We demonstrate a new computational illumination technique that achieves a large space–bandwidth–time product, for quantitative phase imaging of unstained live samples in vitro. Microscope lenses can have either a large field of view (FOV) or high resolution, and not both. Fourier ptychographic microscopy (FPM) is a new computational imaging technique that circumvents this limit by fusing information from multiple images taken with different illumination angles. The result is a gigapixel-scale image having both a wide FOV and high resolution, i.e., a large space–bandwidth product. FPM has enormous potential for revolutionizing microscopy and has already found application in digital pathology. However, it suffers from long acquisition times (of the order of minutes), limiting throughput. Faster capture times would not only improve the imaging speed, but also allow studies of live samples, where motion artifacts degrade results. In contrast to fixed (e.g., pathology) slides, live samples are continuously evolving at various spatial and temporal scales. Here, we present a new source coding scheme, along with real-time hardware control, to achieve 0.8 NA resolution across a FOV with subsecond capture times. We propose an improved algorithm and a new initialization scheme, which allow robust phase reconstruction over long time-lapse experiments. We present the first FPM results for both growing and confluent in vitro cell cultures, capturing videos of subcellular dynamical phenomena in popular cell lines undergoing division and migration. Our method opens up FPM to applications with live samples, for observing rare events in both space and time.
© 2015 Optical Society of America
In vitro microscopy is crucial for studying physiological phenomena in cells. For many applications, such as drug discovery , cancer cell biology , and stem cell research , the goal is to identify and isolate events of interest. Often these events are rare, and so automated high-throughput imaging is needed in order to provide a statistically and biologically meaningful analysis . Thus, an ideal technique should be able to image and analyze thousands of cells simultaneously across a wide field of view (FOV). To observe dynamical processes across various spatial and temporal scales, both high spatial and high temporal resolutions are required. Existing high-throughput imaging techniques [5–10] cannot meet the space–bandwidth–time product (SBP-T) required for wide-field in vitro applications. Here, we introduce a new computational microscopy technique with both high spatial and temporal resolutions over a wide FOV.
Our method is an extension of Fourier ptychographic microscopy (FPM) , which overcomes the physical space–bandwidth product (SBP) limit. Instead of choosing between a large FOV and high resolution, FPM achieves both by trading acquisition speed. Illumination angles are scanned sequentially with a programmable LED array source (Fig. 1), while taking an image at each angle. Tilted illumination samples different regions of Fourier space, as in synthetic-aperture [11,12] and structured-illumination [13,14] imaging. Although the spatial resolution of each measurement is low, the images collected with high illumination angles (dark field) contain subresolution information. The reconstructed image achieves resolution beyond the diffraction limit of the objective—the sum of the objective and illumination NAs. Distinct from synthetic aperture, FPM does not measure phase directly at each angle, but instead uses nonlinear optimization algorithms [7,15–17] similar to translational diversity [18,19] and ptychography [20,21]. Conveniently, the LED array coded source used here is implemented as a simple and inexpensive hardware modification to an existing microscope. Our proposed method uses efficient source coding schemes to reduce the capture time by several orders of magnitude.
The standard approach for large-SBP microscopy is slide scanning, in which the sample is mechanically moved around while imaging with a high-resolution objective to build up a large FOV. Scanning limits throughput  and is unsuitable for the in vitro imaging of dynamic events. Instead of starting with high resolution and stitching together a large FOV, FPM starts with a large FOV and stitches together images to recover high resolution. A major advantage is its ability to capture the required set of images with no moving parts, by varying illumination angles. In addition, scanners use high-magnification objectives with short depths of fields (DOFs), thus requiring extensive autofocusing mechanisms. In contrast, FPM provides robustness to focus errors because its DOF is longer than that, which would be provided by a high-magnification objective of equivalent NA . Further, out-of-focus images can be digitally refocused .
The main limitation of FPM for real-time in vitro applications is long acquisition times and large datasets. Not only are a large number of images () captured for each reconstruction, but also long exposure times are needed for the dark-field images. To shorten the exposure times, we first built a custom hardware setup consisting of high-brightness LEDs and fast control circuitry (see Section 4.A). This allows us to reduce the total acquisition time by , enabling a full 173 LED scan to capture 0.96 gigapixels of data in under 7 s. While this speed is already suitable for many in vitro applications (e.g., cell division processes), fast subcellular dynamics (e.g., vesicle tracking) require subsecond capture to avoid severe motion blur. Since the camera’s data transfer rate is the limiting factor, large-SBP images cannot be captured in less than 1 s unless we can reduce the data requirements. Furthermore, for studies requiring time-series measurements, the large data requirement ( gigapixel per dataset) poses severe burdens on both storage and processing.
