The substantial refractive index contrast between a cell’s cytoplasm and traditional immersion media scatters enough light to obscure the angular scattering from cellular organelles and to reduce contrast in quantitative phase images (QPI). To reduce the whole-cell scattering, we mix cell media with a biocompatible, high-refractive-index liquid to closely match the cytoplasm’s refractive index. We demonstrate with live single-cell images that this enables isolation of organelles’ angular scattering, which will improve angular scattering-based estimates of organelle size. We also explain why index-matching enhances phase contrast within a cell’s QPI and propose a digital method for obtaining this result.

Abstract: Precision spectroscopic measurements of particles in free solution are greatly aided by use of an Anti-Brownian ELectrokinetic (ABEL) trap, which counteracts the effects of diffusion by means of closed-loop feedback to hold a single particle in a diffraction-limited spot for extended-time measurements. Generally, fluorescent emission from the trapped object is used to produce position estimates for the feedback circuit. However, many objects of interest may fluoresce only dimly (such as native fluorescence from a pigment-protein complex) or intermittently (such as a quantum dot). Here we report the development and demonstration of a new trapping modality that incorporates interferometric scattering to produce particle position estimates, called the Interferometric Scattering Anti-Brownian ELectrokinetic (ISABEL) trap. Using the ISABEL trap, we are able to completely decouple trapping from fluorescence detection, permitting trapping of dim and completely dark nanoparticles.

Achieving the isotropic 3D resolution has been one of the most challenging in optical microscopy. When objective lenses with finite numerical aperture, their axial resolution becomes inferior to the lateral resolution. Although various sample rotation methods have been demonstrated to improve the axial resolution, the requirement of invasive sample manipulations has limited their applications for general complex-shaped specimens. Here, we propose a general method for the in-situ isotropic microtomography of freestanding specimens. Exploiting complex wavefront shaping and optical tweezers, we demonstrate that optimally structured 3D light traps can stably rotate a specimen by considering their 3D refractive index distribution, and reconstruct tomograms with isotropic resolution.

Ranging from material science to tissue biology, three-dimensional (3D) optically anisotropic structures have been investigated for versatile purposes in various research areas. However, conventional methods indirectly access information of 3D anisotropic structure, due to the lack of direct imaging modality for 3D anisotropy. Here, we present a method for reconstructing 3D dielectric tensors of anisotropic structures. Dielectric tensor, a physical descriptor for vectorial light-matter interaction, serves intrinsic information of optical anisotropy including principal refractive indices and optic axes. We demonstrate quantitative tomographic measurements of various nematic liquid-crystal structures and their fast 3D nonequilibrium dynamics.

We present a methodology that allows precise determination of the optical and thermal properties of layered materials using thermal perturbation and phase-resolved optical coherence tomography (pOCT). Following heating with a laser pulse of a buried absorbing layer, optical path length changes between the sample’s surface and the absorbing layer are monitored using a line-scan pOCT. Using an axisymmetric thermo-mechanical model for transversely uniform multi-layered media, we fit the absorption coefficient, the heat conductivity, and the thermal expansion coefficient of the polymer. We demonstrate that the temperature distribution can be determined with a precision under 0.1°C, after a single laser pulse.

We report diffractive optical networks designed through a task-specific training process to classify and reconstruct spatially-overlapping phase images. Trained with ~550-million unique combinations of spatially-overlapping phase-encoded handwritten digits (MNIST), our blind testing achieves >85.8% accuracy for all-optical, simultaneous classification of two overlapping phase images of new/unseen handwritten digits. We also demonstrate the reconstruction of these phase images based on a shallow electronic neural network that uses as its input the highly-compressed optical signals synthesized by the diffractive network with ~20-65 times less number of pixels. This framework might find applications in computational imaging, on-chip microscopy and quantitative phase imaging fields.

Fourier Ptychographic Microscopy (FPM) is a computational imaging technique which reconstructs super-resolved amplitude and phase images by combining variably illuminated low-resolution images through an iterative phase retrieval algorithm. However, the phase-retrieval-based reconstruction requires sufficient overlap between spatial frequency bands of the measurements, which creates a trade-off between the number of measurements and the reconstruction quality. We propose a deep-learning-based FPM reconstruction that recovers both amplitude and phase images in high resolution with far fewer measurements than conventional FPM, with model-based constraint. Our model works with almost no overlap between low-resolution measurements in the Fourier domain, only taking into account the total Fourier extent of the measurements.

Three-dimensional (3D) cellular systems have been increasingly adopted over 2D cell monolayers to study disease mechanisms and discover drug therapeutics, as they more accurately recapitulate the in vivo cellular communication and development of extracellular matrices. However, significant challenges exist for the existing optical microscopy techniques when applied to increasingly thick tissue structures. Recently, we have reported artificial confocal microscopy (ACM), a laser scanning QPI system combined with deep learning algorithms, which renders “synthetic” fluorescence confocal images from thick unlabeled specimens. Here, we aim to determine quantitative markers that can report on the viability of mammalian embryos. We start with the identification of nuclei in the embryonic cells, as the presence of anucleated cells indicate low viability of the embryos. Using phase imaging with computational specificity (PICS), we show that nuclear mask can be predicted from GLIM images alone.

