Quantitative Phase Imaging: Recent Advances and Expanding Potential in Biomedicine

Abstract

Quantitative phase imaging (QPI) is a label-free, wide-field microscopy approach with significant opportunities for biomedical applications. QPI uses the natural phase shift of light as it passes through a transparent object, such as a mammalian cell, to quantify biomass distribution and spatial and temporal changes in biomass. Reported in cell studies more than 60 years ago, ongoing advances in QPI hardware and software are leading to numerous applications in biology, with a dramatic expansion in utility over the past two decades. Today, investigations of cell size, morphology, behavior, cellular viscoelasticity, drug efficacy, biomass accumulation and turnover, and transport mechanics are supporting studies of development, physiology, neural activity, cancer, and additional physiological processes and diseases. Here, we review the field of QPI in biology starting with underlying principles, followed by a discussion of technical approaches currently available or being developed, and end with an examination of the breadth of applications in use or under development. We comment on strengths and shortcomings for the deployment of QPI in key biomedical contexts and conclude with emerging challenges and opportunities based on combining QPI with other methodologies that expand the scope and utility of QPI even further.

Keywords: biophysics; diagnostics; holography; interferometry; microscopy; phase retrieval; quantitative phase imaging; tomography.

Figure 1.

QPI has undergone a steady increase in interest driven by advances in different fields of optics. (a) Schematic of four main QPI approaches with interferometry (green timeline), holography (red timeline), wavefront sensing (orange timeline), and phase retrieval algorithms (light blue timeline) are indicated. These methods have improved extensively over time with the emergence of greater computational resources (thick black line). The improved efficiency of computational resources led to technical advances in QPI that include quantitative phase tomography (magenta timeline), in vivo QPI (dark blue timeline), multimodal approaches (brown timeline), and machine learning methods (yellow-green timeline). (b) The growth in interest and advances in QPI over time depicted by the number of publications on Web of Science using search terms “Quantitative Phase Imaging” or “Quantitative Phase Microscopy” by year.

Figure 2.

Examples of the four primary QPI lineages shown in Figure 1. (a) Michelson interferometry uses the interference between light beams passing through a sample and a reference to generate an interferogram that encodes phase information in the image amplitude (e.g. ref 12). The number of visible interference fringes generated depends on the optical setup and coherence of the light source, with low coherence light sources (e.g., white light) producing fewer visible fringes than highly coherent lasers. An in-focus interferogram is then used to generate the phase image, and, in phase shifting interferometry, the reference and sample path lengths are adjusted in steps, e.g., with a piezo, to shift the fringes by a fraction of a wavelength. (b) DHM computationally reconstructs the phase image from an interferogram obtained using an interferometer.^, Here, a Mach–Zehnder interferometer with a slightly off-axis reference beam is used to avoid the twin image problem, where the image and its conjugate sit on top of one another. (c) Wavefront sensing with QWLSI uses a diffraction grating that captures gradients in phase shift as local distortions in the resulting intensity grid pattern on the camera sensor. A comparison of sample images to a reference wavefront image is used to determine the wavefront distortion due to the sample itself, with numerical integration to recover phase. (d) Differential phase contrast (DPC) microscopy, a deterministic phase retrieval method, images a sample using half-circle patterns of illumination that extend beyond the microscope objective numerical aperture. Light refraction through the sample then causes intensity increases (or decreases) in one-half-circle image and decreases (or increases) in image intensity with the opposing half-circle pattern. The normalized difference between these two images approximates the gradient of phase along one axis. Multiple pairs of images are collected, and the phase is numerically reconstructed.

Figure 3.

