The DBSCAN clustering algorithm, an in-house method of unsupervised machine learning, was used to explore specific histotopographical features of the megakaryocytic lineage in the bone marrow biopsies of patients with JAK2 - or CALR -mutated essential thrombocythemia and prefibrotic primary myelofibrosis. Ninety-five bone marrow biopsies of patients with essential thrombocythemia and primary myelofibrosis were investigated to assess the histotopography of megakaryocytes, specifically, the mean number of megakaryocytes in one cluster, as well as the mean number of clusters and megakaryocytes per 1 mm 2 of section area. The logistic regression model was statistically significant: χ 2 = 14.703, p = 0.023, Nagelkerke R 2 = 19.6%. Analysis of the histotopographical features of megakaryocytes allowed correct differentiation between essential thrombocythemia and primary myelofibrosis in 71.6% cases. The differences in the histotopographical features of megakaryocytes in the bone marrow of with JAK2 - or CALR -mutated essential thrombocythemia and primary myelofibrosis revealed by the DBSCAN clustering algorithm indicate that a relationship exists between the disease and the pattern of megakaryocytic lineage development that can be used to create a logistic regression model for differentiating these diseases.

中文翻译：

---

原发性骨髓纤维化和原发性血小板增多症的形态学鉴别诊断与髓样组织巨核细胞谱系的计算机聚类分析

DBSCAN 聚类算法是一种内部无监督机器学习方法，用于探索 JAK2 或 CALR 突变的原发性血小板增多症和前纤维化原发性骨髓纤维化患者骨髓活检中巨核细胞谱系的特定组织学特征。研究了 95 份原发性血小板增多症和原发性骨髓纤维化患者的骨髓活检，以评估巨核细胞的组织学形态，特别是一个簇中的平均巨核细胞数，以及每 1 mm 2 切片的平均簇数和巨核细胞数区域。逻辑回归模型具有统计学意义：χ 2 = 14.703，p = 0.023，Nagelkerke R 2 = 19.6%。71.6% 的巨核细胞组织学特征分析允许正确区分原发性血小板增多症和原发性骨髓纤维化。DBSCAN 聚类算法揭示的 JAK2 或 CALR 突变的原发性血小板增多症和原发性骨髓纤维化的骨髓中巨核细胞组织学特征的差异表明，该疾病与可使用的巨核细胞谱系发育模式之间存在关系创建一个逻辑回归模型来区分这些疾病。