Diffusion MRI for tumor microstructure imaging using VERDICT modeling

Abstract

VERDICT is a method which uses a mathematical model that provides estimates of microstructural tumor tissue parameters based on diffusion-weighted MRI data. It is a promising imaging method for non-invasive in vivo evaluation of whole-tumor tissue. However, model assumptions may introduce systematic errors in parameter estimates.

The aim of this thesis was to assess the use of VERDICT for tumor tissue evaluation and investigate the impact of model assumptions on parameter estimates, as well as to develop and evaluate methods addressing accuracy issues related to some of these assumptions.

The standard clinical approach for evaluating tumor treatment response is by measuring changes in gross tumor volume. However, such changes can be slow, and methods sensitive to microstructural changes may detect response earlier. Paper I investigates the use of VERDICT parameters for radiation treatment response assessment and shows that early parameter changes correlate with treatment outcome.

Histological analysis remains the gold standard for assessing tumor microstructure, but tumor heterogeneity limits biopsy representativeness. Paper II explores the use of VERDICT for whole-tumor tissue classification as a potential complement to histology. The work shows that multidimensional cluster analysis of VERDICT parameters enables classification of distinct tumor tissue types.

Model assumptions can introduce systematic errors in parameter estimates. Paper III investigates the effect of assumptions related to extracellular–extravascular diffu-sion and presents a Monte Carlo-based model which explicitly accounts for diffusion time dependence. Paper IV investigates the impact of including compartment-specific T2 relaxation in the model, in contrast to uniform T2 relaxation across compartments as assumed in conventional VERDICT. These works show that model assumptions can significantly influence parameter estimates and present methods to mitigate their effects.

In conclusion, the results of this thesis highlight the importance of accurate model assumptions in VERDICT, and demonstrate the model’s potential for non-invasive, whole-tumor evaluation of tumor tissue in various applications.