Pharmacokinetic (PK) parameters derived from DCE-MRI data are sensitive to both temporal resolution and signal to noise ratio (SNR)¹. There is a balance to be reached between temporal resolution and SNR such that the data is optimised for a given PK model; however, no method exists to quantitatively assess the PK modelling accuracy of different acquisition approaches, a situation which is exacerbated by potential inaccuracies introduced by fast acquisition techniques such as compressed sensing (CS). To address this shortcoming, a flow phantom was designed and built which can produce contrast time-intensity curves (CTCs) representative of those observed in DCE-MRI data of the prostate. The ability of the device to quantitatively evaluate the effects of different acquisition strategies on derived PK parameters is demonstrated.

A prostate sized object was manufactured containing two measurement chambers, designed for optimum homogeneous mixing of fluids at low flow rates (1-4ml/s)², as well as a further arrangement of chambers (Figure 1(a)) that mimicked the overall enhancement of the prostate observed in vivo. To realistically challenge DCE protocols, the ‘prostate’ was set into a large custom-built anthropomorphic phantom that mimicked the male pelvis (Figure 1(b)).

CTCs were produced using a custom-built computer-controlled multi-pump system (Reglo-Z, ISMATEC, Switzerland), capable of simultaneously producing two different arbitrarily shaped CTCs within the measurement chambers by varying the relative concentration of a gadolinium-based contrast agent (CA) solution (Multihance, Bracco, USA), while keeping the overall flow rate constant.

Calibrated ‘Truth Estimates’ for the CTCs produced by the system after traveling the 11 m from the pumps to the measurement chambers (pumps in the scanner control room) were measured using a custom-built optical imaging system (Fujinon-Σ400 endoscopic light source, and high resolution CMOS Cannon-20D camera), using black dye as a CA surrogate.

To demonstrate the operation of the phantom, DCE-MRI data was acquired using a 3T scanner (Achieva, Philips, Netherlands) and 16-channel phased array detector coil (3D-SPGR, TR/TE = 3.6/1.75ms, flip angle = 10°, voxel size = 1.1 x 1.1 x 4 mm³, FOV = 224 x 224 x 72 mm³, slices = 18). The parallel imaging factor and number of signal averages were varied to give temporal resolutions from 2.3 s to 20.3 s. The DCEMRI.jl toolkit³ was used to derive K^trans values for the truth estimates and MR data from a hand selected region of interest containing 35 voxels using the Extended Tofts model.

Measurement chambers were manufactured with wall thicknesses of 0.3 mm, minimising any susceptibility artefacts. Agar-based tissue mimic materials were produced which closely match the T₁ and T₂ properties of bone, muscle, and fat tissue⁴ (+4%, -0.7%, and +20% respectively).

Good temporal stability was observed for measurements made using the optical scanner (±0.37% over 30 minutes). Full CTC runs were measured using the optical system at flow rates from 0.5 to 2.5 ml/s (Figure 2) and the minimum flow rate at which the CTCs are accurately produced was found to be 1.5 ml/s (R² = 0.98). Two different CTCs (healthy and tumorous) were subsequently run four times at 1.5 ml/s to test for repeatability; good correlation was observed between the repeated CTCs (Figure 3).

K^trans values derived from the MR data were compared with those derived from the truth estimates and were found to differ by -8.1% to -44.6%, with the lowest variance from the truth estimate values achieved at a temporal resolution of 6.8 s (see Figure 4).

The anthropomorphic nature of a phantom allows for repeatable measurements of PK parameters in an environment closely emulating that observed in vivo (see representative MR image of the phantom in Figure 5). The optical scanner used to calibrate the system provided a novel and inexpensive way to validate the CTCs produced by the system using a modality other than MRI, and also allowed for minimum flow rates to be established, which is important on two fronts: the minimisation of flow artefacts and the reduction in the amount of CA used. For the curve shapes used in this study, the lowest variation in K^trans values occurred at 6.8 s temporal resolution. At higher temporal resolutions, inaccuracies in the resultant K^trans values were most likely due to low SNR, and at the lower temporal resolutions due to the reduced number of sampling points. The device can be used for optimising DCE acquisition protocols and investigating the effect of fast acquisition schemes (e.g. CS) on the PK modelling accuracy. The ability of faster acquisition schemes to faithfully measure rapidly-varying CTCs, such as the arterial input function, can also be explored using this device.