Evolution of complexity and information content from QPI measurements of cell dry mass and mass distributions within living cells. Representative images and data analyses are shown in a time series. (a) QPI film image of Tradescantia bractea pollen grain (top) along with QPI pollen grain dry mass measurements (bottom, upward arrowheads are no sucrose estimates and downward arrowheads show measurements within a 5% sucrose solution) and volume (circles) during different phases of development. Adapted with permission from ref .Copyright 1954 Company of Biologists Ltd. (b) QPI of chicken fibroblasts with dry mass densities ranging from 0.01 (darkest gray) to 0.6 (white) pg/μm²(top), processed to measure spread area relative to total cell mass (bottom). Adapted with permission from ref . Copyright 1995 Company of Biologists Ltd. (c) QPI of human H929 multiple myeloma cells (top) showing computationally processed data that simultaneously captures drug responses of hundreds of single cells, shown as initial cell mass versus normalized changes in mass during drug treatment (bottom). Adapted and data set with permission from ref . Copyright 2011 Elsevier. (d) High-resolution QPI of a human buccal epithelial cell (top) and an example of changes in dry mass of HeLa cells undergoing apoptosis triggered by exposure to cytotoxic paclitaxel (bottom). Adapted with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2017 Springer Nature.

Figure 4.

QPI biomechanics measurement evolution. (a) Early QPI biomechanical analyses required physical perturbations, such as actuation of a magnetic bead indenter on NIH3T3 fibroblasts (top) or HeLa carcinoma cells to extract Young’s modulus (E; bottom). Adapted with permission from ref . Copyright 2008 IOP Publishing Ltd. (b) Detailed mechanical modeling from contactless measurements of biomechanical properties of red blood cells (RBCs; top left) using natural fluctuations in phase caused by membrane motion (top right) captures mechanical property variations (bottom) for populations of normal (DC), spiculated (EC), and spherical (SC) shaped RBCs. Adapted with permission from ref . Copyright 2010 National Academy of Sciences, U.S.A. Scale bar = 1.5 μm. (c) QPI phase (top-middle) of more complex cells HT-29 wild-type and shRNA (top left), HT-29 with CSK shRNA-mediated knock down (top middle), A431 epidermoid carcinoma control (top right) and cytochalasin D treated A431 (middle left) cells, and A549 lung adenocarcinoma cells (middle right) used to compute a mean phase disorder strength, related to intracellular cytoskeletal structure and independent measurements of shear stiffness (bottom). Adapted with permission from ref . Copyright 2017 Elsevier. (d) Time lapse QPI data (top) showing the redistribution of mass within single cells and cell clusters, which provides both resistance to deformation and decay terms. These terms were validated by comparisons with AFM measurements of stiffness (bottom left) and viscosity (bottom right) for MCF-7 and BT-474 breast carcinoma cells, and for HeLa endocervical carcinoma cells, treated with different concentrations of cytochalasin B. Adapted with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2020 Springer Nature.

Figure 5.

Progress toward QPI clinical applications as a screening and selection tool for treatments, and as a diagnostic tool to identify healthy versus diseased states. (a) Specific QPI features can identify disease or changes from a healthy or control state. For example, QPI measured differences in RBC membrane fluctuations at 37 and 41 °C in vitro can distinguish between healthy and ring, trophozoite, or schizont diseased states with P. falciparum parasitic infection. Adapted with permission from ref . Copyright 2008 National Academy of Sciences, U.S.A. (b) Once QPI features of interest are identified, validation is sought with an independent, orthogonal method, if available. For example, shown here is an area under the curve (AUC) or receiver operating characteristic (ROC) plot of the true positive (sensitivity) versus false positive (specificity) rate determining malignance from hematoxylin and eosin counter-stained tissue biopsy. This previously validated method is used to validate QPI determined malignant state for breast tissue biopsies. Adapted with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2018 Springer Nature. (c) Validation of a QPI measured feature in a specific context can broaden its utility. For example, validation of QPI measured changes in growth rate was successfully applied to identify effective treatments from a pool of candidate agents against carboplatin-resistant, patient-derived xenograft HCI09 breast carcinoma cells. Adapted with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2019 Elsevier. (d) Example of QPI as a diagnostic tool, with spatial light interference microscopy (SLIM; middle and right columns) identification of benign (top row) versus malignant (bottom row) glandular tissue, validated by pathological classification of hematoxylin and eosin stained biopsy material (left column). Adapted with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2018 Springer Nature.

Figure 6.

