A Monte Carlo simulation framework for histology-informed diffusion MRI cancer characterisation and microstructural parameter estimation

Computer simulations within substrates that mimic the complexity of biological tissues are key to the development of biophysical diffusion Magnetic Resonance Imaging (dMRI) models. Realistic simulations have enabled, for example, the non-invasive estimation of fine neuronal sub-structures, which is playing an increasingly key role in neurology and neuro-science. However, biologically-realistic, simulation-informed dMRI techniques are also needed in other applications, as for example in oncological imaging of body tumours. This article aims to fill this gap by presenting a Monte Carlo (MC) framework tailored for histology-informed simulations in body imaging applications. The framework, which combines free software with custom-written routines, is demonstrated on substrates reconstructed from hematoxylin-eosin (HE) stains of human liver biopsies, including non-cancerous liver and primary/metastatic liver cancer tissues. The article has four main contributions. Firstly, it provides practical guidelines on how to conduct realistic MC diffusion simulations informed by HE histology. Secondly, it reports reference values on cell size (CS), cell density and on other cellular properties in non-cancerous and cancerous liver — information not easily found in the literature, yet essential to inform the design of innovative dMRI techniques. Thirdly, it presents a detailed characterisation of synthetic signals generated for clinically feasible dMRI protocols, shedding light onto patterns of intra-/extra-cellular (IC/EC) water diffusion in liver. Finally, it illustrates the utility of the framework, by devising a strategy where synthetic signals inform the estimation of unexplored microstructural properties, as the EC intrinsic diffusivity and CS distribution skewness. The strategy is demonstrated on actual dMRI scans, acquired on ex vivo mouse tissue and in cancer patients in vivo .