Source-linked AI summary
Computational illumination for high-speed in vitro Fourier ptychographic microscopy
Lei Tian, Ziji Liu, Li-Hao Yeh, Michael Chen, Jingshan Zhong, Laura Waller
TL;DR
Large-SBP FPM provides wide-FOV, high-resolution imaging but is limited by long acquisitions, large datasets, and motion artifacts in live samples. This paper introduces source-coded illumination, real-time hardware control, and improved phase reconstruction, achieving 0.8 NA across a 4× FOV in 0.8 seconds and imaging dynamic cell behavior.
Problem
In vitro imaging needs wide FOV, high spatial and temporal resolution, and high throughput, but existing FPM is limited by long acquisition times, large datasets, and motion artifacts in live samples.
Method
The paper uses multiplexed source-coded LED illumination, fast hardware control, and improved initialization and reconstruction algorithms for unstained live-sample FPM.
Results
46 megapixels/second SBP-T was achieved for a 4× FOV and 0.8 NA captured in 0.8 seconds, while live-cell videos captured subcellular and collective dynamics.
Takeaways & Limitations
Source-coded FPM enables large-SBP, label-free quantitative imaging of live in vitro samples across spatial and temporal scales.
Takeaways & Limitations
Fastest MCF10A dynamics still produced some motion blur with source-coded FPM at 0.8 seconds, missing some information relative to DPC.
Abstract
from arXiv · showhide
We demonstrate a new computational illumination technique that achieves large space-bandwidth-time product, for quantitative phase imaging of unstained live samples in vitro. Microscope lenses can have either large field of view (FOV) or high resolution, 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 wide FOV and high resolution, i.e. large space-bandwidth product (SBP). FPM has enormous potential for revolutionizing microscopy and has already found application in digital pathology. However, it suffers from long acquisition times (on the order of minutes), limiting throughput. Faster capture times would not only improve 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 4x FOV with sub-second capture times. We propose an improved algorithm and 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.
1 Introduction
Source-coded FPM addresses the throughput and motion-artifact limits of conventional large-SBP microscopy by combining coded LED illumination, fast hardware control, and improved phase reconstruction for unstained live samples. It achieves high-resolution, wide-FOV imaging with sub-second acquisition while supporting observations across spatial and temporal scales.
- Motivation: Existing in vitro imaging methods cannot provide the wide FOV, high spatial resolution, and high temporal resolution needed to study rare dynamic cellular events.These requirements motivate a large space-bandwidth-time product (SBP-T) for high-throughput imaging.
- Existing FPM approach: FPM combines a wide-FOV objective with sequentially varied LED illumination to reconstruct resolution beyond the objective’s diffraction limit.Tilted illumination samples different Fourier-space regions, and nonlinear optimization fuses the measurements into a high-resolution image.
- FPM limitations: Long acquisition times and approximately 1 gigapixel datasets limit FPM for real-time in vitro imaging and burden storage and processing.A full 173-LED scan required under 7 seconds after hardware acceleration, but sub-second capture remained necessary for fast subcellular dynamics.
- Source-coded acquisition: Source-coded FPM multiplexes LED illuminations to reduce Fourier-space redundancy, separating brightfield and dark-field LEDs to mitigate noise imbalance.Sequential FPM’s approximately 60% Fourier-space overlap requires approximately 10× more captured data than reconstructed data.
- Performance: 46 megapixels/second SBP-T was achieved for a 4× FOV and 0.8 NA captured in 0.8 seconds, approaching the camera-transfer-rate limit.Some redundancy was retained to ensure robust algorithm convergence, and FOV, resolution, and acquisition time can be traded through illumination-angle range.
- Phase reconstruction: DPC initialization improves recovery of low-frequency phase components in unstained samples, where intensity-only initialization is a poor starting point.Without DPC initialization, reconstructions show a high-pass filtering effect even though both approaches achieve 0.7 NA resolution.
- Applications: Source-coded FPM reconstructs large-SBP videos of growing and confluent cell cultures, revealing subcellular and collective dynamics and rare events across space and time.The demonstrated live-sample videos use label-free quantitative phase and intensity information.
2 Results
The study validates quantitative phase reconstruction for unstained samples and demonstrates fast, large-FOV FPM videos that capture cellular dynamics across multiple spatial and temporal scales.
- Validation with stained and unstained samples: Quantitative phase imaging captures subcellular features in unstained U2OS cells that resemble stained intensity images, supporting a label-free alternative to staining.The unstained intensity image has little contrast, while the phase result clearly reveals cellular structure.
- Validation with stained and unstained samples: DPC initialization recovers low-frequency phase information that intensity-only initialization misses, while both reconstructions achieve 0.7 NA resolution.The recovered low frequencies describe the overall height and shape of cells.
