Fully optical phase recovery and quantitative phase imaging are performed immediately without a computer

Optical imaging and characterization of weakly scattering phase objects, such as isolated cells, bacteria, and thin tissue sections, often used in biological research and medical applications, has been of considerable interest for decades. Due to their optical properties, when these “phase objects” are illuminated with a light source, the amount of scattered light is usually much less than the light passing directly through the sample, resulting in poor image contrast using traditional imaging methods. This low contrast of the image can be overcome with the help of, for example, chemical stains or fluorescent labels. However, these external methods of labeling or coloring are often tedious, expensive, and involve toxic chemicals.

Quantitative Phase Imaging (QPI) is emerging as a powerful approach without labels for optical examination and detection of various low-scattering, transparent phase objects. Over the last few decades, we have witnessed the development of a number of digital methods for quantitative phase imaging based on image reconstruction algorithms running on computers to recover the phase image of the object from various interferometric measurements. These digital QPI techniques, powered by graphics processors (GPUs), have been used in a variety of applications, including pathology, cell biology, immunology, and cancer research, among others.

In a new research paper published in Advanced optical materials, a team of optical engineers led by Professor Idogan Ozcan of the Department of Electrical and Computer Engineering and the California Institute of Nanosystems (CNSI) at the University of California, Los Angeles (UCLA), developed a diffraction optical network to replace the image reconstruction algorithms in QPI systems with a series of passive optical surfaces that are spatially designed using deep learning. Unlike conventional QPI systems, where the phase recovery step is performed on a digital computer using intensity measurement or hologram, the diffraction QPI network directly processes the optical waves generated by the object itself to extract the phase information of the sample as light propagates through the diffraction network. Therefore, all phase recovery and quantitative phase imaging processes are performed at the speed of light and without the need for an external power source, except for the illuminated light. After the light interacts with the object of interest and propagates through the spatially designed passive layers, the reconstructed phase image of the sample appears at the output of the diffraction grid as an intensity image, successfully converting the phase characteristics of the input object into an output intensity image.

These results represent the first fully optical phase extraction and phase-to-intensity transformation achieved by diffraction. According to the results presented by the UCLA team, QPI diffraction networks trained through deep learning can not only be aggregated to invisible, new phase objects that statistically resemble training images, but also can be aggregated to completely new types of objects. with different spatial characteristics. In addition, these diffraction QPI networks are designed so that the quantification of the input phase is invariant with respect to possible changes in light intensity or image sensor detection efficiency. The team also showed that QPI diffraction networks can be optimized to maintain their quantitative phase image quality even in the event of mechanical inconsistencies in its diffraction layers.

The diffraction QPI networks reported by the UCLA team represent a new concept of phase imaging that, in addition to its superior computational speed, completes the phase recovery process when light passes through thin, passive diffraction surfaces and therefore eliminates energy consumption and the use of memory needed in digital QPI systems, potentially paving the way for various new applications in microscopy and sensors.