In order to reconstruct a large SBP from fewer images, we need to fundamentally change our capture strategy. To do this, we eliminate redundancy in the data by designing new source coding schemes. The redundancy arises because of a 60% overlap requirement in Fourier space for neighboring LEDs [7,23]. This means that we must capture more data than we reconstruct if using “sequential” FPM. Our angle-multiplexing scheme instead turns on multiple LEDs simultaneously for each measurement, allowing better coverage of the Fourier space with each image. Without eliminating the overlap, we therefore fill in the Fourier space faster. Previous work employed a random coding strategy across both bright-field and dark-field regions . One problem with this scheme is that the bright-field images and dark-field images have large differences in intensity (). Considering Poisson noise statistics, this means that images with mixed bright-field and dark-field components may suffer from the dark-field signal being overwhelmed by the bright-field noise. As a result, we significantly improve our multiplexing results here by separating bright-field from dark-field LEDs. Furthermore, we note that asymmetric illumination-based differential phase contrast (DPC) [24,25] provides a means for recovering quantitative phase and intensity images out to the objective NA with only four images. Hence, there is no need to individually scan the bright-field LEDs. Our new method, termed source-coded FPM, uses a hybrid illumination scheme: it first captures four DPC images (top, bottom, left, right half-circles) to cover the bright-field LEDs, and then uses random multiplexing with eight LEDs to fill in the dark-field Fourier space region [Fig. 1(B)].
Source-coded FPM approaches the theoretical limit for the SBP-T [26,27], set by the camera’s data transfer rate. We achieve an SBP-T of 46 megapixels per second, calculated according to  ( FOV and 0.8 NA captured in 0.8 s). Although the data rate of our camera is larger (138 megapixels per second), we leave some redundancy in order to ensure robust algorithm convergence. Reconstruction algorithms that explicitly take a priori information into account, such as sparsity-based methods [29,30], could further improve the capture speed; here, for generality, we choose not to make any assumptions on the sample. Both sequential FPM and source-coded FPM require the number of images in the dataset to grow quadratically with improved final resolution. This is due to the coverage area in Fourier space increasing in proportion to the square of the final NA. However, our source-coded method has a slower growth rate [Fig. 1(C)] and stays flat for less than resolution improvement. Conveniently, these techniques can flexibly trade off the FOV, resolution, and acquisition time by choosing the illumination angle range.
For in vitro applications, stain-free and label-free are particularly attractive because they are noninvasive and nontoxic. FPM provides both intensity and phase , which contain morphological and cell mass  information that can be used for quantitative study [33,34]. In order to achieve accurate, high-quality results for unstained samples, we needed to make several modifications to the FPM algorithm. It is well known that nonconvex problems such as phase retrieval will often get stuck in the local minima [35–37]; the best way to avoid this is to provide a good initial guess [8,36,38]. Previous work in FPM used a low-resolution intensity image as the initial guess [7,15,16,31]. Stained samples with strong intensity variations thus reconstruct successfully, since the intensity-only initialization is close to the actual solution. However, unstained samples are phase objects and so the intensity-only initial guess does not provide a good starting point. Furthermore, FPM does not measure phase directly at each angle, and the phase contrast provided by the asymmetric illumination results in uneven sensitivity to phase at different spatial frequencies. A detailed phase transfer function analysis  based on the weak-object (Born) approximation [39,40] shows that low-frequency phase information is captured poorly, since it results only from illumination angles that are close to the objective NA. Thus, low-spatial-frequency phase information is more difficult to reconstruct than high-spatial-frequency phase information, contrary to the situation for intensity reconstructions. To improve our reconstruction, we use a linearly approximated phase solution based on DPC deconvolution  as a close initial guess for spatial frequencies within the 2 NA bandwidth. We then run a nonlinear optimization algorithm to solve the full phase problem (see Section 4.B), resulting in high-quality phase reconstructions with high resolution [Fig. 2(A)] and good low-spatial-frequency phase recovery [Fig. 2(B)].
