Recently, the State Key Laboratory of Transient Optics and Photonics Technology of Xi'an Institute of Optics and Fine Mechanics has made new progress in the research of intelligent optical microscopic imaging, and the research results were published online in the international high-level academic journal Opto Electronic Advances (IF: 15.3). The first author of the paper is Tian Xuan, a 2024 doctoral candidate of Xi'an Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, and Li Runze, a special research assistant. The corresponding authors are Associate Researcher Bai Chen and Researcher Yao Baoli.
The phase information carried by light waves can reveal the thickness, refractive index, geometric morphology, and other characteristics of matter. As it cannot be directly sensed by optical sensors, interference methods are usually required for detection. Digital in-line holographic microscopy (DIHM) has become a commonly used method for quantitative phase imaging due to its high spatial bandwidth product, label free, non-invasive, and fast imaging speed. However, in practical applications, the interference of twin images during holographic reconstruction and the loss of sub-pixel information caused by the use of large pixel size detectors can hinder high-quality DIHM imaging. Deep learning, with its noise suppression and inverse problem solving capabilities, has become a powerful tool for DIHM imaging and pixel super resolution (PSR). However, most current deep learning based methods rely on supervised learning and training instances to optimize their weights and biases. Collecting a large number of holograms and their corresponding high-resolution raw phase maps is not only time-consuming in experiments, but also very difficult to collect training data. In addition, the trained model has very limited generalization to samples that are different from the training data.
Figure MPPN-PSR phase imaging: (a) Full field pixel super-resolution phase imaging of TOMM20 antibody cells, (b) Comparison of different PSR phase reconstruction methods, and (c) corresponding optical thickness maps.
In response to the above issues, the research team proposed a non trained neural network for DIHM pixel super-resolution phase imaging, namely the Multi Prior Physical Enhancement Neural Network (MPPN-PSR), which can reconstruct phase information of samples from coaxial holograms with high throughput, high accuracy, and high resolution. MPPN-PSR combines neural networks and physical models, encapsulating physical model priors, sparsity priors, and depth image priors in an untrained deep neural network. This avoids the need for a large amount of training data for neural networks and does not require any additional hardware design. It can achieve twin image suppression, pixel super-resolution, and high-throughput phase imaging with only a single hologram. Compared with the phase recovery method without PSR, the MPPN-PSR method increases the pixel resolution of the image by three times. Compared with the classic phase recovery method Twist TV-PSR that combines pixel super-resolution, the optical resolution is increased by about two times. Moreover, due to the use of the inherent large field of view of the low magnification objective lens, MPPN-PSR improves the spatial bandwidth product of the imaging. This research result is expected to provide reference for other digital holographic imaging schemes and be widely applied in the fields of biomedical and industrial measurement.
In recent years, the Yao Baoli team of the State Key Laboratory of Transient Optics and Photonics Technology has conducted in-depth research on intelligent optical microscopic imaging technology, and formed a variety of new optical microscopic imaging technologies, which have achieved significant improvements in imaging functions, information acquisition dimensions, performance indicators, etc., including three-dimensional imaging of full-color wide field micro light slices, fast super resolution three-dimensional imaging of confocal microscopy, fast three-dimensional microscopic imaging of light slices, etc., as well as high-resolution and high signal noise ratio microscopic imaging of light slices, computational imaging through scattering media, etc., which are achieved using compression sensing technology. The relevant research results were published in Photon Journals such as Res, Opt Lett, Opt Express, etc. In addition, the team has conducted long-term theoretical and experimental research on optical microscopy imaging and optical micro manipulation based on light field regulation. They have published more than 300 papers in journals such as PNAS, Nature Com., PRL, Rep. Prog. Phys., Adv. Opt. Photon. They have been granted multiple national invention patents and have won awards and honors such as the first and second prizes of Shaanxi Provincial Science and Technology Innovation Team and Shaanxi Provincial Key Science and Technology Innovation Team.
Source: Opticsky