Neurosci 2012;19:820-823. [9] Hsu YH, Ferl GZ, Ng CM. GPU-accelerated nonparametric kinetic analysis of DCE-MRI data from glioblastoma patients treated with bevacizumab. Magn Reson Imaging 2013;31:618-623. [10] 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;14:487-492.

23

ACCEPTED MANUSCRIPT [11] Yun TJ, Park CK, Kim TM, et al. Glioblastoma treated with concurrent radiation therapy and temozolomide chemotherapy: differentiation of true progression from pseudoprogression with quantitative dynamic contrast-enhanced MR imaging. Radiology 2015;274:830-840. [12] Tofts PS. Modeling tracer kinetics in dynamic Gd-DTPA MR imaging. J Magn Reson Imaging 1997;7:91-101.

PT

[13] Ewing JR, Knight RA, Nagaraja TN, et al. Patlak plots of Gd-DTPA MRI data yield bloodbrain transfer constants concordant with those of 14C-sucrose in areas of blood-brain opening.

RI

Magn Reson Med 2003;50:283-292.

SC

[14] Sourbron SP, Buckley DL. On the scope and interpretation of the Tofts models for DCEMRI. Magn Reson Med 2011;66:735-745.

NU

[15] Heye AK, Culling RD, Valdes Hernandez Mdel C, et al. Assessment of blood-brain barrier disruption using dynamic contrast-enhanced MRI. A systematic review. Neuroimage Clin

MA

2014;6:262-274.

[16] Wang HZ, Riederer SJ, Lee JN. Optimizing the precision in T1 relaxation estimation using

D

limited flip angles. Magn Reson Med 1987;5:399-416.

PT E

[17] van Osch MJ, van der Grond J, Bakker CJ. Partial volume effects on arterial input functions: shape and amplitude distortions and their correction. J Magn Reson Imaging 2005;22:704-709. [18] Rohrer M, Bauer H, Mintorovitch J, et al. Comparison of magnetic properties of MRI

CE

contrast media solutions at different magnetic field strengths. Invest Radiol 2005;40:715-724.

AC

[19] Hansen AE, Pedersen H, Rostrup E, et al. Partial volume effect (PVE) on the arterial input function (AIF) in T1-weighted perfusion imaging and limitations of the multiplicative rescaling approach. Magn Reson Med 2009;62:1055-1059. [20] McGrath DM, Bradley DP, Tessier JL, et al. Comparison of model-based arterial input functions for dynamic contrast-enhanced MRI in tumor bearing rats. Magn Reson Med 2009;61:1173-1184. [21] Weinmann HJ, Laniado M, Mutzel W. Pharmacokinetics of GdDTPA/dimeglumine after intravenous injection into healthy volunteers. Physiol Chem Phys Med NMR 1984;16:167-172.

24

ACCEPTED MANUSCRIPT [22] Parker GJ, Roberts C, Macdonald A, et al. Experimentally-derived functional form for a population-averaged high-temporal-resolution arterial input function for dynamic contrastenhanced MRI. Magn Reson Med 2006;56:993-1000. [23] Alsop DC, Wedmid A, Schlaug G. Defining a local input function for perfusion quantification with bolus contrast MRI. Proc Intl Soc Mag Reson Med 2002;10:Abstract 659.

PT

[24] Yang C, Karczmar GS, Medved M, et al. Multiple reference tissue method for contrast agent arterial input function estimation. Magn Reson Med 2007;58:1266-1275.

RI

[25] Bleeker EJ, van Buchem MA, van Osch MJ. Optimal location for arterial input function

SC

measurements near the middle cerebral artery in first-pass perfusion MRI. J Cereb Blood Flow Metab 2009;29:840-852.

NU

[26] 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

MA

population. Magn Reson Imaging 2010;28:1420-1430.

