Neural network for blood cell classification in a holographic microscopy system

Modern clinical laboratories are equipped with high-throughput flow cytometers for fast and accurate cell sorting. Most cytometers use selective biomarkers which often induce significant changes in the cell morphology, sometimes leading to cell death. However, for purposes like cell imaging there exist label-free techniques, for example digital inline holographic microscopy. Yet the image reconstruction algorithms needed to analyze the images do not scale up easily to large numbers of cells. We suggest an integrated, optical neural network to deal with the high-speed image classification with the promise of dense integration for ultrafast, cell sorting. A ternary classification task, distinguishing between monocytes, granulocytes, and lymphocytes resulted in 89% accuracy.