We demonstrate our new source-coded FPM by reconstructing large-SBP videos of popular cell types in vitro on a petri dish, for both growing and confluent samples. We observe subcellular and collective cell dynamics happening on different spatial and temporal scales, allowing us to observe rare events in both space and time, due to the large space–bandwidth–time product (SBP-T) and the flexible tradeoff of time and SBP.
A. Validation with Stained and Unstained Samples
A major advantage of quantitative phase imaging is that it can visualize transparent samples in a label-free, noninvasive way. Figure 2(A) compares our reconstructions before and after staining, for the same fixed human osteosarcoma epithelial (U2OS) sample. With staining, the intensity image clearly displays detailed subcellular features. Stained samples also display strong phase effects, proportional to the local shape and density of the sample. Without staining, the intensity image contains very little contrast and is nearly invisible; however, the phase result clearly captures the subcellular features. Due to the strong similarity between the stained intensity and unstained phase, it follows that a quantitative phase may provide a valid alternative to staining.
To demonstrate the importance of using a good initial guess to initialize the phase recovery for unstained samples, we compare the FPM results both with and without our DPC initialization scheme [Fig. 2(B)]. Both achieve the same 0.7 NA resolution, with high-spatial-frequency features (e.g., nucleus and filopodia) being reconstructed clearly, as expected. However, without DPC initialization, the low-frequency components of the phase are not well recovered, resulting in a high-pass-filtering effect on the reconstructed phase, much like Zernike phase contrast (PhC). With DPC initialization, low frequencies, which describe the overall height and shape of the cells, are recovered correctly.
Next, we verify the accuracy of the recovered phase values by comparing to images captured directly with a higher-resolution objective (, 0.65 NA). The reconstruction resolution for our method matches its expected value (0.7 NA), as shown in Fig. 2(C), which shows images of fixed human mammary epithelial (MCF10A) cells. To validate our quantitative phase result, we compare with that recovered from a through-focus intensity stack captured with the high-resolution objective, and then input into a phase retrieval algorithm  [Fig. 2(C) and Section 4.C]. The quantitative phase can be further used to simulate popular phase-contrast modes such as differential interference contrast (DIC) and PhC . In Fig. 2(D), we compare an actual PhC image to that simulated from our method’s reconstructed result (see Section 4.D). Since the PhC objective uses annular illumination (Ph2, 0.25 NA) to achieve resolution corresponding to 0.9 NA, it should have slightly better resolution than our reconstruction. Note that the simulated PhC image is effectively high-pass-filtered phase information, and so small details appear with better contrast.
B. Fast Sequential FPM Video of HeLa Cells Dividing In Vitro
Figure 3 shows a few frames from a time-lapse video of the human cervical adenocarcinoma epithelial (HeLa) cell division process over the course of 4 h, as well as an automated quantitative analysis of cell dry mass evolution. These data were captured using our improved sequential FPM (173 images)—sample raw images and a schematic of the Fourier space coverage are shown in Fig. 3(A). One time frame of the full-FOV phase reconstruction is shown in Fig. 3(B) and a few frames of the video for a single zoom-in are shown in Fig. 3(C). In this region, one cell is undergoing mitosis and dividing into four cells, during which the cells detach from the surface and become more globular (see Visualization 1). This situation, if imaged with a high-magnification objective, would result in the detached cells moving out of the focal plane and blurring. However, FPM provides a longer DOF than a high-magnification objective with equivalent NA, and so the entire sample stays in focus across the FOV. Subcellular features are visible and the dynamics of actin filament formation can be tracked over time. We show only phase results (omitting intensity) since the samples are unstained and have little intensity contrast.
Since our reconstructed video contains gigapixels of quantitative phase data with cells in each frame, the practical extraction of relevant information requires automated analysis. It is well known that PhC/DIC images cannot be segmented automatically (due to lack of low-spatial-frequency information); however, our quantitative phase results do not have this problem. Using an automated cell segmentation software (CellProfiler ) applied directly to the full-FOV phase result, we first segment each frame of the video to find each of the cells [Fig. 3(D)]. Next, we compute each cell’s dry mass [32,34] over time through the division process, with a few sample cells plotted in Fig. 3(E). The dry mass is calculated by integrating the phase over each segmented cell region for each time frame (see Section 4.E). Note that the automated cell segmentation and dry mass calculation are sensitive to the quality of the phase result, and often fail when low spatial frequencies are not well reconstructed; hence, the DPC initialization and other algorithm improvements implemented in this work are crucial for automated quantitative studies.
