Cross-modality deep learning brings bright-field microscopy contrast to holography

TL;DR

Holographic images often contain speckle and artifacts that limit their contrast relative to bright-field microscopy. The paper uses a GAN to transform back-propagated holograms into bright-field-equivalent images, reporting artifact-reduced outputs at multiple sample depths.

• Problem

  Holographic images can contain speckle and other artifacts, limiting their ability to match the artifact-free contrast of bright-field microscopy.

  Source: p. 1

• Method

  A GAN transforms digitally back-propagated holograms at selected depths into images equivalent to bright-field images acquired at those depths.

  Source: p. 1

• Results

  For pollen samples, network outputs were free of observed artifacts, matched bright-field contrast, and reproduced color; 3D tests also matched contrast and depth-of-field across sample depths.

  Source: p. 2

• Takeaways & Limitations

  The approach replaces mechanical scanning of a volumetric sample with digital propagation and transformation of fields into bright-field-equivalent images.

  Source: p. 2