Artificial intelligence recognition of bone marrow cells can be applied to diagnosis of minimal residual disease in acute leukemia
10.3760/cma.j.cn114452-20220928-00562
- VernacularTitle:人工智能识别骨髓细胞可应用于诊断急性白血病微小残留病
- Author:
Siheng LIU
1
;
Jia LI
;
Wuchen YANG
;
Luo ZHAO
;
Xiangui PENG
Author Information
1. 陆军军医大学附属新桥医院血液病医学中心，重庆 400037
- Keywords:
Leukemia;
Minimal residual disease;
Bone marrow cells;
Artificial intelligence
- From:
Chinese Journal of Laboratory Medicine
2023;46(3):280-285
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the diagnostic value and problems of artificial intelligence (AI) bone marrow cell recognition technology in the detection of minimal residual disease (MRD) of leukemia.Methods:A total of 65 cases with minimal residual disease of leukemia confirmed by flow cytometry from the Hematology Medical Center of Xinqiao Hospital affiliated to the Army Medical University (AMMU) from November 1 to December 31, 2020 were collected. The bone marrow Wright′s staining smears were obtained, and all bone marrow smears were scanned and classified automatically without artificial intervention by the analysis system based on Artificial Intelligence platform (morphogo). AI-MRD was defined to positive when the proportion of primary cells was more than 3%. According to the number of AI automatic recognition cells, the cases were divided into 18 cases of less than 500 (L500), 35 cases of 500 to 1900 (between 500 and 1900, B1900), and 12 cases of more than 1900 (M1900), no overlap or omission between groups. Kappa consistency test was performed on the results of artificial intelligence test and the results of flow cytometry for minimal residual disease of leukemia (MFC-MRD) in each group. The receiver operating characteristic curve (ROC) of the artificial intelligence test results of each group of patients was drawn based on the MFC-MRD results, and the sensitivity, specificity and accuracy of the area under the curve (AUC) value and AI results were calculated.Results:After grouping according to the number of cells automatically recognized by AI, the detection results of L500 group were MFC-MRD+/AI-MRD+7 cases, MFC-MRD+/AI-MFC-2 cases, MFC-MRD-/AI-MRD+6 cases, MFC-MRD-/AI-MRD-3 cases; In B1900 group, MFC-MRD+/AI-MRD+13 cases, MFC-MRD+/AI-MFC-6 cases, MFC-MRD-/AI-MRD+6 cases, MFC-MRD-/AI-MRD-10 cases; The results of M1900 group were MFC-MRD+/AI-MRD+5 cases, MFC-MRD+/AI-MFC-0 cases, MFC-MRD-/AI-MRD+1 case, MFC-MRD-/AI-MRD-6 cases. Taking MFC-MRD as the determination standard, the sensitivity of AI-MRD detection in L500 group, B1900 group and M1900 group was 53.8%, 68.4% and 83.3%, the specificity was 60%, 62.5% and 100%, the accuracy was 55.6%, 65.7% and 91.7%, and the AUC value were 0.568 P=0.654, 0.678 P=0.069,1.000 P=0.000. Conclusions:This study preliminarily explored the diagnostic value and problems of AI bone marrow cell recognition in the detection of minimal residual disease of leukemia. It was confirmed that when 3% of the proportion of blasts in AI cell classification is set>3% as the positive threshold of AI-MRD, the consistency between AI and MFC-MRD detection increases with the increase of the number of cells recognized by AI.
