Evaluation of the application of AI morphological assisted analysis system in the pre-classification of blood cells of AML-MR patients
10.3760/cma.j.cn114452-20240709-00361
- VernacularTitle:AI形态学辅助分析系统在AML-MR患者血细胞预分类中的应用评价
- Author:
Rui ZHENG
1
;
Zhiying SHEN
1
;
Ziyi YAN
1
;
Yini YU
1
;
Jun GAN
1
;
Baoguo CHEN
1
Author Information
1. 浙江省台州医院中心实验室，临海 317000
- Publication Type:Journal Article
- Keywords:
Cytology;
Acute myeloid leukemia, myelodysplasia-related;
Blood smear;
Bone marrow smear;
Artificial intelligence;
Dysplasia
- From:
Chinese Journal of Laboratory Medicine
2025;48(3):357-363
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the application value of the artificial intelligence (AI) morphological assisted analysis system in the pre-classification of blood cells in patients with acute myeloid leukemia, myelodysplasia-related (AML-MR).Methods:A retrospective analysis was conducted on the bone marrow and peripheral blood cell morphology of patients initially diagnosed with AML-MR at Taizhou Hospital in Zhejiang Province from September 1, 2022, to December 31, 2023. A total of 44 patients, including 25 males and 19 females, with a median age of 71 (63.5, 75.3) years. Bone marrow and peripheral blood morphology were examined using the Morphogo cell morphology assisted analysis system, with the artificial classification results serving as the gold standard. A confusion matrix was constructed to evaluate the precision, sensitivity, and specificity of the AI system in identifying various cell types in bone marrow and peripheral blood for AML-MR diagnosis. The impact of dysplastic hematopoiesis on AI pre-classification was analyzed by comparing AI and manual classification results.Results:The AI system completed the pre-classification of 44 bone marrow smears and 42 corresponding peripheral blood smears from AML-MR patients. For bone marrow smears, the precision, sensitivity, and specificity of AI in pre-classifying blast cells were 85.78%, 91.01%, and 94.58%, respectively. For peripheral blood smears, these values were 87.11%, 87.05%, and 98.29%, respectively. The precision and sensitivity of AI in pre-classifying promyelocytes were 54.26% and 46.93%, respectively, while for monocytes, they were 58.16% and 68.34%, both lower than those for blast cells. The precision and sensitivity of AI in identifying myelocytes and metamyelocytes also decreased (77.47%, 66.25% and 81.91%, 63.29%, respectively). The precision and sensitivity of AI in pre-classifying erythroblasts/proerythroblasts (67.71%, 69.89%) were lower than those for polychromatic and orthochromatic normoblasts (83.43%, 85.53% and 92.97%, 86.96%, respectively). The confusion matrix and comparative analysis of AI and manual classification indicated that the decline in AI pre-classification precision and sensitivity was due to frequent misclassification between promonocytes and monocytes, as well as between monocytes and promyelocytes. Additionally, this decline is associated with dysplasia. However, the impact of dysplasia on the AI pre-classification of mature-stage granulocytes was minimal.Conclusion:The AI system demonstrated high precision, sensitivity, and specificity in pre-classifying blast cells in bone marrow and peripheral blood smears from AML-MR patients. The AI-assisted morphological analysis system can be effectively utilized for the pre-classification of blood cells in AML-MR patients.
