Abstract
INTRODUCTION: Primary myelofibrosis (MF) is characterized by MF, splenomegaly, and extramedullary hematopoiesis. MF initially presents with reticular fibers (RFs) and progresses to increased collagen fiber deposition in the advanced stages. Although recent clinical trials have adopted MF improvement as an evaluation criterion, current diagnostic methods rely primarily on qualitative assessments based on the presence or absence of reticular and collagen fibers. Therefore, detecting subtle changes during the early stages of MF can be challenging. METHODS: We developed a novel deep learning-based method for quantitatively evaluating MF by measuring RFs as an indicator. Unlike collagen fibers, RFs are detectable from the early stages of MF and increase as the disease progresses. Moreover, based on the hypothesis that splenic fibrosis progresses in parallel with MF, we applied this method to evaluate RFs in the spleen. Using these methods, we analyzed temporal changes in fibrosis, splenomegaly, and hematopoietic stem cell dynamics over time in two MF models: a drug-induced fibrosis model using romiplostim and a Jak2V617F gene-transformed mouse. Additionally, we examined correlations between our quantitative fibrosis measurements and clinical data, including MF grade and genetic mutations. RESULTS: Our findings revealed that in Jak2V617F gene-transformed mice, splenomegaly and extramedullary hematopoiesis in the spleen occurred earlier than MF. Furthermore, the quantitative fibrosis method significantly correlated with MF grade in patients with myeloproliferative neoplasms and the JAK2V617F mutant allele burden. CONCLUSION: Our novel deep learning-based method successfully captured temporal changes in bone marrow and spleen fibrosis and shows potential for clinical application. TRIAL REGISTRATION: The authors have confirmed clinical trial registration is not needed for this submission.