Diffraction neural networks enable quantitative phase imaging through the scattering medium

Full optical phase recovery and quantitative phase imaging through scattering media using diffraction neural networks are used. Without the need for a digital image reconstruction algorithm, diffractive neural networks reconstruct phased objects hidden behind a randomly scattered medium. Image source: Ozcan Lab @ UCLA

Quantitative phase imaging

For a long time, imaging weakly scattered (transparent) phase objects such as cells has been a popular research direction in many fields, including biomedicine. Exogenous methods such as chemical stains or fluorescent labels are often used to improve image contrast in weakly scattered samples. However, such methods often require relatively complex sample preparation steps, and the staining process can be toxic and have damaging effects on the sample. In contrast, Quantitative phase imaging (QPI), as a label-free method, can perform non-invasive, high-resolution imaging of transparent samples without the use of any external reagents, and can quantitatively reflect the phase information of the sample to be measured, thus effectively overcoming the shortcomings of chemical staining methods. However, due to the need to use digital image reconstruction and phase recovery algorithms, traditional quantitative phase imaging systems are computationally intensive and slow. In addition, the imaging of biological tissues is often disturbed by random scattering media, the effects of which are not considered by the vast majority of quantitative phase imaging methods.

In order to solve these problems, Professor Ozcan’s research team from the University of California, Los Angeles (UCLA) has proposed a new method to achieve quantitative phase imaging through random unknown scattering media using diffractive neural networks.

The related results were published in Light: Advanced Manufacturing under the title “Quantitative phase imaging (QPI) through random diffusers using a diffractive optical network.”

Diffractive neural network is a kind of optical computing architecture based on free space, which has attracted more and more research attention in recent years. In this work, the researchers carefully designed a set of diffraction neural networks composed of transmissive spatial structural surfaces in a data-driven manner. The network is compact in size and has an axial length of only 70 illumination wavelengths. The researchers trained diffractive optical networks on randomly generated scattering media to improve the robustness of the network to phase perturbations caused by random unknown scattering media. And this deep learning-based network training only needs to be trained once, and the resulting diffraction neural network can perform full optical phase recovery and quantitative phase imaging of unknown objects hidden behind random scattering media.

Simulation results show that the diffraction neural network has the ability to reconstruct object phase information through the scattering medium and perform quantitative phase imaging. In addition, the researchers also studied many factors such as the number of diffractive surface layers, the limiting relationship between image reconstruction quality and energy efficiency. They found that deeper diffractive neural networks (i.e., having more diffraction surfaces) generally performed better than shallower networks. It is worth mentioning that the proposed optical computing framework can be scaled equally according to the wavelength of the illumination light, so as to be suitable for different electromagnetic wave bands without the need to redesign or retrain the diffraction surface.

This all-optical computing architecture has the advantages of low power consumption, fast processing speed, and small size. The UCLA research team expects that this method can be integrated with existing image sensing (CMOS/CCD) to effectively convert standard microscopes into diffractive quantitative phase imaging microscopes, so as to have the ability to use optical diffraction to achieve phase recovery and image reconstruction of on-chip objects through passive structural surfaces. (Source: Advanced Manufacturing WeChat public account)

Related paper information:

Special statement: This article is reproduced only for the need to disseminate information, and does not mean to represent the views of this website or confirm the authenticity of its content; If other media, websites or individuals reprint and use from this website, they must retain the “source” indicated on this website and bear their own legal responsibilities such as copyright; If the author does not wish to be reprinted or contact the reprint fee, please contact us.

Source link

Related Articles

Leave a Reply

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

Back to top button