C. Source-Coded FPM Video of Neural Stem Cells In Vitro
Source-coded FPM can be used to observe samples across long time scales (up to 4.5 h) with subsecond acquisition speeds (1.25 Hz). An example frame from a reconstructed large-SBP phase video of adult rat neural stem cells (NSCs) is shown in Fig. 4(B), in a petri dish. We use source-coded FPM to achieve the same SBP as in sequential FPM (0.8 NA resolution across a FOV), but with only 21 images [sample raw images shown in Fig. 4(A)], as opposed to 173. As a result, we significantly decrease the capture time, from 7 to 0.8 s. This alleviates motion artifacts that would otherwise blur the result due to subcellular dynamics that happen on time scales shorter than 7 s. Since our source-coded FPM reduces the number of images needed, we can also capture longer video sequences without incurring data management issues. Hence, we create videos of both fast-scale dynamics and slow-scale evolution of NSCs with details at both the subcellular level and across the entire cell population. For example, while vesicle transport and other organelle motions tend to occur on a short time scale and at a small length scale (see Visualization 2), NSC differentiation into neuronal and glial lineages occurs on a longer time scale (i.e., days) and at a larger length scale. In the middle time scale, we observe NSCs sensing their microenvironment by retracting and extending processes, reorganizing their cytoskeletons, migrating, and maturing their axonal and dendritic processes.
With live samples imaged in vitro, dynamics create motion blur artifacts that can destroy the resolution improvements gained by large-SBP methods. Thus, the final effective resolution is always coupled with acquisition speed and sample-dependent motion. In general, smaller subcellular features tend to move at faster speeds; hence, we find that capture times of the order of minutes always incur motion blur. For this reason, faster acquisition is essential for in vitro applications with typical cell types. To demonstrate this point, we compare the results of our source-coded FPM, sequential FPM with and without real-time hardware controls, and high-magnification DPC. The DPC method is considered our benchmark since it achieves faster acquisition speeds, avoiding motion blur, albeit with a small FOV. In Fig. 5, we show zoomed-in phase reconstructions for two cell types with each capture scheme, from slowest to fastest. In each case, the final result has a nominal NA of 0.8 and so each of these results should have the same resolution. However, it is clear that the slower capture schemes blur out features, rendering much of the small-scale structure and dynamics invisible.
The MCF10A cells in Figs. 5(A) and 5(B) exhibit the fastest dynamics observed, due to rapid shuttling of small vesicles and fluctuations of cytoskeletal fibers (e.g., actin filaments, microtubules). Using sequential FPM with a 60 s capture time, we retain almost no details about the structure of the fibers [Fig. 5(A)]. Even with our faster hardware setup, requiring only 7 s acquisition time, the fibers are still completely blurred out. By switching to our source-coded FPM, we obtain the same SBP as in the previous two schemes, but with only 0.8 s acquisition time. Now fiber cytoskeletal details become more discernible, although there is still some motion blur, as compared with the DPC result (Visualization 3). Hence, even our source-coded FPM is missing some information in this case. In Fig. 5(B) we zoom in on another fast process of vesicle transport, where we find that our method captures more of the dynamics than in the fiber fluctuation case, still with some blurring. In Visualization 3, we show time-lapse videos over 8 h to observe cell migration and proliferation with subcellular details, captured using our source-coded FPM. Thus, the trade-off in FOV and time may be more appropriate, depending on what one aims to observe.
In Figs. 5(C) and 5(D), example results for two different NSCs are shown. In both cases, sequential FPM results in significant motion blur, particularly along thin, extended processes and within intracellular vesicles and organelles. In contrast, source-coded FPM is able to accurately capture the full details of the sample without motion blur artifacts. While it is difficult to compare directly because the live cells were moving in between capture schemes, in general we can say that most vesicle transport, retraction and extension of processes, and other organelle motion can be clearly visualized using source-coded FPM (see Visualization 4).
The flexibility of our system in trading off FOV, resolution, and time means that experiments can be tailored to the sample. For example, the HeLa and NSC samples shown here display slower subcellular dynamics, and so are more suited to high-SBP imaging with our current scheme, whereas the MCF10A sample necessitates a trade-off of SBP in order to capture data with sufficient speed. When dynamics are on time scales faster than 0.8 s, one should reduce the number of images captured, either by sacrificing FOV (using a larger-NA objective) or by reducing the resolution improvement factor (using a smaller range of LEDs). In the limit, one can eliminate all dark-field LED images and simply implement DPC for maximum speed of capture. However, for most biological dynamics that are studied in vitro (e.g., differentiation, division, apoptosis), we find that subsecond acquisition is sufficient, and so the SBP should be maximized within this constraint.