[27] Mehrabian H, Chandrana C, Pang I, et al. Arterial input function calculation in dynamic

D

contrast-enhanced MRI: an in vivo validation study using co-registered contrast-enhanced

PT E

ultrasound imaging. Eur Radiol 2012;22:1735-1747. [28] Chen J, Yao J, Thomasson D. Automatic determination of arterial input function for dynamic

2008;11:594-601.

CE

contrast enhanced MRI in tumor assessment. Med Image Comput Comput Assist Interv

AC

[29] Peruzzo D, Bertoldo A, Zanderigo F, et al. Automatic selection of arterial input function on dynamic contrast-enhanced MR images. Comput Methods Programs Biomed 2011;104:e148157.

[30] Bleeker EJ, van Osch MJ, Connelly A, et al. New criterion to aid manual and automatic selection of the arterial input function in dynamic susceptibility contrast MRI. Magn Reson Med 2011;65:448-456. [31] Lin YC, Chan TH, Chi CY, et al. Blind estimation of the arterial input function in dynamic contrast-enhanced MRI using purity maximization. Magn Reson Med 2012;68:1439-1449.

25

ACCEPTED MANUSCRIPT [32] Azahaf M, Haberley M, Betrouni N, et al. Impact of arterial input function selection on the accuracy of dynamic contrast-enhanced MRI quantitative analysis for the diagnosis of clinically significant prostate cancer. J Magn Reson Imaging 2016;43:737-749. [33] Othman AE, Falkner F, Kessler DE, et al. Comparison of different population-averaged arterial-input-functions in dynamic contrast-enhanced MRI of the prostate: Effects on

PT

pharmacokinetic parameters and their diagnostic performance. Magnetic resonance imaging 2016;34:496-501.

RI

[34] Fedorov A, Fluckiger J, Ayers GD, et al. A comparison of two methods for estimating DCE-

SC

MRI parameters via individual and cohort based AIFs in prostate cancer: a step towards practical implementation. Magnetic resonance imaging 2014;32:321-329.

NU

[35] Huang W, Chen Y, Fedorov A, et al. The Impact of Arterial Input Function Determination Variations on Prostate Dynamic Contrast-Enhanced Magnetic Resonance Imaging

MA

Pharmacokinetic Modeling: A Multicenter Data Analysis Challenge. Tomography 2016;2:56-66. [36] Fram EK, Herfkens RJ, Johnson GA, et al. Rapid calculation of T1 using variable flip angle

D

gradient refocused imaging. Magn Reson Imaging 1987;5:201-208.

PT E

[37] Stanisz GJ, Henkelman RM. Gd-DTPA relaxivity depends on macromolecular content. Magn Reson Med 2000;44:665-667.

[38] Murase K. Efficient method for calculating kinetic parameters using T1-weighted dynamic

CE

contrast-enhanced magnetic resonance imaging. Magn Reson Med 2004;51:858-862.

AC

[39] Jackson A, Jayson GC, Li KL, et al. Reproducibility of quantitative dynamic contrastenhanced MRI in newly presenting glioma. Br J Radiol 2003;76:153-162. [40] 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;62:205-217. [41] Mills SJ, Soh C, O'Connor JP, et al. Enhancing fraction in glioma and its relationship to the tumoral vascular microenvironment: A dynamic contrast-enhanced MR imaging study. AJNR Am J Neuroradiol 2010;31:726-731.

26

ACCEPTED MANUSCRIPT [42] Jia ZZ, Geng DY, Liu Y, et al. Microvascular permeability of brain astrocytoma with contrast-enhanced magnetic resonance imaging: correlation analysis with histopathologic grade. Chin Med J (Engl) 2013;126:1953-1956. [43] 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

PT

contrast-enhanced MRI. Magn Reson Med 2011;65:108-119. [44] Parker GJ, Padhani AR (2003) T1-w DCE-MRI: T1-weighted Dynamic Contrast-enhanced

RI

MRI. In: Tofts PS (ed) Quantitative MRI of the Brain: Measuring Changes Caused by Disease. .

