Abstract
BACKGROUND: Automated quantitation of marrow fibrosis promises to improve fibrosis assessment in myeloproliferative neoplasms (MPNs). However, analysis of reticulin-stained images is complicated by technical challenges within laboratories and variability between institutions. METHODS: We have developed a machine learning model that can quantitatively assess fibrosis directly from H&E-stained bone marrow trephine tissue sections. RESULTS: Our haematoxylin and eosin (H&E)-based fibrosis quantitation model demonstrates comparable performance to an existing reticulin-stained model (Continuous Indexing of Fibrosis [CIF]) while benefitting from the improved tissue retention and staining characteristics of H&E-stained sections. CONCLUSIONS: H&E-derived quantitative marrow fibrosis has potential to augment routine practice and clinical trials while supporting the emerging field of spatial multi-omic analysis.