In summary, we have demonstrated a high-speed, large-SBP microscopy technique, providing label-free quantitative phase and intensity information. Our source-coded FPM method overcomes the limitations of existing large-SBP methods, permitting fast, motion-free, imaging of unstained live samples. This work opens up large-SBP imaging to high-throughput in vitro applications across a large range of both spatial and temporal scales. A gallery of interactive full-FOV high-resolution images from our experimental system can be found at http://www.gigapan.com/profiles/WallerLab_Berkeley.
A. Experimental Setup
We place a custom-built surface-mounted LED array (4 mm spacing, central wavelength 513 nm with 20 nm bandwidth) above the sample [Fig. 1(A)], replacing the microscope’s standard illumination unit (Nikon TE300). All LEDs are driven statically using 64 LED controller chips (MBI5041) to provide independent drive channels. A controller unit based on an ARM 32-bit Cortex M3 CPU (STM32F103C8T6) provides the logical control for the LEDs by the C interface at 5 MHz, with an LED pattern transfer time of . The camera (PCO.edge 5.5, 6.5 μm pixel pitch) is synchronized with the LED array by the same controller via two coaxial cables that provide the trigger and monitor the exposure status. All raw images are captured with an exposure time of 14 ms. We experimentally measure the system frame rate to be for capturing full-frame () 16-bit images. The data are transferred to the computer via a CameraLink interface. All in vitro experiments are performed in petri dishes placed inside a temperature- and -controlled stage-mounted incubator (In Vivo Scientific).
B. Large-SBP Quantitative Phase and Intensity Reconstruction
Our new FPM reconstruction algorithm can be described in two steps. First, we calculate a low-resolution initialization based on DPC. Next, we implement our quasi-Newton’s method iterative reconstruction procedure  to include the higher-order scattering and dark-field contributions, for 3–5 iterations. In our source-coded FPM, the four bright-field images are directly used in the deconvolution-based DPC reconstruction algorithm  to calculate the phase within the objective’s NA. The initial low-resolution intensity image is calculated by the average of all bright-field images corrected by the intensity falloffs  and then de-convolved by the absorption transfer function . In our improved sequential FPM algorithm, the four DPC images are numerically constructed by taking the sum of single-LED images corresponding to the left, right, top, and bottom half-circles on the LED array.
In the reconstruction, we divided each full-FOV raw image ( pixels) into subregions ( pixels each), with a 160-pixel overlap on each side of neighboring subregions. Each set of images was then processed by our algorithm above to create a high-resolution complex-valued reconstruction having both intensity and phase ( pixels). Finally, all high-resolution reconstructions were combined using the alpha-blending stitching method  to create the full-FOV high-resolution reconstruction. Using a desktop computer (Intel i7 CPU), the processing time for each subregion was in Matlab. The total processing time for each full FOV was .
C. Phase Reconstruction from a Through-Focus Intensity Stack
We capture intensity images with a high-resolution objective (, 0.65 NA) while moving the sample axially to 17 exponentially spaced positions from to 64 μm using a piezostage (MZS500-E-Z, Thorlabs). The images are then used to reconstruct the phase [Fig. 2(C)] using a transport-of-intensity-type algorithm based on spatial frequency domain fitting .
D. Simulation of Phase-Contrast Images
Our phase-contrast simulation [Fig. 2(D)] fully accounts for the partially coherent annular illumination (inner , outer , measured experimentally) and apodized phase-contrast pupil . The pupil consists of a -phase shifting ring with 75% attenuation (size matching the source) and two apodization rings with 50% attenuation (widths are calculated based on ). In our simulation, a tilted plane wave from each source point shifts the sample’s spectrum in the Fourier space, which is then filtered by the pupil function before calculating the intensity image in the real space. The phase-contrast image is the incoherent sum of all the intensity contributions from all the points on the annular source .