SC

John Wiley & Sons, Ltd p353

[45] van der Schaaf I, Vonken EJ, Waaijer A, et al. Influence of partial volume on venous output

NU

and arterial input function. AJNR Am J Neuroradiol 2006;27:46-50. [46] Bergamino M, Saitta L, Barletta L, et al. Measurement of blood-brain barrier permeability

MA

with t1-weighted dynamic contrast-enhanced MRI in brain tumors: a comparative study with two different algorithms. ISRN Neurosci 2013;2013:905279.

D

[47] Henderson E, Rutt BK, Lee TY. Temporal sampling requirements for the tracer kinetics

PT E

modeling of breast disease. Magn Reson Imaging 1998;16:1057-1073. [48] Heisen M, Fan X, Buurman J, et al. The influence of temporal resolution in determining pharmacokinetic parameters from DCE-MRI data. Magn Reson Med 2010;63:811-816.

CE

[49] Cutajar M, Mendichovszky IA, Tofts PS, et al. The importance of AIF ROI selection in DCE-

AC

MRI renography: reproducibility and variability of renal perfusion and filtration. Eur J Radiol 2010;74:e154-160.

27

ACCEPTED MANUSCRIPT

AC

CE

PT E

D

MA

NU

SC

RI

AIF – arterial input function CA – closest artery CNS – central nervous system Cp – plasmatic concentration of contrast agent Cp(t) – plasmatic concentration of contrast agent over time Ct – tissue concentration of contrast agent Ct(t) – tissue concentration of contrast agent over time DCE-MRI – dynamic contrast-enhanced MRI ICA – internal carotid artery ICC – intraclass correlation coefficient kep – transfer constant of contrast agent reflux Ktrans – transfer constant of contrast agent efflux ROI – region of interest SSS – superior sagittal sinus ve – volume fraction of extravascular-extracellular space vp – plasma volume fraction

PT

List of Abbreviations (in alphabetical order)

28

AC

CE

PT E

D

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

29

AC

CE

PT E

D

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

30

ACCEPTED MANUSCRIPT

Age (mean±S D; in years)

Gender (male/f emale)

NA WM

necro sis

NEL

IAC EL

EAC EL

Tiss ue 1

Tissu e5

Tiss ue 7

Tiss ue 8

Tiss ue 10/1 1

255

85

85

-

85

-

28

14

PT

Patie nts

tR OIs

-

14

-

-

5

-

5

-

-

68

34

-

34

-

-

18

9

-

9

-

-

48

22

4

18

4

-

12

6

-

6

-

-

12

3

-

-

9

-

10

5

-

5

-

-

40

13

13

-

14

-

8

4

-

4

-

-

18

10

-

6

2

-

56

28

-

-

-

28

230

100

-

-

-

130

Intraaxial lesions

AA WHO III° gliomato sis WHO II° PA WHO I° DNET

Tis sue 10 Tis sue 11

metasta ses

6

2

4

2

1

1 3

2

MS

Extraaxial lesions meningi oma WHO III° meningi oma WHO I°

43.0

1/0 10

1

PCNSL

2/0

1

AC

Tis sue 12 Tis sue 14 Tis sue 13

57.0±1.4

48.8±6.9

4/2

47.5±1.5

54.5±14. 3 43.5±24. 7

2/0

47.0

24.0 59.7±4.1

5

22

60.0

38.5±17. 7

NU

OA WHO II°

12/9

MA

OA WHO III°

2

60.7 ±11.0

2/2

1/1

D

ODG WHO II°

21

PT E

GB WHO IV° ODG WHO III°

CE

Tis sue 4 Tis sue 3 Tis sue 2 Tis sue 3 Tis sue 2 Tis sue 3 Tis sue 2 Tis sue 9

RI

Histolo gy

SC

Table 1 Overview of 821 tissue ROI used for AIF analyses in 74 patients

0/1

1/0 2/1

1/0

0/2

53.0±2 1.7

4/1

56.3±1 5.6

6/16

31

ACCEPTED MANUSCRIPT Tis sue 15

chordo ma

61.0

0/1 8

74

57.4±1 3.4

38/36

821

4

-

-

-

4

342

102

101

114

162

PT

total

1

RI

Tissue stratification was performed based on histology (left column, bold print, numbers also in bold print) or appearance on T2W /contrast-enhanced T1W MRI (top right line, five tissues, all

