Toward five‐part differential of leukocytes based on electrical impedances of single cells and neural network

Abstract

The five-part differential of leukocytes plays key roles in the diagnosis of a variety of diseases and is realized by optical examinations of single cells, which is prone to various artifacts due to chemical treatments. The classification of leukocytes based on electrical impedances without cell treatments has not been demonstrated because of limitations in approaches of impedance acquisition and data processing. In this study, based on treatment-free single-cell impedance profiles collected from impedance flow cytometry leveraging constriction microchannels, two types of neural pattern recognition were conducted for comparisons with the purpose of realizing the fivepart differential of leukocytes. In the first approach, 30 features from impedance profiles were defined manually and extracted automatically, and then a feedforward neural network was conducted, producing a classification accuracy of 84.9% in the fivepart leukocyte differential. In the second approach, a customized recurrent neural network was developed to process impedance profiles directly and based on deep learning, a classification accuracy of 97.5% in the five-part leukocyte differential was reported. These results validated the feasibility of the five-part leukocyte differential based on label-free impedance profiles of single cells and thus provide a new perspective of differentiating white blood cells based on impedance flow cytometry.

Electrical micro flow cytometry with LSTM and its application in leukocyte differential

“…Previously, based on constrictional microchannels with cross sectional areas smaller than leukocytes, both three-part and five-part leukocyte differentials were realized based on single-cell impedance only [39,40]. However, they can only process purified leukocyte subtypes rather than cell mixtures due to dimensional mismatches between constrictional microchannels and leukcoytes with significant variations in cell sizes.…”

“…Previously, constrictional microchannels were used for single-cell impedance measurements in which single cells traveled through microchannels with constricted cross-sectional areas and effectively blocked electric field lines, generating significant differences in impedance variations [37,38]. More specifically, based on constrictional microchannels, Chen et al realized three-part and five-part differential of leukocytes [39,40] while Wang et al classified three human breast cell lines [41]. In addition, mechanical and electrical properties of individual cells were also quantified based on constrictional microchannels, simultaneously [42,43].…”

Leukocyte differential based on an imaging and impedance flow cytometry of microfluidics coupled with deep neural networks

“…Within the developments of microfluidic impedance flow cytometry, impedance variations of traveling single cells can be significantly enhanced than conventional hematology analyzers since electrical lines can be better confined to penetrate single cells due to microfabricated detection channels, leading to improved accuracies of leukocyte differentials [12][13][14][15][16][17]. More specifically, classification accuracies of 93.5% [12] and 97.5% [13] for 3-part and 5-part leukocyte differentials were reported based on constrictional microchannels plus feedforward and recurrent neural networks.…”

Rapid assessment of susceptibility of bacteria and erythrocytes to antimicrobial peptides by single-cell impedance cytometry

“…Its merits with respect to current state-of-the-art AST methods are comprehensively discussed in (Spencer et al, 2020). The technique measures the electrical phenotype of individual biological cells and has been applied to mammalian cells, human pathogens, yeast cells, and plant cells (Daguerre et al, 2020;de Bruijn et al, 2021;Gökçe et al, 2021;Honrado et al, 2021b;Kruit et al, 2022;Petchakup et al, 2018;Tang et al, 2021;Wang et al, 2022). The sensitivity of the technique to alterations of cell size, membrane, and interior composition makes it particularly suitable for cell viability applications (De Ninno et al, 2020;Honrado et al, 2022Honrado et al, , 2021aZhong et al, 2021).…”