Development of an Imaging and Impedance Flow Cytometer Based on a Constriction Microchannel and Deep Neural Pattern Recognition

Abstract

This article presents an imaging and impedance flow cytometer based on a constriction microchannel with corresponding cell-type classification based on deep neural pattern recognition. When an incoming cell reached the entrance of the constriction microchannel, the image of the nucleus labeled with fluorescence was captured by a high-speed camera without the concern of losing focus, whereas when the cell deformed through the constriction microchannel by effectively blocking electrical lines, large impedance variations were sampled by an impedance analyzer. Six key biostructural and bioelectrical parameters (e.g., cell diameter, nuclear roundness, and cytoplasmic conductivity) from thousands of single cells were extracted, producing a successful rate of 88.3% in classifying A549 versus Jurkat versus K562 cells based on the feedforward neural network. In addition, multilayer neural networks of deep learning (e.g., VGG16 of CNN and LSTM of RNN) were also used to process fluorescent images and impedance profiles, producing an almost 100% successful rate in cell-type classification. In summary, the microfluidic flow cytometer reported in this article could characterize single-cell biostructural and bioelectrical properties in a high throughput manner, realizing high successful rates of cell-type classification.