E. Image Segmentation and Cell Dry Mass Calculation
Image segmentation for each frame is performed by CellProfiler  open-source software, which implements a series of automated operations, including thresholding, watershedding, and labeling, to return a 2D map containing segmented regions representing different cells. The 2D maps are then loaded into Matlab to extract the phase within each individual cell. The total dry mass for each cell is calculated as the sum of the dry mass density . The dry mass density is directly related to phase by , where is the center wavelength and is the average of reported values for the refractive increment of protein . The background fluctuations are characterized from a region without any cells [white square in Fig. 3(D)] and having an area similar to the average cell size. Background fluctuations are shown by the black curve and histogram in Fig. 3(E). We achieve a standard deviation of 1.5 pg, in units of dry mass, indicating good stability of the phase measurement.
F. Sample Preparation
HeLa cells were cultured with DMEM (Dulbecco’s modified Eagle’s medium) supplemented with 10% fetal bovine serum, glutamine, and penicillin/streptomycin. The cells were plated on a p75 flask and cultured in a 37°C incubator with 5% . Confluent cells were treated with 0.2% trypsin and passed at a 1:8 ratio. Two drops of trypsinized cells were then added into a poly-d-lysine-coated 35 mm MatTek glass-bottom plate with 2 ml of medium. After 6 h of incubation, the cells became fully attached to the plate. U2OS cells were prepared using the same procedure. They were fixed in 4% formaldehyde at room temperature for 10 min and later stained with 1% toluidine blue O at room temperature for 5 min and rinsed in three changes of double-distilled water. Adult rat NSCs were isolated from the hippocampi of 6-week-old female Fischer 344 rats. To promote adhesion, tissue culture polystyrene plates were first coated with 10 μg/ml poly-L-ornithine (Sigma) overnight at room temperature, followed by 5 μg/ml laminin (Invitrogen) overnight at 37°C. NSCs were cultured in monolayers in 1:1 DMEM/F12 high-glucose medium (Life Technologies), supplemented with N-2 (Life Technologies) and 20 ng/ml recombinant human FGF-2 (Peprotech). The medium was changed every other day and cells were subcultured with accutase upon reaching confluency. MCF10A cells were cultured in DMEM/F12 (Invitrogen), supplemented with 5% horse serum (Invitrogen), 1% penicillin/streptomycin (Invitrogen), 0.5 μg/ml hydrocortisone (Sigma), 100 ng/ml cholera toxin (Sigma), 10 μg/ml insulin (Sigma), and 20 ng/ml recombinant human epidermal growth factor (Peprotech). The medium was changed every other day and cells were passed with trypsin upon reaching confluency. In preparation for imaging, cells were washed once with phosphate-buffered saline and detached with either accutase or trypsin. A total of 300,000 cells were seeded onto 35 mm glass-bottom microwell dishes (MatTek) and allowed to attach.
Gordon and Betty Moore Foundation (GBMF4562).
We would like to thank Zachary Phillips for help with experiments, and Olivia Scheideler, Lydia Sohn, David Schaffer, Peiwu Qin, and Ahmet Yildiz for providing the cell samples and incubator. Funding was provided by the Gordon and Betty Moore Foundation’s Data-Driven Discovery Initiative through Grant GBMF4562 to Laura Waller (UC Berkeley).
1. P. Lang, K. Yeow, A. Nichols, and A. Scheer, “Cellular imaging in drug discovery,” Nat. Rev. Drug Discov. 5, 343–356 (2006). [CrossRef]
2. M. R. Boyd and K. D. Paull, “Some practical considerations and applications of the national cancer institute in vitro anticancer drug discovery screen,” Drug Develop. Res. 34, 91–109 (1995). [CrossRef]
