Automation Of Flow Cytometric Data Analysis For Acute Lymphocyte Leukemia Based On Machine Learning

Object: In view of the complexity and low efficiency of traditional manual analysis of flow cytometry(FCM)data,unsupervised clustering method and supervised classification method were proposed for FCM data analysis to simulate the whole process of manual analysis and solve the problem of disconnection of each part of the automatic analysis method of FCM data,so as to realize the full automation of ALL bone marrow flow cytometry data analysis.To provide practical analytical tools for disease diagnosis.Methods: The data were derived from the data of 528 bone marrow test cases in flow laboratory of Xinjiang Uygur Autonomous Region People’s Hospital in 2021.The original data of FCM was preprocessed by compensation,conversion and deadhesion.The unsupervised clustering method was used to perform cluster analysis on the preprocessed data,and the supervised classification model was trained to divide all cell subsets into 20 categories by using the distribution law of the central location(macro cell)of the generated cell subsets.Finally,the 20 categories were manually recognized and labeled to obtain 9known cell types,and a mapping of 20 to 9 was established.Any cell subpopulation belongs to one of 20 categories and is mapped to the cell type it belongs to.Then,according to the subgroup labeling results,the composition ratio of each cell subgroup,the positive expression rate of cell surface or intracellular antigens,and the abnormal related characteristics in each subgroup were extracted.Refer to the diagnostic criteria specified in the guidelines to prepare the disease diagnosis procedures.According to the feature extraction results,diagnosis and classification were carried out.Finally,sensitivity,specificity,Yoden index and area under the curve(AUC)were used to evaluate the automated diagnosis results according to the gold standard of manual analysis by flow laboratory experts.The results of automated immune typing were evaluated using error rate,Rand Index and F-Measure values.Results: After checking the process files of desglutination,cell clustering,subgroup classification and labeling and the visualization results,it was found that there was no abnormality in the pretreatment part,cell clustering,subgroup classification and labeling part,which was basically consistent with manual analysis.In the automatic diagnosis results,the sensitivity and specificity of the training samples were 87%,97%,0.84 and 0.92(0.87,0.97).The error rate for immune typing was 8%,the RAND index was 0.89,and the F-Measure was 0.92.In the test samples,the sensitivity and specificity of disease diagnosis were 82%,97%,0.79,and 0.89(0.82,0.95).The error rate for immune typing was 5%,the RAND index was 0.91,and the F-Measure was 0.95.Conclusion: The automatic analysis method proposed in this study solves the problem of disconnection of each part in the analysis of FCM data.It is true in the flow diagnosis of ALL,and provides the middle process of diagnosis,which has certain clinical application value for the automatic application of FCM data diagnosis.