Simulation-Informed Evaluation of Microvascular Parameter Mapping for Diffusion MR Imaging of Solid Tumours

We aim to inform the design of new diffusion MRI (dMRI) approaches for microvasculature quantification that enhance the biological specificity of imaging towards cancer.

We adopted simulation-informed modelling of the vascular dMRI signal. We synthesised signals from 1500 synthetic vascular networks, for a variety of protocols (flow-compensated [FC], non-compensated [NC], hybrid), featuring different $b$ samplings and diffusion times. We estimated the number of independent, recoverable signal degrees of freedom in presence of noise (signal-to-noise ratio of 5), and ranked 12 microvascular metrics depending on the quality of their estimation. Lastly, we demonstrated the feasibility of estimating the top-ranking metrics on 3T dMRI of a healthy volunteer and of a metastatic colorectal cancer (CRC) patient.

Both NC and FC synthetic vascular signals exhibited complex behaviour as, for example, non-zero kurtosis and diffusion time dependence. Two independent degrees of freedom appeared recoverable from directionally-averaged vascular signals (SNR of 5). Mean volumetric flow rate $q_m$ and an Apparent Network Branching (ANB) index maximised correlations between ground truth and estimated values in silico. In the patient, both $q_m$ and $ANB$ detected re-vascularisation after 3 months of targeted therapy against liver metastases, consistently with Intra-Voxel Incoherent Motion (IVIM) metrics.

Simulation-based modelling of the vascular dMRI signal suggests $q_m$ and $ANB$ as the most promising metrics for tissue microvasculature characterisation. Their estimation in vivo appears feasible to capture general trends, and demonstrates contrasts that are biologically plausible, encouraging their usage in future studies.