3. M. R. Costa, F. Ortega, M. S. Brill, R. Beckervordersandforth, C. Petrone, T. Schroeder, M. Götz, and B. Berninger, “Continuous live imaging of adult neural stem cell division and lineage progression in vitro,” Development 138, 1057–1068 (2011). [CrossRef]
4. N. Rimon and M. Schuldiner, “Getting the whole picture: combining throughput with content in microscopy,” J. Cell Sci. 124, 3743–3751 (2011). [CrossRef]
5. G. Zheng, S. A. Lee, Y. Antebi, M. B. Elowitz, and C. Yang, “The ePetri dish, an on-chip cell imaging platform based on subpixel perspective sweeping microscopy (SPSM),” Proc. Natl. Acad. Sci. USA 108, 16889–16894 (2011). [CrossRef]
6. A. Greenbaum, W. Luo, T.-W. Su, Z. Göröcs, L. Xue, S. O. Isikman, A. F. Coskun, O. Mudanyali, and A. Ozcan, “Imaging without lenses: achievements and remaining challenges of wide-field on-chip microscopy,” Nat. Methods 9, 889–895 (2012). [CrossRef]
7. G. Zheng, R. Horstmeyer, and C. Yang, “Wide-field, high-resolution Fourier ptychographic microscopy,” Nat. Photonics 7, 739–745 (2013). [CrossRef]
8. A. Greenbaum, Y. Zhang, A. Feizi, P.-L. Chung, W. Luo, S. R. Kandukuri, and A. Ozcan, “Wide-field computational imaging of pathology slides using lens-free on-chip microscopy,” Sci. Transl. Med. 6, 267ra175 (2014). [CrossRef]
9. W. Luo, A. Greenbaum, Y. Zhang, and A. Ozcan, “Synthetic aperture-based on-chip microscopy,” Light 4, e261 (2015). [CrossRef]
10. K. Goda, K. Tsia, and B. Jalali, “Serial time-encoded amplified imaging for real-time observation of fast dynamic phenomena,” Nature 458, 1145–1149 (2009). [CrossRef]
11. T. Gutzler, T. R. Hillman, S. A. Alexandrov, and D. D. Sampson, “Coherent aperture-synthesis, wide-field, high-resolution holographic microscopy of biological tissue,” Opt. Lett. 35, 1136–1138 (2010). [CrossRef]
12. A. E. Tippie, A. Kumar, and J. R. Fienup, “High-resolution synthetic-aperture digital holography with digital phase and pupil correction,” Opt. Express 19, 12027–12038 (2011). [CrossRef]
13. M. G. Gustafsson, “Surpassing the lateral resolution limit by a factor of two using structured illumination microscopy,” J. Microsc. 198, 82–87 (2000). [CrossRef]
14. K. Wicker and R. Heintzmann, “Resolving a misconception about structured illumination,” Nat. Photonics 8, 342–344 (2014). [CrossRef]
15. X. Ou, G. Zheng, and C. Yang, “Embedded pupil function recovery for Fourier ptychographic microscopy,” Opt. Express 22, 4960–4972 (2014). [CrossRef]
16. L. Tian, X. Li, K. Ramchandran, and L. Waller, “Multiplexed coded illumination for Fourier ptychography with an LED array microscope,” Biomed. Opt. Express 5, 2376–2389 (2014). [CrossRef]
17. L. Tian and L. Waller, “3D intensity and phase imaging from light field measurements in an LED array microscope,” optica 2, 104–111 (2015). [CrossRef]
18. J. M. Rodenburg and H. M. Faulkner, “A phase retrieval algorithm for shifting illumination,” Appl. Phys. Lett. 85, 4795–4797 (2004). [CrossRef]
19. M. Guizar-Sicairos and J. R. Fienup, “Phase retrieval with transverse translation diversity: a nonlinear optimization approach,” Opt. Express 16, 7264–7278 (2008). [CrossRef]
20. A. M. Maiden and J. M. Rodenburg, “An improved ptychographical phase retrieval algorithm for diffractive imaging,” Ultramicroscopy 109, 1256–1262 (2009). [CrossRef]
21. P. Thibault, M. Dierolf, O. Bunk, A. Menzel, and F. Pfeiffer, “Probe retrieval in ptychographic coherent diffractive imaging,” Ultramicroscopy 109, 338–343 (2009). [CrossRef]
22. A. Orth and K. Crozier, “Microscopy with microlens arrays: high throughput, high resolution and light-field imaging,” Opt. Express 20, 13522–13531 (2012). [CrossRef]
23. O. Bunk, M. Dierolf, S. Kynde, I. Johnson, O. Marti, and F. Pfeiffer, “Influence of the overlap parameter on the convergence of the ptychographical iterative engine,” Ultramicroscopy 108, 481–487 (2008). [CrossRef]
