Myeloproliferative neoplasms (MPNs) are blood cancers characterised by well-defined driver mutations. Some patients develop bone marrow (BM) fibrosis, which is associated with a poor clinical prognosis. Complex interactions between immune, stromal and haematopoietic stem and progenitor cells in the BM are implicated in the development of fibrosis. However, our understanding of how the BM microenvironment underpins this process is limited by the lack of objective, quantitative descriptions of BM topology. We use spatial transcriptomic (ST) analysis to explore the BM microenvironment in MPN. We have developed a workflow that integrates H&E-based annotation and AI-based image analysis with ST data to provide high-resolution, whole section profiling of the human BM. We identify spatially-restricted patterns of haematopoiesis and cellular neighbourhoods that define both the normal BM, and the perturbation in MPN. Using an AI-based fibrosis detection algorithm, we identify microenvironmental features associated with regions of fibrosis. We then develop novel AI-based approaches to identify spatial signatures that characterise the normal BM, and which capture cohort and sample-level microenvironmental heterogeneity. Our observations open the door to a new era of spatial biomarker discovery.