SC

numbers for the respective columns). NAWM (tissue 1) was examined for both stratification

NU

groups. SD: standard deviation; GB: glioblastoma; WHO: World Health Organization (tumor grading); ODG: oligodendroglioma, OA: oligoastrocytoma, AA: anaplastic astrocytoma, PA:

MA

pilocytic astrocytoma, DNET: dysembryoblastic neuroepithelial tumor, PCNSL: primary CNS lymphoma, MS: multiple sclerosis; tROIs: number of target tissue region of interest; IACEL: intra-axial contrast-enhancing lesion, EACEL: extra-axial contrast-enhancing lesions; NEL: non-

AC

CE

PT E

D

enhancing T2-hyperintense lesion; NAWM: normal-appearing white matter.

32

ACCEPTED MANUSCRIPT Table 2 AIF-dependent variability of kinetic parameters from 821 tissue ROI and four AIF sources AIF source

8.20 (4.71 – 12.43) 103.10 (70.75 – 123.37)

53.22 (37.14 – 73.50) 494.59 (316.90 – 643.22)

5.1 (4.45 – 6.91) 26.42 (23.00 – 29.71)

796.76 (658.54 – 1009.74)

64.39 (59.27 – 69.69)

0.86 (0.36 – 2.44) 5.11 (1.77 – 10.41) 0.38 (0.07 – 0.78)

79.72 (59.49 – 122.81) 575.7 (470.12 – 723.65) 1672.66 (1372.96 – 1849.45) 4.31 (2.67 – 9.97) 24.38 (8.17 – 74.50) 3.36 (2.39 – 6.43)

3.06 (0.95 – 6.66) 11.53 (6.10 – 37.62) 1.81 (0.67 – 3.11)

2.23 (1.67 – 2.63) 1.83 (0.85 – 4.50) 2.26 (2.08 – 2.60)

300.10 (248.38 – 359.50) 384.54 (337.75 – 412.76) 800.28 (631.24 – 904.99) 113.08 (0.0 – 316.91) 44.18 (0.0 – 174.18)

258.53 (229.36 – 290.30) 308.80 (269.51 – 345.84) 575.00 (531.79 – 651.28) 167.10 (0.0 – 327.05) 0.0 (0.0 – 109.79)

251.80 (197.36 – 277.7) 309.07 (279.54 – 344.06) 617.53 (577.53 – 669.88) 81.85 (0.0 – 236.07) 0.0 (0.0 – 110.49)

280.94 (255.78 – 296.73) 160.22 (142.14 – 179.84) 354. 27 (307.81 – 388.0) 394.56 (311.59 – 444.38) 29.75 (0.0 – 94.36) 347.02 (303.74 – 386.75)

167.06 (141.77 – 186.63)

NEL necrosis NAWM all IACEL kep (*103 /min)

EACEL NEL necrosis

0.0

PT E

NAWM

0.0 15.73 (10.14 – 23.68) 155.38 (122.40 – 205.17) 127.19 (114.26 – 155.45) 0.41 (0.0 – 1.61) 0.0 (0.0 – 10.59)

necrosis

0.31 (0.0 – 3.94)

NAWM

0.0

0.0

0.0

2.0 (1.50 – 2.70) 8.94 (7.07 – 11.32) 9.48 (7.82 – 12.32) 0.45 (0.28 – 0.63) 0.49 (0.25 –

14.12 (12.0 – 17.20) 33.97 (28.19 – 41.06) 82.39 (70.62 – 94.88) 4.44 (2.91 – 6.43) 2.80 (1.6 –

