Optimization of b-value Acquisition and Parameter Estimation of Intravoxel Incoherent Motion Measurements

Sammanfattning: Radiotherapy is commonly used for cancer treatment, given to approximately half of all cancer patients in Sweden. Despite its widespread usage, there is currently no non-invasive and reliable imaging method that can identify regions of radiation resistance. These regions may be identified by hypoxia, given that radiobiological research has demonstrated a significant correlation between oxygen levels and the extent of radiation-induced cellular damage. A recent study has shown that magnetic resonance imaging (MRI) of intravoxel incoherent motions (IVIM) using diffusion-weighted MRI of the prostate shows promise for hypoxia mapping. The basis of these hypoxia maps are the IVIM parameters perfusion fraction and diffusion coefficient. IVIM is recognized for being sensitive to noise and showing low reproducibility of parameter values. Existing literature highlights two issues that need to be resolved: optimization of image acquisition and reliable parameter estimation. The aim of this work is to address these issues by optimizing an acquisition protocol and to evaluate two new fitting algorithms; MIX and TopoPro; by comparing their estimation errors with commonly used algorithms to date. Protocol optimization was performed with respect to diffusion weightings (b-values), and the number of averages as free parameters. The Cramér-Rao Lower Bound (CRLB) was used to provide estimates for the variances of the IVIM parameters for a set of b-values. Numerical minimization of the CRLB was employed to find optimal sets of b-values. Three protocols with different properties were produced and evaluated on a wide variety of simulated IVIM signals with respect to accuracy and precision. Comparison was made to a generic IVIM protocol typically found in literature. The best performing protocol was used in an in-vivo evaluation on the prostate and compared to the generic protocol with respect to estimation convergence. The optimized acquisition of in-vivo data improved convergence where the generic protocol otherwise would have failed to provide estimates due to noisy data. The optimized protocol increased the number of converged voxels by up to 20\% and was found to vary between fitting algorithms. In conclusion, an optimized acquisition protocol may yield improved parameter estimates. The evaluation of fitting algorithms was performed on simulated IVIM signals for three signal-to-noise ratios, where the signals were produced using a wide range of IVIM parameter combinations. MIX and TopoPro were compared to four conventional algorithms. Three of the conventional algorithms were found to be the most precise, but less accurate in certain parameter limits. MIX was found to be both precise and accurate. Furthermore, MIX showed a smaller dependence on the acquisition scheme than the others, both in simulations and in-vivo. MIX is therefore recommended as it is accurate, precise, and reduces the acquisition-dependency of IVIM parameter estimates. Furthermore, the usage of the Linear fit is not recommended as it showed the lowest accuracy, and lowest precision.