Deep learning enables real-time 3D quantitative phase imaging

Machine Learning


The new technique, developed by University of California scientists, offers a practical way to avoid the bottlenecks caused by existing 3D QPI techniques, which are more laborious and computationally intensive. The new research, using a wavelength-multiplexed diffractive optical processor, presents a state-of-the-art method for 3D QPI, Advanced Photonics.

Artistic depiction of a wavelength-multiplexed diffractive optical processor for 3D quantitative phase imaging.
Artistic depiction of a wavelength-multiplexed diffractive optical processor for 3D quantitative phase imaging. Image courtesy of the Ozcan Lab at the University of California.

Light waves experience a time delay as they pass through a medium. This delay can reveal important details about the underlying structural and compositional properties. A cutting-edge optical technique called Quantitative Phase Imaging (QPI) displays the change in optical path length as light passes through materials, biological samples, and other transparent structures.

In contrast to traditional imaging techniques that rely on stains or labels, QPI produces high-contrast images that enable non-invasive examinations essential to fields such as biology, materials science and engineering. These images allow researchers to observe and quantify phase variations.

The UCLA team developed a wavelength-multiple diffractive optical processor that optically converts the phase distributions of many 2D objects at different axial positions into intensity patterns that are each stored in a different wavelength channel.

This design uses an intensity-only image sensor to obtain quantitative phase images of input objects in different axial planes, eliminating the need for digital phase retrieval techniques.

“We are excited by the possibilities this new approach brings to biomedical imaging and sensing. Our wavelength-multiplexed diffractive optical processor offers a new solution for high-resolution, label-free imaging of transparent specimens, which could bring significant benefits to biomedical microscopy, sensing and diagnostic applications.”.

Aydgan Ozkan, Principal Investigator and Chancellor's Professor, University of California

Deep learning optimizes wavelength multiplexing and passive diffractive optical elements in a novel multi-plane QPI design that provides high-speed quantitative phase imaging of samples across many axial planes by performing a spectrally multiplexed phase-to-intensity conversion.

The system's compact design and all-optical phase recovery capability make it a competitive analog alternative to traditional digital QPI techniques.

The method was validated by proof-of-concept experiments successfully imaging intrinsic phase objects at different axial positions in the terahertz band. The scalability of this design will facilitate its adaptation to different electromagnetic spectral regions, such as the visible and infrared bands, using appropriate nanofabrication techniques.

This opens new avenues for integrating phase imaging solutions with focal plane arrays or image sensor arrays to create effective on-chip imaging and sensing devices.

This research will have a profound impact on many fields, including biological imaging, sensing, materials science, and environmental analysis. By providing a faster and more effective way to perform 3D QPI, this technique can improve a variety of applications, including environmental sample monitoring, materials characterization, and disease diagnosis and research.

Journal References:

Shen, C.-Y. others(2024) Multiplanar quantitative phase imaging using wavelength-multiplexed diffractive optical processor. Advanced Photonics. doi.org/10.1117/1.ap.6.5.056003.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *