Blood cell characterization based on deep learning and diffraction phase microscopy.

Quantitative characterization of blood cells is very important for medical diagnosis and life science research. This study presents a novel label-free and high-throughput approach for analyzing flowing blood cells in a microfluidic channel by combining deep learning with partially coherent illumination-based diffraction phase microscopy (PCI-DPM). Diluted blood was injected into a microchannel and imaged with PCI-DPM, yielding high-contrast phase images for the blood cells. A modified YOLOv5-s neural network was used to perform automatic identification and analysis of blood cells in the phase image of PCI-DPM, revealing the diameter, dry mass, volumetric density, and speed of blood cells precisely. The proposed approach allows quick and accurate assessment of blood samples, providing valuable insights for patients in different pathological states, such as hematological disorders and infections. [ABSTRACT FROM AUTHOR]