Automatic detection and counting of phase objects in raw holograms of digital holographic microscopy via deep learning

Abstract

A method to automatically detect and count phase objects in off-axis digital holographic microcopy (DHM) is presented. While traditional procedures to detect and count microscopic objects in DHM require a reconstruction stage, our proposal can detect and count phase objects in the raw holograms with no reconstruction by using a convolutional neural network (CNN). In comparison with the visual counting executed over fully processed DHM holograms, which includes a phase unwrapping stage, the accuracy of our deep learning-based method can reach 100% in samples with up to 2.1 × 103 cells per mm2. Although this proposal was validated in the counting of red blood cells, it can be straightforwardly implemented to detect and count other microscopic phase objects in raw DHM holograms after proper training of the CNN.