Development and application of deep learning and spatial statistics within 3D bone marrow imaging

Abstract

The bone marrow is a highly specialised organ, responsible for the formation of blood cells. Despite 50 years of research, the spatial organisation of the bone marrow remains an area full of controversy and contradiction. One reason for this is that imaging of bone marrow tissue is notoriously difficult. Secondly, efficient methodologies to fully extract and analyse large datasets remain the Achilles heels of imaging-based research. In this thesis I present a pipeline for generating 3D bone marrow images followed by the large-scale data extraction and spatial statistical analysis of the resulting data. Using these techniques, in the context of 3D imaging, I am able to identify and classify the location of hundreds of thousands of cells within various bone marrow samples. I then introduce a series of statistical techniques tailored to work with spatial data, resulting in a 3D statistical map of the tissue from which multi-cellular interactions can be clearly understood. As an illustration of the power of this new approach, I apply this pipeline to diseased samples of bone marrow with a particular focus on leukaemia and its interactions with CD8+ T cells. In so doing I show that this novel pipeline can be used to unravel complex multi-cellular interactions and assist researchers in understanding the processes taking place within the bone marrow.