News Release

Is there a new path for managing the resolution of digital holographic microscopy?

Peer-Reviewed Publication

Light Publishing Center, Changchun Institute of Optics, Fine Mechanics And Physics, CAS

Fig. 1. Schematics of Digital holographic microscopy (DHM)

image: * view more 

Credit: by Peng Gao and Caojin Yuan

Since the 17th century, optical microscopy has been acting as a powerful tool for detecting subtle structures of samples, playing an irreplaceable role in many fields. However, traditional optical microscopy can only obtain image contrast once specimens have absorption or scattering on the illumination light. Transparent samples, such as living cells, cannot be investigated.

 

Fluorescence microscopy has been used to highlight structures of interest in advance for many years. However, it is difficult to continuously observe living cells over a long time without changing the behavior of the cells. Thus, there is an urgent need for label-free microscopic tools to track the dynamics of live cells in the long term.

 

Digital holographic microscopy (DHM) is a label-free, quantitative phase microscopy approach. It records a hologram between an object wave and a reference wave by using a digital camera (CCD or CMOS). Both the amplitude and phase images of the imaged sample can be reconstructed from the hologram. However, the resolution of DHM is relatively low due to its limited diffraction. DHM has been widely applied in industrial inspection, liquid and gas flow visualization, and biomedical imaging. Advances in the field are necessary to expand its commercial applications and opportunities.

 

Professors Peng Gao (Xidian University) and Caojin Yuan (Nanjing Normal University) have collaborated on a review paper published today in Light: Advanced Manufacturing. This report, entitled "Resolution enhance of digital holographic microscopy via synthetic aperture: a review," reviewed various resolution enhancement approaches in DHM and sought to examine their various advantages and disadvantages.

 

During the past decades, super-resolution optical imaging techniques have been developed. The super-resolution optical imaging field pioneers, Stefan W. Hell, Eric Betzig, and W. E. Moerner, were awarded the Nobel Prize in Chemistry in 2014. Yet, these methods utilize the switching of fluorophores. For DHM, many efforts have been made to improve the spatial resolution of DHM while preserving a large field of view (FOV).

 

The researchers classified the resolution enhancement approaches of DHM into three types: illumination modulation techniques, holographic recording enhancements, and deep-learning assistance.

 

Illumination modulation techniques comprise oblique illumination, structured illumination, or speckle illumination. They all seek to increase illumination's numerical aperture (NA) by scanning mirrors, using diffraction gratings or spatial light modulators. All three techniques are designed to create a broader frequency spectrum, which in turn provides a resolution-enhanced reconstruction after a Fourier transform. The researchers found that these techniques allow for 3D imaging while enhancing the spatial resolution and keeping a large field of view.

 

The researchers were also optimistic about the holographic recording enhancement approaches primarily used for lens-free DHM. The significant advantages of lens-free DHM lie in their reduced size and cost, the larger FOV, and better integration with on-chip devices. The resolution of lens-free DHM systems can be improved via self-extrapolation of holograms, hologram expansion, and pixel super-resolution for on-chip DHM.

 

Rapid developments in artificial intelligence and deep learning techniques provide more opportunities to improve DHM resolution. Neural networks allow mapping a low-resolution hologram input against a low-resolution image output. This follows training with a vast quantity of paired images. Deep learning has also been used to enhance the resolution of different kinds of DHM schemes. Despite it being incredibly time-consuming to train the system, there is enormous potential for improving the networks. Incorporating physical models into the networks would reduce the training data and expand the generality and reliability of these methods.


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.