The analysis of dynamic contrast-enhanced magnetic resonance imaging data: treatment effects, sampling rates and repeatability

Abstract

This dissertation describes the analysis of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) data, primarily by means of pharmacokinetic models, in order to yield parameters which reflect the physiological properties of tumours. Historically, these analysis procedures have been complicated by varied sources of error, both systematic and random. This dissertation investigates improvements to such techniques and the hypothesised gain in accuracy when implementing a more elaborate analysis strategy. The theory of MR imaging as it relates to DCE-MRI is described, and the pharmacokinetic models most often used for data analysis are critically appraised. The problem of extracting an accurate and precise arterial input function (AIF) is reviewed, as are current proposed solutions. The general principles underlying the implementation of a customised software analysis package for the analysis of DCE-MRI data are described. Data from studies involving four different human tumour types are investigated and the analysis methods applied to each are incrementally optimised. In the first two studies, with data from ovarian and prostate tumours, it is hypothesised that the effects of platinum-based chemotherapy and androgen deprivation therapy (ADT) respectively can be correlated with changes in standard DCE-MRI parameters (Ktrans, ve and vp). The results show that ADT is reflected by decreasing tumour perfusion and vascularity (Ktrans and vp) whereas platinum chemotherapy has less clear anti-vascular effects. This is in accord with the known action of these agents. In the third study it is hypothesised that very high temporal resolution image acquisitions can improve dual-input pharmacokinetic modelling in the liver. Rapid data acquisition was performed using a dualsection saturation-recovery imaging method at 3T. It is shown that the acquired data were sub-sampled to a moderate temporal resolution. A separate study investigated the effects of B1 non-uniformity and found that the parameter results were only minimally affected by application of a method correcting for variations in the saturation pulse flip-angle. The effect of contrast agent signal non-linearity was investigated using an additional CT study which provided a sample of vascular input functions (VIFs) for comparison with those measured by MR. VIFs from each modality showed a similar degree of inter-patient variability. In the final study the hypothesis is that the parameter results from DCE-MRI examinations are repeatable between patient visits. Data from two identical examinations, performed pre-treatment in patients with metastatic renal cell carcinoma, were used to investigate the intra-subject repeatability of AIF extraction and DCE-MRI parameter evaluation. Analysis methods were developed to include histogram analysis of pharmacokinetic model parameter maps and the use of differing AIF forms, whether modelled, individually-measured or sample-averaged. The use of ‘error-code mapping’ to improve the accuracy of parameter extraction was introduced and the combined effects on repeatability of differing analysis strategies were analysed. The results show that parameter evaluation is repeatable (to a defined degree) but that no single analysis technique, complex or minimal, significantly out-performs the others.