UCLA Researchers Introduce a Multispectral QPI System Designed Based on a Broadband Diffractive Optical Neural Network

Quantitative Phase Imaging (QPI) is a cutting-edge imaging method in many scientific and microscopy domains. It makes it possible to quantify and see the smallest differences in light’s optical path length as it travels through transparent or semi-transparent materials. The refractive index distribution and thickness changes inside a sample can be learned much with this non-invasive, label-free technique.

Multispectral Quantitative Phase Imaging (QPI) systems build upon this fundamental principle by acquiring multiple phase images across a spectrum of wavelengths or spectral bands of interest. QPI derives information about the refractive index and thickness of the sample by evaluating the phase shifts that light undergoes when interacting with a specimen. 

QPI is a flexible technique with uses outside the traditional biomedical disciplines like cell biology, pathology, and biophysics. It is useful in several scientific fields, including surface science for evaluating biological interfaces and materials science for characterizing optical components and thin films and nanoparticles. Its capabilities include the research of subcellular structures and processes, the monitoring of cell growth and behavior in real-time, the detection of cancer, the detection of pathogens, the measurement of thin film thickness, the evaluation of optical quality, and the analysis of surface roughness.

So, there has been thorough research conducted by the researchers on QPI and thus researchers from the Electrical and Computer Engineering Department at the University of California, Los Angeles (UCLA) have introduced a new design for multispectral QPI. 

This approach uses deep learning to create a broadband diffractive optical network, enabling the acquisition of quantitative phase images across various spectral bands within a single snapshot. The optical network uses several spatially structured dielectric diffractive layers, each with hundreds of thousands of transmissive diffractive features optimized for deep learning.

The optical network performs as an all-optical phase-to-intensity transformer after fabricating the resulting diffractive layers. It does this by optically routing the multispectral QPI signals onto predetermined spatial positions at the output plane, where a monochrome focal plane array measures the resulting intensity distributions, and extracting the phase profiles of the input objects at predetermined wavelengths.

This optical network optimizes the multispectral phase information of the input objects through deep learning, transforming it into a distinct intensity distribution at the output field-of-view that spatially encodes the object phase information corresponding to each target spectral band separately.

QPI is composed of two primary components. One component is an imaging frontend, which is responsible for performing optical interferometry to transform the needed phase information into intensity levels that can be recorded using a digital image sensor, and second is a digital processing backend task to perform the essential image processing and reconstruction of quantitative phase images based on these signals.

To test the system’s accuracy, the researchers validated the ability by imaging new types of never-before-seen objects. The study has put that this is a versatile, general-purpose multispectral quantitative phase imager suitable for various applications.

Check out the Paper. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 30k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..