Progress in QPI tomography from applications with static optical fibers to multicellular organisms. (a) QPI tomography analysis of cross sections of optical fibers (top). A common feature of QPI tomography is recovery of the 3D refractive index distribution, rather than the integrated refractive index through the sample thickness, as in 2D QPI. This is shown by the refractive index distribution measured as a line profile through the sample (bottom). Reproduced from ref . Copyright 2000 Elsevier. (b) QPI tomography of single cell protozoan, Hyalosphenia papilio, with refractive index reconstructions shown as different 2D slices. Adapted from ref . Copyright 2006 The Optical Society. (c) Multiplexed intensity diffraction tomography of multicellular Caenorhabditis elegans embryos. Shown are in-focus refractive index (top row) and depth-coded projections of volumetric reconstruction (bottom row). Red and orange arrows indicate developmental stages of the embryos. Individual developing tissues, the buccal cavity (white box), intestine (blue box), and native bacteria (blue arrow), are visible. Reproduced with permission from ref . Copyright 2019 The Optical Society.

Figure 7.

Progression of in vivo QPI approaches. (a) Sample preparation for an in vivo technique called spectral-domain optical coherence phase microscopy (SD-OCPM; top) which generated optical path difference maps for human epithelial check cells (bottom). Adapted from ref . Copyright 2005 The Optical Society). (b) Diagram of live mouse heating stage setup for in vivo QPI (top). Representative QPI data from a live mouse mesentery showing mouse microvasculature represented as optical phase delay maps reconstructed from holograms (bottom). Adapted with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2016 Springer Nature. Scale bar = 10 μm. (c) Schematic of a fiber-based quantitative oblique back-illumination microscopy (qOBM) platform for imaging tumor tissue in excised rat brain (top), thereby generating QPI images from deconvolution of intensity images (bottom). Adapted from ref . Copyright 2021 The Optical Society. Scale bar = 50 μm.

Figure 8.

Examples of the opportunities available from coupling QPI with additional imaging modalities. (a) QPI of kidney cells paired with fluorescence detection enables the identification and quantification of dry mass changes, represented by phase shifts, within subcellular regions (right), such as the nucleus, identified by Hoechst staining (bottom left). Reproduced with permission from ref . Copyright 2006 The Optical Society. (b) Enhanced fast image acquisition of dual 3D fluorescence (top right) and refractive index measurements from tomographic QPI (top left). This accelerated approach provides the necessary capture speed in image scanning to reconstruct 3D tomograms of A549 cells for both fluorescence (bottom right) and QPI (bottom left) measurements from z-step data Adapted with permission from ref . Copyright 2017 The Optical Society. (c) Molecular vibrational spectroscopy paired with QPI of COS7 cells (top left) examined for molecular signatures, such as CH₂ (top center) and peptide bending (top left), corresponding to subcellular phase shifts within the nucleus (orange), cytoplasm (blue), relative to empty space control (gray) (bottom). Reproduced with permission from ref . Copyright 2020 The Optical Society..

Figure 9.

Machine learning has been applied to all three stages of a typical QPI processing and analysis pipeline: (1) computation of phase data, (2) labeling of phase images, and (3) feature-based cell classification. (a) Phase image reconstruction from a single overfocus or under-focus image using deep learning and TIE algorithm. The error of phase calculation using the combined deep learning TIE method is under 0.06 π for the ‘Network+’ learning-based method using one overfocus image and the ‘Network-’ method using an under-focus image when compared to the ground truth calculated from three images using TIE. Reproduced with permission from ref . Copyright 2020 Elsevier. (b) PhaseStain is a digital staining method developed using deep learning on holographic microscopy images, to perform virtual staining of tissues from label-free QPI images. The stained images produced are similar to histological staining observed under a brightfield microscope. (c) A zoomed-in view comparing the liver tissue section stained using PhaseStain and Masson’s trichrome staining. Reproduced with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2019 Nature. (d, e) Machine learning to classify cell states during the epithelial-to-mesenchymal transition (EMT). M-phase, pro-apoptotic, and growth-arrested cell states occurring during EMT can be distinguished from untreated control cells using machine learning, utilizing cell features identified from QPI. Reproduced with permission under Creative Commons Attribution (CC BY) license from ref . Copyright 2017 Springer Nature.