24. S. Mehta and C. Sheppard, “Quantitative phase-gradient imaging at high resolution with asymmetric illumination-based differential phase contrast,” Opt. Lett. 34, 1924–1926 (2009). [CrossRef]
25. L. Tian and L. Waller, “Quantitative differential phase contrast imaging in an LED array microscope,” Opt. Express 23, 11394–11403 (2015). [CrossRef]
26. W. Lukosz, “Optical systems with resolving powers exceeding the classical limit,” J. Opt. Soc. Am. 56, 1463–1471 (1966). [CrossRef]
27. W. Lukosz, “Optical systems with resolving powers exceeding the classical limit. II,” J. Opt. Soc. Am. 57, 932–939 (1967). [CrossRef]
28. G. Zheng, R. Horstmeyer, and C. Yang, “Corrigendum: wide-field, high-resolution Fourier ptychographic microscopy,” Nat. Photonics 9, 621 (2015). [CrossRef]
29. E. J. Candès and M. B. Wakin, “An introduction to compressive sampling,” IEEE Signal Process. Mag. 25(2), 21–30 (2008). [CrossRef]
30. Y. Shechtman, A. Beck, and Y. Eldar, “GESPAR: efficient phase retrieval of sparse signals,” IEEE Trans. Signal Process. 62, 928–938 (2014). [CrossRef]
31. X. Ou, R. Horstmeyer, C. Yang, and G. Zheng, “Quantitative phase imaging via Fourier ptychographic microscopy,” Opt. Lett. 38, 4845–4848 (2013). [CrossRef]
32. G. Popescu, Y. Park, N. Lue, C. Best-Popescu, L. Deflores, R. R. Dasari, M. S. Feld, and K. Badizadegan, “Optical imaging of cell mass and growth dynamics,” Am. J. Physiol. 295, C538–C544 (2008). [CrossRef]
33. A. R. Cohen, F. L. Gomes, B. Roysam, and M. Cayouette, “Computational prediction of neural progenitor cell fates,” Nat. Methods 7, 213–218 (2010). [CrossRef]
34. M. Mir, Z. Wang, Z. Shen, M. Bednarz, R. Bashir, I. Golding, S. G. Prasanth, and G. Popescu, “Optical measurement of cycle-dependent cell growth,” Proc. Natl. Acad. Sci. USA 108, 13124–13129 (2011). [CrossRef]
35. J. R. Fienup, “Phase retrieval algorithms: a comparison,” Appl. Opt. 21, 2758–2769 (1982). [CrossRef]
36. J. R. Fienup and C. C. Wackerman, “Phase-retrieval stagnation problems and solutions,” J. Opt. Soc. Am. A 3, 1897–1907 (1986). [CrossRef]
37. C. Yang, J. Qian, A. Schirotzek, F. Maia, and S. Marchesini, “Iterative algorithms for ptychographic phase retrieval,” arXiv:1105.5628 (2011).
38. E. J. Candès, X. Li, and M. Soltanolkotabi, “Phase retrieval via Wirtinger flow: theory and algorithms,” arXiv:1407.1065 (2014).
39. H. Rose, “Nonstandard imaging methods in electron microscopy,” Ultramicroscopy 2, 251–267 (1977). [CrossRef]
40. D. Hamilton and C. Sheppard, “Differential phase contrast in scanning optical microscopy,” J. Microsc. 133, 27–39 (1984). [CrossRef]
41. Z. Jingshan, R. A. Claus, J. Dauwels, L. Tian, and L. Waller, “Transport of intensity phase imaging by intensity spectrum fitting of exponentially spaced defocus planes,” Opt. Express 22, 10661–10674 (2014). [CrossRef]
42. E. D. Barone-Nugent, A. Barty, and K. A. Nugent, “Quantitative phase-amplitude microscopy I: optical microscopy,” J. Microsc. 206, 194–203 (2002). [CrossRef]
43. A. E. Carpenter, T. R. Jones, M. R. Lamprecht, C. Clarke, I. H. Kang, O. Friman, D. A. Guertin, J. H. Chang, R. A. Lindquist, J. Moffat, P. Golland, and D. M. Sabatini, “CellProfiler: image analysis software for identifying and quantifying cell phenotypes,” Genome Biol. 7, R100 (2006). [CrossRef]
44. Z. F. Phillips, M. V. D’Ambrosio, L. Tian, J. J. Rulison, H. S. Patel, N. Sadras, A. V. Gande, N. A. Switz, D. A. Fletcher, and L. Waller, “Multi-contrast imaging and digital refocusing on a mobile microscope with a domed led array,” PLoS ONE 10, e0124938 (2015). [CrossRef]
45. T. Otaki, “Artifact halo reduction in phase contrast microscopy using apodization,” Opt. Rev. 7, 119–122 (2000). [CrossRef]
46. L. Mandel and E. Wolf, Optical Coherence and Quantum Optics (Cambridge University, 1995).
47. R. Barer, “Interference microscopy and mass determination,” Nature 169, 366–367 (1952). [CrossRef]