Subtyping Of Label-free Leukemia Cells Based On Deep Learning And Two-dimensional Light Scattering Flow Cytometry

Leukemia is one of the most common hematologic malignancies in the world and the most common cancer among children and adolescents.Different subtypes of leukemia have different therapeutic and prognostic strategies,for example,the targeted drugs of Acute Lymphocytic Leukemia(ALL)are different from those of Acute Myeloid Leukemia(AML).Tyrosine kinase inhibitors(TKIs)are the common targeted drugs of ALL,while IDH inhibitors and gemtuzumab ozogamicin(Mylotarg)are often used in AML targeted therapies.Tissue biochemical staining and immunohistochemical examination are used in the diagnosis of leukemia.Although staining and labeling methods have better accuracy,they also have the limitations of professional operation and high cost.Two-dimensional(2D)light scattering technology has the advantages of rapid,accurate and nondestructive in cell detection,which provides a new idea for the realization of labeling free detection and typing of leukemia cells.As an automated image analysis method,deep learning is a potential tool to fully mine the effective information of cell light scattering images and realize high-precision analysis.Therefore,the development of 2D light scattering cell detection technology based on deep learning has a strong research significance and application prospect in the early diagnosis of leukemia.In this paper,we present a deep learning based 2D light scattering flow cytometry and explore its application in the classification of leukemia cells.In the single cell image acquisition module,the 2D light scattering technology,microfluidic technology and lightsheet illumination technology were combined to build a light scattering imaging flow platform,which was used in the light scattering imaging experiments of leukemia single cells and normal single cells.In image analysis and classification module,this paper proposes a new algorithm named ISSC-Net(Inception V3-SIFT-Scattering Net)for 2D light scattering image analysis.The paper focuses on the principle of ISSC-Net,and tests the performance of ISSCNet,which proves that ISSC-Net can fully mine the effective information of 2D light scattering patterns of single cells,and is a potential method for effective classification of different types of single cells.ISSC-Net based 2D light scattering flow cytometry is used to further classify the 2D light scattering images of leukemia cell line cells and normal cells,and classify the 2D light scattering images of four leukemia subtype cell line cells.In summary,ISSC-Net is proposed in this paper,and it is combined with light-sheet based 2D light scattering microfluidic cytometer.This article focuses on the application of deep learning based 2D light scattering flow cytometry in leukemia subtyping,including the classification of leukemia cell line cells and normal cells,the classification of four leukemia subtypes cell line cells,and the classification of human T cell and human B cell acute lymphocytic leukemia cell line cells.The results of this study indicate that the twodimensional light scattering flow cytometry based on deep learning has an excellent performance in the classification of leukemia cell lines,and is expected to be used in clinical screening of label-free leukemia cells.