Digital holographic microscopy can reconstruct the images of 3D samples from a single hologram at a fraction of the size and cost of a standard bright-field microscope. It has enabled a plethora of hand-held holographic devices for biomedical diagnostics. Despite these benefits, holographic microscope images generally suffer from light interference-related spatial artifacts, which can limit the achievable contrast in the reconstructed hologram. Researchers have developed a novel artificial neural network-based method, bright-field holography, to overcome these limitations of holographic 3D imaging.
Bright-field holography combines image contrast (microscopy) and snapshots volumetric imaging (holography). The researchers trained a deep neural network using co-registered pairs of holograms and their corresponding bright-field microscope images.
Although the training of such a neural network takes ~40 hours, the network remains fixed after the training. Subsequently, it can rapidly create its output image within a second for a hologram with millions of pixels.
Bright-field holography bridges the contrast gap between traditional hologram reconstruction methods and high-end bright-field microscopes while rapidly removing the need for complex hardware and mechanical scanning to image sample volumes. Rapid volumetric imaging of dynamic events within large volumes is one application that will immediately benefit from this technology, opening up new avenues for significantly advancing high-throughput imaging of liquid samples through deep learning.