Rapid screening of acute promyelocytic leukaemia in daily batch specimens: A novel artificial intelligence‐enabled approach to bone marrow morphology

作者：Y. J. Xiao, Huang Zheng, Jun Wu, Yan Zhang, Yuanyuan Yang, Chunping Xu, Fangyu Guo, Xiong Ni, Xinhua Hu, Jianmin Yang, Yang Song, Hui Cheng, Gusheng Tang · 发表于：Clinical and Translational Medicine · 年份：2024 · DOI：10.1002/ctm2.1783 · 被引用次数：6 · 研究领域：Retinoids in leukemia and cellular processes、Chemokine receptors and signaling、interferon and immune responses

Dear Editor, Acute promyelocytic leukaemia (APL) is a malignant haematological disease characterised by abnormal proliferation of promyelocytes and represents a distinct subtype of acute myeloid leukaemia (AML), constituting about 15% of AML cases.1 According to real-world data, the early mortality rate for APL varies between 17% and 40%.2-4 In order to avoid premature deaths, rapid and accurate diagnosis is crucial for early identification and initiation of treatment with all-trans retinoic acid (ATRA) and arsenic trioxide (ATO) or chemotherapy.5 Although definitive diagnosis of APL requires confirmation of chromosome t (15;17) or PML::RARA fusion gene,6 cytomorphology remains the fastest technique for initial diagnosis. Manual microscopic examination of cytomorphology, however, often demonstrates significant inter-observer variability, potentially resulting in missed or misdiagnosis. Developing an automated, accurate and universally applicable intelligent diagnosis system for APL would hold significant clinical importance by mitigating intra- and inter-observer variability and enabling early diagnosis. This study investigated the enhanced potential of artificial intelligence (AI)-assisted morphology for the rapid screening and diagnosis of APL in daily batch specimens. The CELLSEE we proposed, an AI-powered APL morphological diagnostic system featuring a convolutional neural network (CNN) with embedded attention mechanisms, which would recognise APL at 10× and 100× magnific...