Statistical reconstruction of histology data based on magnetic resonance imaging (HistoStat)

One of the major aims in medicine and human biology is to understand biological processes and their impact on tissue structure and function. This is of paramount importance for gaining insights into mechanisms found in health and disease, which will eventually lead to new discoveries and identification of novel biomarkers. In this regard, the field of medical imaging was born with the ultimate aim of providing such information in vivo and non-invasively. Numerous medical imaging methodologies have been developed to study soft tissue structure, from which magnetic resonance imaging (MRI) stands out of the rest. The reasons are based on the flexibility of the technique to provide different tissue contrasts using non-ionising radiation, and therefore avoiding long-lasting effects. Although MRI has proven its value for understanding tissue and material properties, it has a practical image resolution limit that prevents going beyond the millimetre scale. This is a very important constraint for obtaining microscopic information that may help to understand the causes of such macroscopic measurements.

Since the 1990s, researchers have presented techniques for estimating tissue microstructures from MRI. Despite all the advances in the field, existing methodologies have intrinsic limitations for providing accurate and complete descriptions of tissue microarchitecture. All of the available techniques developed to estimate structural dimensions are based on hypotheses regarding the morphology of the cellular compartments, e.g. cells represented as perfect cylinders or spheres as appropriate. The reason is based on the need of mathematical formulas relating microstructural features with MRI signals, which become intractable for non-trivial geometries. This imposes a limitation on tissue depiction, since these models cannot account for features such as cell's shape and arrangement that are of the utmost significance to specialists. These constraints highlight the importance of evaluating alternative routes to generate histology-like images from non-invasive MRI acquisitions. The development of a flexible framework to reconstruct tissue accurately would present a step change not only in the medical imaging community, but also in extending its applicability to countless problems in biology, medicine, and engineering.

In this project, we propose a paradigm shift in the area of MRI-based microstructure characterisation by introducing HistoStat, a statistical framework that allows reconstruction of tissue architecture without imposing limiting models. MRI provides macroscopic measurements from which the description of microstructural compartments is targeted. The variability of morphometries, densities, and general magnetic properties within a measurement unit means that such a description can only be done statistically. HistoStat takes this into account by presenting a comprehensive methodology for describing microstructure from a statistical standpoint. More specifically, HistoStat will be based on the use of the MRI scanner to estimate a series of statistical descriptors (SDs) representing different tissue micro compartments. By SD, we mean a function that encodes certain aspects of the relative arrangement and shapes of microstructural constituents. These descriptors arise, for example, in mathematical expressions of macroscopic physical properties, which are known to depend on the micro-environment. The estimated SDs will be then utilised, without any further information, to reconstruct histology-like images resembling tissue micro-architecture, providing insights into tissue structure not accessible by any other non-invasive imaging modality. We will demonstrate the framework in controlled experiments, preparing the grounds for further applications in healthcare and engineering problems.

Non-invasive characterisation of tissue structures at the cellular level is crucial for understanding biological processes in health and disease. Of the several available methodologies for doing so, MRI holds the greatest promise due to its flexibility and the non-ionising radiation used. Scholars in the field have designed methods for describing a variety of tissues and materials at the microscale. By fitting parametric signal models to MRI data, researchers were able to estimate microstructural features such as cell's radii and density within a voxel. However, most of these models are constrained by the hypotheses behind the models themselves. Such hypotheses often include assumptions regarding cellular morphology and other magnetic properties that are not necessarily detailed enough. This imposes a limitation for proper tissue depiction, since these models cannot account for features such as cells morphology and arrangement that are of the utmost importance to specialists.
In this project, we propose to tackle the problem by introducing HistoStat, a statistical framework that allows reconstruction of tissue microstructure from MRI data without imposing limiting models. Unlike other existing approaches, HistoStat will result in the generation of histology-like images containing similar information to that available in real histological data from a statistical standpoint. This will be done by employing the MR-scanner to measure a series of statistical descriptors (SDs) encoding the relative arrangements and shapes of tissue components. The acquisition of SDs will be based on different properties that are known to impact on MRI signal generation, such as diffusivities, relaxations, and magnetic susceptibilities. These SDs will be used to generate statistically accurate tissue reconstructions. The accuracy of the results will be guaranteed by the high-resolution information contained in the SDs, which would otherwise be impossible to obtain by standard MRI methods.

Early and differential diagnosis of diseases and syndromes are amongst society's biggest medical challenges. As our population continues to live longer into old age, the number of people suffering and living with medical conditions continues to grow too. For example, neurodegenerative diseases alone, such as epilepsy and dementia, have been estimated to cost the UK more than £23b per year and are expected to triple by 2030. For the same year, prostate cancer is predicted to become the most common cancer in the UK. Given the variety of causes and aetiologies of these conditions, reliable identification of the underlying pathologies is a requisite for therapy effectiveness. It is in the provision of differential and early diagnosis that the technologies enabled by this project can have a serious impact. By working towards technological innovations in non-invasive microstructure characterisation, HistoStat will foster advances aimed at several conditions linked to other organs, such as bone and liver. The development of a mechanistic framework linking measurements and tissue microstructure will therefore provide plenty of insights for early medical diagnosis, helping to reduce the economic burden associated with them. This is of great importance for delivering more efficient treatments to topical conditions affecting a large proportion of the society, as well as to reduce the costs of late-stage therapies.

In particular, the following sectors will benefit from the research plan:

1. The society as a whole, who will experience more effective medical treatments as a result of the microstructural information available through MRI.

2. Medical doctors and the public health system, counting with a novel tool to aid in the early diagnosis of diseases without the need of costly and potentially dangerous invasive procedures.

3. Drug developers (e.g. GlaxoSmithKline), who will benefit from new tissue biomarkers of efficacy, and improved targeting of drugs to tissue, using predictive models that account for microstructure.

4. The MR scanners industry (e.g. Siemens and Philips), who will be able to plan and design new prototype systems and/or sequences specifically engineered for maximising the sensitivity of the instruments to HistoStat's requirements. For example, it may be possible to design MR fingerprinting sequences for measuring the required statistical descriptors in a very reduced time.