Using neural networks for high-speed blood cell classification in a holographic-microscopy flow-cytometry system

Abstract

High-throughput cell sorting with flow cytometers is an important tool in modern clinical cell studies. Most cytometers use biomarkers that selectively bind to the cell, but induce significant changes in morphology and inner cell processes leading sometimes to its death. This makes label-based cell sorting schemes unsuitable for further investigation. We propose a label-free technique that uses a digital inline holographic microscopy for cell imaging and an integrated, optical neural network for high-speed classification. The perspective of dense integration makes it attractive to ultrafast, large-scale cell sorting. Network simulations for a ternary classification task (monocytes/granulocytes/lymphocytes) resulted in 89% accuracy.

Fast particle characterization using digital holography and neural networks

“…Our research was highly motivated by these experiments because its high-speed, on-line image classification requirements are without doubt. Indeed we proved recently [12] that our method successfully applies to the classification of experimentally recorded inline holograms of white blood cells in a microfluidic channel. In that sense the method still allows for parallel computation of many spatially isolated particles but excludes currently the possibility of characterizing and tracking multiple particles that are close to each other.…”

“…We emphasize that at those optical depths neither the Fraunhofer nor the Fresnel approximation holds. We chose fixed values for the optical depth as it agrees with the fact that the depth is well controlled in microfluidic flow channels, which triggered our interest in this research topic (see [11,12]). Nonetheless our method is not restricted to any specific choice of the optical depth, which is often unknown.…”

Neuroblastoma Cells Classification Through Learning Approaches by Direct Analysis of Digital Holograms

“…Actually, the concept of direct classification without image reconstruction has been also investigated in other kinds of optical imaging systems and it can be implemented in diffractive imaging [24], scattering imaging [25] and single-pixel imaging [26]. In DH, the possibility to solve classification problems by using digital holograms as input of a learning-based classifier has been recently explored [27], [28], [29]. In particular, in ref.…”

“…[27], the method named "ensemble deep learning invariant hologram classification" has been demonstrated for the classification of digital holograms of macroscopic objects, specifically handwritten digits. Instead, for the classification of holograms recorded in microscope configuration, the paper in [28] demonstrates, but only through a numerical simulations, the possibility to use neural networks in holographic flow cytometers for the classification of white blood cells. The work in [29] focuses on the classification of individual cells according to the number of cell-bound microbeads.…”

Cell detection by surface imprinted polymers SIPs: A study to unravel the recognition mechanisms

“…Similarly, the prevention of large-scale outbreaks of infectious diseases does not only depend on personal hygiene, but also on the detection of pathogens such as bacteria and viruses in water, food and air. The most important cell detection techniques are based on polymerase chain reaction (PCR), oligonucleotide DNA microarrays, microscopy, flow cytometry, fluorescent in situ hybridization, and pyrosequencing [1][2][3][4][5][6].…”