Artificial intelligence of digital morphology analyzers improves the efficiency of manual leukocyte differentiation of peripheral blood

• Published in: BMC medical informatics and decision making Vol. 23; no. 1; p. 50
• Main Authors: Xing, Ying, Liu, Xuekai, Dai, Juhua, Ge, Xiaoxing, Wang, Qingchen, Hu, Ziyu, Wu, Zhicheng, Zeng, Xuehui, Xu, Dan, Qu, Chenxue
• Format: Journal Article
• Language: English
• Published: England BioMed Central Ltd 29-03-2023; BioMed Central; BMC
• Subjects: Accuracy; Algorithms; Analysis; Analytical instruments; Analyzers; Artificial Intelligence; Blood; Cell Differentiation; Cells; Classification; Cytology; Differentiation; Digital imaging; Digital morphology analyzer; Evaluation; Health informatics; Hematology; Humans; Identification; Innovations; Laboratories; Leukocyte differentiation; Leukocytes; Lymphocytes; Morphology; Neutrophils; Peripheral blood; Professionals; Sensitivity; Sensitivity and Specificity; Specificity; Standardization; Technologists; China; Beijing China; Accuracy; Sensitivity; Specificity; Digital morphology analyzer; Peripheral blood smear; Leukocyte differentiation; Artificial intelligence; Specimen turnaround time
• ISSN: 1472-6947; 1472-6947
• DOI: 10.1186/s12911-023-02153-z

Morphological identification of peripheral leukocytes is a complex and time-consuming task, having especially high requirements for personnel expertise. This study is to investigate the role of artificial intelligence (AI) in assisting the manual leukocyte differentiation of peripheral blood.
A total of 102 blood samples that triggered the review rules of hematology analyzers were enrolled. The peripheral blood smears were prepared and analyzed by Mindray MC-100i digital morphology analyzers. Two hundreds leukocytes were located and their cell images were collected. Two senior technologists labeled all cells to form standard answers. Afterward, the digital morphology analyzer unitized AI to pre-classify all cells. Ten junior and intermediate technologists were selected to review the cells with the AI pre-classification, yielding the AI-assisted classifications. Then the cell images were shuffled and re-classified without AI. The accuracy, sensitivity and specificity of the leukocyte differentiation with or without AI assistance were analyzed and compared. The time required for classification by each person was recorded.
For junior technologists, the accuracy of normal and abnormal leukocyte differentiation increased by 4.79% and 15.16% with the assistance of AI. And for intermediate technologists, the accuracy increased by 7.40% and 14.54% for normal and abnormal leukocyte differentiation, respectively. The sensitivity and specificity also significantly increased with the help of AI. In addition, the average time for each individual to classify each blood smear was shortened by 215 s with AI.
AI can assist laboratory technologists in the morphological differentiation of leukocytes. In particular, it can improve the sensitivity of abnormal leukocyte differentiation and lower the risk of missing detection of abnormal WBCs.