Performance evaluation of AI-enabled blood cell morphology system for peripheral blood smear and application in grading screening network of primary medical care system
10.13602/j.cnki.jcls.2025.04.02
- VernacularTitle:外周血人工智能阅片系统性能评价及在基层医疗分级诊疗网络中的应用
- Author:
Xiaobing SUN
1
;
Gusheng TANG
;
Kaiying YUAN
;
Duanqin DIAO
;
Jun HU
;
Xiaoyuan SHI
;
Hao YUAN
;
Anmei WANG
;
Yan FANG
;
Liqin JIANG
;
Xueliang QIN
;
Chun XU
;
Qi HOU
;
Jiong WU
Author Information
1. 上海嘉会国际医院检验科,上海 200233
- Publication Type:Journal Article
- Keywords:
artificial intelligence;
digital cell morphology;
hierarchical diagnosis;
multicenter;
peripheral blood smear;
hematological disease
- From:
Chinese Journal of Clinical Laboratory Science
2025;43(4):246-252
- CountryChina
- Language:Chinese
-
Abstract:
Objective To evaluate the recognition capability of AI-enabled Cellsee CS-BM1 automatic cell morphology analyzer for pe-ripheral blood smears and its roles in assisting manual classification,and explore the application value of AI system in the diagnosis network of tiered primary medical units.Methods The blood samples which triggered the re-examination rules were collected from six primary medical units,including the Laboratory Department of Shanghai Jiahui International Hospital,and so on,from March to No-vember 2023.The smears of peripheral blood were prepared and AI analyzer was used for pre-classification to evaluate its recognition performance in identifying the samples with abnormal WBC and RBC.The sensitivity,specificity,and accuracy of WBC classification by six junior and intermediate technicians,both with and without AI assistance,were analyzed.Additionally,the roles of the AI system in tiered diagnosis of primary medical units were also evaluated.Results The sensitivity,specificity,and accuracy of AI system in recognizing malignant primitive cells were 92.86％,95.16％,and 95.10％,respectively.The sensitivities of AI system in recognizing immature granulocytes,reactive lymphocytes,and nucleated RBCs were all greater than 90％.The sensitivity of AI system in identif-ying abnormal morphology of RBCs reached 99.59％,along with rapid quantitative analysis for various anomalous types of RBCs.In AI-assisted mode,the sensitivity of recognition for all cell types was improved to varying degrees by junior and intermediate technicians,and the sensitivity for recognizing malignant primitive cells,reactive lymphocytes,and immature granulocytes increased to 58.24％,53.39％,and 62.37％for junior technicians,and to 92.06％,83.24％,and 83.12％for intermediate technicians,respectively.The improvements for junior technicians were particularly significant,with increases of 12.46％,10.61％,and 3.71％for each cell type,respectively.Both groups achieved higher specificity and accuracy.Through AI pre-classification and manual review,a variety of pe-ripheral blood cell-related diseases were accurately diagnosed in the tiered healthcare practice of primary medical units,including 339 cases(11.13％)of red blood cell diseases,5 cases(0.16％)of platelet diseases,2 343 cases(76.90％)of infection-related disea-ses,and 28 cases(0.92％)of malignant hematological diseases.In addition,332 cases(10.90％)which lacked an obvious related cause or required further examinations were identified as well.Conclusion AI pre-classification has demonstrated strong cell recogni-tion capabilities and may assist technicians in improving the sensitivity,specificity,and accuracy of blood cell classification.AI could en-hance the disease-screening capabilities in the tiered diagnosis network of primary medical units,presenting a broad application prospect.