- Fast sequential FPM video of HeLa cells dividing in vitro: Fast sequential FPM records HeLa cell division over 4 hours, including mitosis, cell detachment, four-way division, and actin-filament formation.FPM’s longer depth of field keeps detached cells in focus across the field of view.
- Fast sequential FPM video of HeLa cells dividing in vitro: Automated segmentation identifies approximately 3,400 cells directly from full-FOV quantitative phase images, enabling quantitative analysis of large datasets.The reconstructed video contains approximately 20 gigapixels of quantitative phase data and about 3,000 cells per frame.
- Source-coded FPM video of neural stem cells in vitro: Source-coded FPM achieves 0.8 NA resolution across a 4× field of view in 0.8 seconds using 21 images, enabling NSC videos up to 4.5 hours at 1.25 Hz.The method captures both fast subcellular dynamics and slower population-scale evolution.
- Source-coded FPM video of neural stem cells in vitro: Longer acquisition times blur live-cell dynamics, whereas sub-second source-coded FPM reveals subcellular motion with fewer artifacts, though DPC remains clearer for the fastest processes.The comparison uses schemes with the same nominal 0.8 NA resolution but different capture times.
3 Discussion
Source-coded FPM enables high-speed, large-SBP imaging of live samples, but effective resolution remains coupled to sample motion and acquisition time. The system can be tailored by trading field of view, resolution, and speed according to the dynamics being observed.
- Discussion: Sub-second source-coded FPM captures live-cell dynamics with less motion blur than slower acquisition schemes at the same nominal 0.8 NA.Slower schemes blur small-scale structure and dynamics, while source-coded FPM preserves more detail in live samples.
- Discussion: 0.8-second source-coded FPM reveals more MCF10A cytoskeletal and vesicle dynamics than 60-second or 7-second sequential FPM, though some motion blur remains.The fastest MCF10A processes still exceed the method’s capture speed, causing information loss relative to high-speed DPC.
- Discussion: Source-coded FPM captures most NSC vesicle transport, process retraction and extension, and organelle motion without the significant blur seen in sequential FPM.Direct comparison is difficult because live cells moved between capture schemes.
- Discussion: The trade-off among field of view, resolution, and time should be selected according to sample dynamics; processes faster than 0.8 seconds require fewer captured images.Speed can be increased by sacrificing field of view or resolution, ultimately approaching DPC for maximum capture speed.
- Discussion: The demonstrated technique provides label-free quantitative phase and intensity imaging for high-throughput in vitro applications across spatial and temporal scales.The authors present source-coded FPM as enabling fast, motion-free imaging of unstained live samples.
4 Methods
The methods combine a programmable LED-array system, DPC-initialized iterative reconstruction, phase-contrast simulation, automated segmentation, and dry-mass analysis. Full-field raw images are processed in overlapping subregions and stitched into high-resolution complex-valued reconstructions.
- 4.1 Illumination system: A custom 32×32 LED array with independently controlled channels provides programmable illumination for FPM acquisition.The array uses 4 mm spacing, a 513 nm central wavelength, and static LED drive through dedicated controller chips.
- 4.2 Reconstruction: The reconstruction first uses DPC to initialize phase, then applies quasi-Newton iterative reconstruction for 3–5 iterations.The iterative step incorporates higher-order scattering and dark-field contributions.
- 4.2 Reconstruction: Four brightfield images are used for deconvolution-based DPC, while the initial intensity image averages corrected brightfield measurements before absorption-transfer-function deconvolution.The DPC reconstruction calculates phase within 2× the objective’s NA.
- 4.3 Full-field reconstruction: Each 2560×2160 full-field image is divided into overlapping 560×560 subregions, reconstructed independently, and alpha-blended into a full-field high-resolution image.Neighboring subregions overlap by 160 pixels on each side, and each reconstruction contains intensity and phase.
- 4.4 Phase reconstruction: Axial intensity images acquired at 17 exponentially spaced positions are reconstructed into phase using a transport-of-intensity-type spatial-frequency fitting algorithm.The axial range extends from −64 µm to 64 µm.
- 4.5 Segmentation and dry mass: CellProfiler segments each frame into cell regions, after which MATLAB extracts cell phase and computes dry mass from phase-derived dry-mass density.The reported dry-mass background fluctuation has a standard deviation of 1.5 pg.
Funding Information
The work was funded by the Gordon and Betty Moore Foundation’s Data-Driven Discovery Initiative.
- Funding Information: Funding was provided by the Gordon and Betty Moore Foundation’s Data-Driven Discovery Initiative through Grant GBMF4562 to Laura Waller.