Microfluidic Device for Flow-Based Immune Cell Quantification in Whole Blood Using Machine Learning

This paper presents an innovative approach for flow-based quantification of immune cells in whole blood using microfluidics and machine learning. Target immune cells were labeled with antibody-coated microbeads and flowed inside a microfluidic device, and a convolutional neural network (CNN)-based object detection algorithm was utilized for the detection of bead-labeled cells. The detection range of this platform was evaluated by analyzing blood samples spiked with 10 mu m-diameter polystyrene beads, which could be accurately quantified over a wide range of concentrations from 300 to 3,500 beads/mu L. Proof-of-concept was demonstrated by quantifying CD4(+) T cells in three blood samples from human volunteers, which offered similar accuracy as cell counts determined by flow cytometry while being at least 1.5-fold faster and simpler to perform.