Generalizable deep learning approach for 3D particle imaging using holographic microscopy (HM).

Area of Science:

• Optics and Photonics
• Biomedical Imaging
• Artificial Intelligence in Diagnostics

Background:

• Holographic microscopy offers label-free particle diagnostics but is hindered by processing methods lacking generalizability.
• Current techniques struggle with diverse particle characteristics and environmental conditions, limiting widespread adoption.

Purpose of the Study:

• To introduce a novel deep learning architecture for holographic microscopy.
• To enhance the generalizability and processing speed of 3D particle analysis.
• To overcome limitations of existing methods in handling complex and varied particle samples.

Main Methods:

• Developed a deep learning architecture inspired by human perception of longitudinal variations in diffracted patterns.
• Trained the model using minimal synthetic and real hologram data of simple particles.
• Employed a novel approach to analyze three-dimensional (3D) particle information from holograms.

Main Results:

• Achieved orders of magnitude improvement in processing speed for holographic data.
• Demonstrated highly generalizable analysis across diverse and challenging cases.
• Showcased exceptional performance with high particle concentrations, noise, varied sizes, complex shapes, and optical properties, surpassing training data diversity.

Conclusions:

• The proposed deep learning method significantly advances holographic microscopy for particle diagnostics.
• The architecture provides a robust and efficient solution for analyzing complex particle systems.
• This approach holds promise for broader applications in label-free particle characterization and analysis.