Dynamic contrast-enhanced (DCE) MRI is a promising imaging tool to non-invasively estimate tissue perfusion and permeability, giving insights into a tumor microvascular environment. For 3D DCE MRI, fast spoiled gradient echo (SPGR) imaging acquisition is performed before, during, and after intravenous bolus administration of gadolinium-based contrast agent, thus capturing tumor kinetic information [1]. DCE MRI data supports qualitative, semi-quantitative and fully quantitative analysis [2]. In a typical qualitative analysis an observer manually categorizes the time-intensity curves. In semi-quantitative methods, aspects of the time intensity curves such as time-to-peak or area under curve are measured. Qualitative and semi-quantitative methods suffer from higher inter- and intra-observer variation as well as inter- and intra-patient variation than fully quantitative. Thus, fully quantitative analysis is preferred with the most common approach employing deconvolutional two-compartmental models characterized by the contrast agent concentrations in tissue, arterial blood, and two first order kinetic rate constants (Ktrans, kep) [3]. Quantitative analysis provides a more precise assessment of the pharmacokinetic parameters of tumor sub-regions for the potential clinical application of precision medicine.

The deconvolutional two-compartmental models used in the analysis of DCE-MRI data, rely heavily on the arterial input function (AIF) [[4], [5], [6], [7]]. Some advocate using the population-based AIF [8,9] over individually measured AIF [[10], [11], [12]] due to the simplicity and standardization of the analysis. However, patient related factors, such as heart rate, ejection fraction, or kidney function, affect AIF and may contribute to large variations of AIF. Port et al. measured the arterial input function in 47 patients with breast cancer and assessed the variability of kinetic parameters based on individual AIF compared to the average AIF [11]. They concluded that AIFs should be monitored individually to estimate DCE parameters accurately. DCE MR for head and neck cancer is generally acquired in the axial plane and uses AIF in the cervical vessels, including the internal carotid artery [13], which runs in the craniocaudal direction approximately perpendicular to the axial plane. Previous publications reported variations of Ktrans, kep, and other DCE output based on the location of AIF on the internal carotid artery, external carotid artery, or vertebral artery in the head and neck [14,15].

Often the AIF is calculated from the DCE signal in an artery in the image field of view. The SPGR signal models used in 3D DCE analysis assume that the tissue of interest, including the blood from which the AIF is measured, remain stationary within the imaging volume. The tissues must receive multiple excitations in order to approach the steady state value described by the SPGR signal model. Depending on its velocity, flowing blood may not remain within the imaging volume long enough to experience the excitations necessary to reach or approach steady state. Previous investigators have noted that patient based AIF must be selected with care [[15], [16], [17], [18], [19], [20]]. Garpebring et al. demonstrated that when the AIF is measured in the presence of flow it significantly reduces the accuracy and increases the variability of the DCE model outputs [18]. They found that the effect could be reduced but not eliminated by measuring the AIF in a vessel downstream of its entry into the imaging volume. The DCE model outputs may be made repeatable but with a flow dependent bias, precise but not accurate.

The purpose of this study was to investigate the effect of sampling location along the axial direction (Superior to inferior: cranial-caudal direction) on the AIF itself and the resulting quantitative values of DCE metrics for muscle in head and neck cancer (HNC) patients. We consider that factors in addition to inflow effects may determine the ideal AIF sampling location. The standard SPGR signal model was employed and the AIF measured at the different axial locations were compared against a published population average AIF [21]. Both image-based and population average AIF were considered when solving the extended Tofts DCE model [16] and the results compared. This publication expands upon preliminary work presented at conference [22].

In SPGR acquisitions, spins are repeatedly tipped into the imaging plane and spoiled prior to the next excitation. The longitudinal magnetization, Mz, available prior to the (n + 1)th excitation is given by:MzTRn+1=MzTRncosαexp−TRT1+M01−exp−TRT1

WithMzTR0=M0where M0 is proportional to the tissue spin density, α is the flip angle, TR is the repetition time and T1 is the tissue-dependent longitudinal decay constant. The signal available at echo time, TE, is given as:S∝MzTRn+1sin∝exp−TET2∗where T2* is another tissue-dependent time constant.

After many repetitions, the signal converges to value given by the equation:S∝M0sinα1−exp−TRT11−cosαexp−TRT1exp−TET2∗

In 3D SPGR imaging, the acquisition of k-space is typically ordered from outer to central k-space. Successive k-space lines are acquired during successive excitations with the initial lines recording the approach to equilibrium described by Eq. (1). When the central portion of k-space is acquired, the stationary tissues within the imaging volume will have experienced enough excitations to reach the steady state described by Eq. (4). Because the overall intensities in the DCE images are governed primarily by central k-space, after the Fourier transform of k-space, the image intensities will reflect convergence to Eq. (4). If a short TE is used during acquisition, one can ignore T2* effects, leaving the only unknowns the spin density term M0 and the longitudinal decay constant, T1.

In 3D SPGR DCE MRI, we can measure T1 in the absence of contrast using the variable flip angle acquisition [23,24]. Multiple image volumes are acquired with short TE at varied flip angles and Eq. (4) applied to compute the unknowns, M0 and T1. For the DCE time series, the flip angle is fixed and images are acquired before and during the passage and uptake of contrast. Given M0, we apply Eq. (4) to our DCE MRI time series and solve for T1 as a function of time.

The longitudinal decay constant, T1, may be related to contrast concentration as follows:T1obs−1=T1c=0−1+R1Cwhere C is the contrast concentration and R1 is the specific relaxivity (Prohance, 3.41 s−1 mM−1) [25]. We can then convert the T1 values from our VFA and DCE MRI image series into concentration time curves. These then serve as input to the physical models for tissue perfusion and permeability.

In this work, we employ the extended Tofts-Kermode model [26,27] to describe the concentration time curves for tissue, Ct(t):Ctt=υpCpt+Ktrans∫0te−kept−τCpτdτwhere vp is the blood plasma volume per unit volume of tissue, Cp(t) is the contrast concentration in blood plasma, Ktrans is the permeability, and kep is the efflux rate constant from the extra-vascular extra-cellular space into plasma. The contrast concentration in the blood plasma is taken from the concentration computed for a region of vessel, suitably scaled to account for the effect of hematocrit volume percentage on contrast concentration.