| dc.description.abstract | Introduction: Dynamic contrast enhanced (DCE) MRI, combined with pharmacokinetic (PK) modelling of the resulting data, has shown great promise for the detection, monitoring, and grading of prostate cancer. However, to date a method has been lacking to quantitatively assess the accuracy of DCE-MRI measurements, stemming from a lack of knowledge of the ground-truth conditions within the patient or object being scanned. This lack has hindered the widespread clinical implementation of this technique. In this thesis a novel DCE-MRI phantom device is presented which addresses this need. The operation of the device was demonstrated in several experiments which quantified the effects of various methodological factors on the absolute accuracy and precision of DCE-MRI measurements.
Methodologies: Initially, the design and construction of a novel DCE-MRI phantom system was outlined, wherein programmed concentration time-intensity curves, mimicking those observed in healthy and tumorous prostate tissue (CTC(tissue)), as well as an arterial input function (AIF; established from the mean of 32 prostate patient DCE-MRI datasets), were generated using a custom-built four-pump flow system. The device was anthropomorphic in nature, mimicking the properties and distribution of tissue in the male pelvic-region in terms of size and complexity. A highly-precise custom-built optical imaging system was then described, and used to validate the stability of the phantom and to establish ground truth CTC(tissue) and AIF for the device. Next, four sets of experiments were performed wherein DCE-MRI datasets were acquired from the phantom device using a 3T scanner (Philips, Netherlands) and a 32-channel detector coil. Collectively these experiments investigated the effects of: (i) temporal resolutions (T(res); varied overall in the range [0.09 ? 30.6 s]), (ii) CTC acquisition duration (AD; varied in the range [30 - 600 s]), (iii) voxel-wise flip-angle correction (VFAC), (iv) PK model fitting regime (non-linear verses linear), and (v) image reconstruction methods for under-sampled data (coil-by coil (CbC), parallel imaging (PI) only, and PI with compressed sensing (PICS)). Experiments (i) ? (iv) utilised a 3D Cartesian spoiled-gradient-echo (SPGR) sequence, while experiment (v) used a 2D continuous golden-angle radial SPGR sequence. For all experiments, errors in DCE-MRI measurements were calculated against the precisely-known ground truth values.
Results: The anthropomorphic phantom device was demonstrated to be capable of producing both CTC(tissue) and AIF curve-shapes simultaneously, with profiles mimicking those observed in patient data. This essentially provided a ?model patient? wherein the ground-truth conditions were precisely known, thereby allowing for a quantitative assessment of each DCE-MRI approach tested. For the measurement of CTC(tissue) (coupled with use of a model AIF), initial PK modelling revealed errors in output PK parameter values of up to 42%, 31%, and 50% for K(trans) (volume transfer coefficient), v(e) (extravascular-extracellular space volume fraction), and k(ep) (rate constant), respectively, following a simple variation of the T(res), using a Cartesian 3D SPGR acquisition. Further, it was demonstrated that errors in all derived PK parameter values were < 14% for acquisitions with Tres <= 8.1s and AD >= 360s, but increased dramatically outside of this range. Flip angle errors deriving from B1+-field non-uniformities of -29% to -33% were measured in the phantom. For the simultaneous measurement of both the CTC(tissue) and AIF, the application of VFAC to the data to correct for the B1+-field non-uniformities increased the PK parameter estimation accuracy by 12.9%, 9.2% and 20.2% for K(trans), v(e) and k(ep) respectively, and increased intra-session precision by up to 11.2%. The use of a linear implementation of the standard Tofts PK model, rather than the more often-used non-linear form, almost doubled the accuracy of K(trans) estimates and increased K(trans) and k(ep) measurement precision by approximately 4%; this also relaxed the dependence on the T(res) used. The PICS reconstruction approach was demonstrated to provide K(trans), v(e), and k(ep) estimations with errors less than 12%, 6.6%, and 12%, respectively, for image sequences reconstructed with as little as 3% of the data, corresponding to a twenty-one-fold gain in acquisition speed compared with a fully-sampled Cartesian approach with the same spatial resolution and geometry. PICS reconstruction was also shown to provide a gain in the accuracy in PK parameter estimations of up to 34% and 42%, compared with the PI and CbC approaches, respectively.
Conclusions: The work of this thesis highlighted the critical dependence of the accuracy and precision of DCE-MRI measurements on the methodologies used, and demonstrated a novel phantom-based method whereby DCE protocols can be refined and optimised. Using this phantom-based method to perform a series of quantitative DCE-MRI experiments, absolute errors in PK parameter estimation were reduced from 50% (using a standard Cartesian acquisition) to 2% (using a continuous GA radial acquisition with VFAC applied and linear Tofts PK model fitting), with a similar improvement in precision from 18% to 4%. The use of quantitative phantom-based approaches, such as the one described herein, to access and optimise the accuracy and precision of DCE-MRI techniques, offers the prospect of standardising DCE acquisition protocols for the prostate and beyond, and ultimately wider acceptance of the technique into routine clinical use. | en |