9.83 (8.86 – 11.50) 30.98 (25.84 – 36.22) 66.15 (59.16 – 73.44) 4.23 (3.29 – 5.83) 1.77 (0.93 –

CE

IACEL EACEL ve (%)

0.0 (0.0 – 0.38)

AC

NEL

all IACEL vp (%)

0.0

25.64 (14.15 – 46.61) 162.13 (114.5 – 216.90) 248.83 (223.88 – 288.72) 0.57 (0.0 – 1.62) 0.0 (0.0 – 33.67)

all

2.14 (1.25 – 5.70) 25.66 (20.93 – 32.55) 22.78 (19.97 – 25.28)

PT

EACEL

PM

RI

Ktrans (*103 /min)

CA

SC

IACEL

ICA

NU

all

SSS

MA

Tissue

D

Kinetic parameter

EACEL NEL necrosis

1.94 (1.47 – 4.02) 20.12 (17.53 – 23.13) 21.56 (20.36 – 24.61) 0.50 (0.37 – 0.59) 0.32 (0.0 – 1.70) 0.57 (0.50 – 0.62) 0.23 (0.19 – 0.32) 0.47 (0.39 – 0.52) 1.1 (0.95 – 1.21) 0.09 (0.08 – 0.11) 0.05 (0.02 –

33

ACCEPTED MANUSCRIPT

NAWM

1.14) 0.75 (0.63 – 0.84)

5.41) 6.24 (5.75 – 7.09)

4.70) 5.09 (4.70 – 5.64)

0.11) 0.11 (0.10 – 0.12)

The table depicts medians and confidence intervals (CI in brackets) of kinetic parameters measured in a total of 821 tissue ROI based on four different AIF sources. AIF: arterial input function; Ktrans: transfer constant of contrast agent efflux; k ep: transfer constant

PT

of contrast agent reflux; ve: ratio of transfer constants of contrast agent efflux and reflux; vp: plasma volume fraction; SSS: superior sagittal sinus; ICA: internal carotid artery; CA: closest

RI

artery to the lesion; PM: population-derived Parker model. IACEL: intra-axial contrast-enhancing

SC

lesion, EACEL: extra-axial contrast-enhancing lesions; NEL: non-enhancing T2-hyperintense

AC

CE

PT E

D

MA

NU

lesion; NAWM: normal-appearing white matter

34

ACCEPTED MANUSCRIPT Table 3 Goodness of fit intraclass correlation of kinetic parameters from different AIF sources

Intraclass correlation coefficients of kinetic parameters (n=821 tROI)

ve vp

ICA vs. SSS 0.30 (0.23 – 0.36) 0.55 (0.50 – 0.59) 0.20 (0.13 – 0.26) 0.13 (0.06 – 0.19)

ICA vs. CA 0.57 (0.52 – 0.62) 0.71 (0.67 – 0.74) 0.52 (0.47 – 0.57) 0.87 (0.85 – 0.89)

PT

PM vs. CA 0.24 (0.18 – 0.31) 0.10 (0.03 – 0.17) 0.36 (0.30 – 0.42) 0.56 (0.51 – 0.60)

RI

kep

PM vs. ICA 0.64 (0.60 – 0.68) 0.08 (0.01 – 0.15) 0.18 (0.11 – 0.24) 0.67 (0.63 – 0.70)

SC

Ktrans

PM vs. SSS 0.51 (0.46 – 0.56) 0.18 (0.11 – 0.24) 0.60 (0.55 – 0.64) 0.42 (0.37 – 0.48)

NU

Tissue

CA vs. SSS 0.09 (0.02 – 0.16) 0.47 (0.42 – 0.52) 0.43 (0.38 – 0.49) 0.19 (0.05 – 0.19)

AC

CE

PT E

D

MA

Upper lines: intraclass correlation coefficient, lower lines: corresponding confidence intervals. AIF: arterial input function; SSS: superior sagittal sinus; ICA: internal carotid artery; CA: closest artery to the lesion; PM: population-derived Parker model

35