Abstract
For centuries, microscopes have had to trade field of view (FOV) for resolution. Recently, a new computational imaging technique, termed Fourier ptychographic microscopy (FPM), circumvents this limit in order to capture gigapixel-scale images having both wide FOV and high resolution. FPM has enormous potential for revolutionizing biomedical microscopy; however, it has until now been limited to fixed samples, since acquisition time is on the order of minutes. Live biological samples are continuously evolving on multiple spatial and temporal scales, which can cause motion blur. Here, we present a Fast FPM method to achieve sub-second capture times for FPM results with 0.8 NA resolution across a 4x objective's FOV. We demonstrate the first FPM quantitative phase results for both growing and confluent in vitro cell cultures. Experiments capture real-time videos of HeLa and human mammary epithelial (MCF10A) cell division and migration and subcellular dynamical phenomena in adult rat neural stem cells.
Introduction
In optical microscopy, a fundamental trade-off exists between the Field of View (FOV) and Resolution. High-resolution imaging typically requires high-magnification objectives that restrict the FOV to a few cells. Conversely, wide-field objectives lack the numerical aperture (NA) required to resolve fine subcellular details. Fourier ptychographic microscopy (FPM) was developed to overcome this limitation by using an LED array to illuminate samples from multiple angles, computationally stitching the resulting images in the Fourier domain to produce high-resolution, wide-field gigapixel images.
However, traditional FPM is constrained by long acquisition times—often taking minutes to scan through hundreds of LEDs. For live biological samples (in vitro), this delay leads to motion blur, as cellular processes occur faster than the capture rate. This paper introduces a "source-coded" FPM technique designed to achieve sub-second acquisition, enabling high-speed imaging of live cell dynamics.
Methods: Source-Coded FPM
The primary innovation is a transition from sequential LED scanning to a multiplexed illumination strategy.
Hybrid Illumination Coding
Instead of the traditional 173-image sequential scan, the researchers implemented a hybrid scheme. The system first captures four images using half-circle LED patterns to perform Differential Phase Contrast (DPC) for the brightfield region. Subsequently, random multiplexing is used for the darkfield regions, where eight LEDs are activated simultaneously. This strategy reduces the required number of raw images from 173 to just 21, drastically cutting acquisition time.
DPC Initialization and Hardware
Phase retrieval in FPM is a non-convex optimization problem. While standard FPM starts with a low-resolution intensity image—often insufficient for transparent, unstained cells—this method uses a DPC-based initial guess. This provides a more accurate starting point for the phase reconstruction of live cells. To support this, the authors developed a custom high-speed LED controller synchronized with a scientific camera, achieving a data transfer rate of 138 megapixels per second.
Key Results and Biological Applications
The source-coded FPM system achieved a capture time of 0.8 seconds for a full reconstruction, a significant improvement over the ~7 seconds required by optimized sequential hardware. Using a 4x objective (0.2 NA), the system reconstructed images with an effective 0.8 NA resolution, representing a four-fold improvement in resolving power across the entire FOV.
Quantitative Phase Imaging
The technique demonstrates that quantitative phase imaging is a viable, non-invasive alternative to chemical staining. Reconstructed phase images revealed subcellular features, such as actin filaments and vesicles, with clarity comparable to traditional stains without the need for cell fixation. The researchers successfully tracked the "dry mass" of approximately 3,400 individual HeLa cells simultaneously over four hours, capturing events like mitosis and migration across a massive area.
Discussion and Future Work
While the system achieves sub-second capture, the computational reconstruction remains an offline process, taking approximately 20 minutes for a full FOV on a desktop CPU. Future work will focus on utilizing GPU acceleration to move toward real-time viewing. Additionally, while 0.8 seconds is sufficient for most cellular movements, extremely fast subcellular dynamics still exhibit minor motion blur, suggesting that further trade-offs between FOV and speed may be necessary for specific high-speed biological phenomena.
Summary updated