The selection of sperm cells possessing normal morphology and motility is crucial for in-vitro fertilization (IVF). Since stains cannot be used in human IVF, clinicians examine sperm cells using imaging methods that only provide gross 2D morphology, leading to lack of consistency in sperm selection. I will review our latest advances in quantitative phase imaging with deep-learning-powered virtual staining called HoloStain. This method can take images of sperm cells retrieved from their off-axis holograms, acquired without cell staining, and make them look as if they have been chemically stained. I will also review our new approach for ultra-rapid 3D refractive-index imaging, providing high-resolution tomographic phase microscopy for acquisition of the entire sperm cell (head with organelles and tail) during free swim and without staining. These novel tools are now becoming available for direct clinical use, giving rise to new and exciting opportunities for IVF.

Despite the progress in QPI during the last decades, disseminating QPI to broader communities is still in its infancy. A critical limiting factor has been the limited biochemical specificity of QPI. We hypothesized that high-specificity information could be directly retrieved by incorporating the surrounding RI values in 3D space with aid from machine vision. Specifically, we trained deep convolutional networks to transform RI tomograms to the corresponding fluorescence tomograms. This approach achieved state-of-the-art prediction accuracy and generalization across cell types, which enabled applications to new samples without retraining. Together, QPI data do have substantial biochemical specificity that can be accessed.

Caenorhabditis elegans can survive upon harsh environments by entering dauer diapause with reduced metabolic activity and distinctive structural changes. We employed optical diffraction tomography (ODT) to quantitatively measure the transition of mass density distribution of living C. elegans larvae in the reproductive and diapause stages. ODT revealed that the mass density of C. elegans larvae increased upon entry into dauer diapause, and surprisingly, the harshly desiccated dauer larvae exhibited very high refractive index values (n ~ 1.5). Moreover, mutants that are sensitive to desiccation displayed structural abnormalities in the anhydrobiotic stage that were not observed by conventional microscopy.

The COVID pandemic prompted the need for rapid detection of the SARS-CoV-2 virus and potentially other pathogens. In this study, we report a rapid, label-free optical detection method for SARS-CoV-2 that is aimed at detecting the virus in the patient’s breath condensates. We show in the published pre-clinical study that, through phase imaging with computational specificity (PICS), we can detect and classify SARS-CoV-2 versus other viruses (H1N1, HAdV and ZIKV) with 96% accuracy, within a minute after sample collection. PICS combines ultrasensitive quantitative phase imaging (QPI) with advanced deep-learning algorithms to detect and classify viral particles. The second stage of our project, currently under development, involves clinical validation of our proposed testing technique. Breath samples collected from patients in the clinic will be imaged with QPI and a U-Net model trained on the breath samples will identify the SARS-CoV-2 in the sample within a minute.

We combine ultrasensitive microscopy with a novel, highly precise, and fast phase control to enable quantitative phase imaging of weakly scattering objects such as single microtubules. We demonstrate 3D mapping of microtubule networks and real-time 3D localization of single proteins labeled with small metallic nanoparticles. In particular, the 3D trajectories of microtubule-associated proteins acquired at the kHz rate revealed a complex trajectory of the protein diffusion on the surface of microtubules. The fast and accurate phase shaping technique combined with interferometric scattering microscopy pushes the limits of the quantitative phase imaging deep into the subdiffraction regime.

In this study, we use phase imaging with computational specificity (PICS) to detect unlabeled mitochondria in live cells and monitor their dynamics over time.This is a two-step study with first phase involving detection of mitochondria in phase images using deep learning. HCT116 cells with GFP tagged mitochondria were imaged with a correlative SLIM and fluorescence imaging instrument, resulting in pairs of registered phase and fluorescence images per field of view. A deep neural network, EfficientNetB2+U-Net, was trained on the phase - fluorescence image pairs. Our network can predict mitochondria from the SLIM images with a SSIM of 0.9. The second step involves monitoring the effects of anticancer drugs on the mitochondria network dynamic, dry mass of mitochondria content, and their correlation with the overall cell health and drug efficacy. This method can potentially be translated into a tool for label-free efficacy evaluation of mitochondria inhibiting drugs for cancer therapy.

Collagen orientation is one of the major indicators of cancer tumor aggressiveness and can be evaluated via tumor-associated collagen signature-3 (TACS-3). Quantitative phase imaging (QPI) is a rising imaging field in which the map associated with the optical pathlength through different regions of biological specimens are determined. Here we present the application of phase imaging with computational specificity (PICS) to image unlabeled collagen. Thus, we combine spatial light interference microscopy (SLIM), a highly sensitive QPI method, with an image-to-image deep neural network to extract phase-resolved collagen features from the input SLIM captures of histology tissue samples.

The current gold standard for tissue histopathology is based on the examination of stained tissue slices by pathologists. To save time and minimize human efforts, computerized methods in nuclei segmentation and classification have been advanced and popularized for histopathology applications. Digital histopathology has been increasingly applied to cancer diagnosis and prognosis. We proposed a label-free digital histopathology method based on refractive index maps measured from a large field of view optical diffraction tomography system. We measured breast cancer tissue slices and digitally processed the data for cancer grading.