Super-resolution (SR) microscopy techniques, particularly structured illumination microscopy (SIM), have revolutionized biological imaging. SIM offers high speed and spatial resolution (~100 nm), making it suitable for live-cell imaging of dynamic cellular processes. Current limitations to increasing SIM's speed include the trade-off between sCMOS camera readout speed and fringe modulator switching time, which often leads to a smaller effective exposure area and reduced signal-to-noise ratio (SNR). Previous attempts to increase frame rates have focused on algorithmic improvements like rolling reconstruction and reduced frame reconstruction, or on using expensive high-speed cameras. This paper addresses these limitations by introducing PAR-SIM, which aims to fundamentally increase SIM speed at the hardware level, maximizing the use of the sCMOS sensor without requiring expensive camera upgrades or complex algorithms. The need for ultra-high-speed imaging in live-cell studies, particularly for observing rapid organelle movements, is a strong motivation behind the development of PAR-SIM. Existing techniques, including HessianSIM, Sparse-SIM, and GI-SIM, have achieved impressive frame rates, but further improvements are needed to capture even finer details of rapid cellular dynamics. PAR-SIM proposes a solution to this challenge by parallelizing both the acquisition and readout processes of the camera.

The development of super-resolution microscopy techniques has significantly advanced our understanding of biological processes. Structured illumination microscopy (SIM) has emerged as a powerful tool due to its speed and resolution. The paper discusses several existing high-speed SIM techniques, including HessianSIM, which achieves a frame rate of 188 Hz, and Sparse-SIM, which reaches 564 Hz using 3-rolling reconstruction. GI-SIM further enhances the temporal resolution, enabling observation of organelle interactions. However, these methods are limited by acquisition frame rates. The authors highlight that advancements, such as rolling reconstruction and the use of advanced sCMOS cameras and spatial light modulators (SLMs), have improved acquisition rates but still face fundamental limitations imposed by the camera's exposure-readout speed and the SLM switching time. The trade-off between speed and SNR remains a significant challenge. Prior efforts to improve speed focused on parallel computing and GPU-based processing of SIM reconstructions. However, the authors argue that these approaches only partially address the issue and that a hardware-level solution is necessary to achieve significant speed enhancements.

PAR-SIM leverages the rolling shutter mechanism of the sCMOS camera to achieve parallel acquisition and readout. The system utilizes a set of galvanometers to project sequentially time-lapse images of sub-regions of interest (sub-ROIs) onto distinct areas of the sCMOS detector. The SLM generates the structured illumination patterns. Precise synchronization between the SLM, galvanometers, and camera is crucial to avoid crosstalk. In contrast to conventional SIM, where the entire sensor area is exposed and then read out, PAR-SIM exploits the sloped exposure-readout times inherent to the rolling shutter. This allows simultaneous exposure of one set of sub-ROIs while other sub-ROIs are being read out. Specifically, the system projects three sub-ROIs with different SIM phases/angles onto parallel sensor areas, and simultaneously reads out another three sub-ROIs from the previous frame. This results in a six-fold increase in acquisition speed. The system uses a custom-designed LabVIEW program for controlling the synchronization of the various components. The synchronization involves careful control of several timing parameters: the galvo stopping time (T₁), SLM duration (Tᵢ), SLM high voltage duration (Tₛ), and three delay signals (T_D1, T_D2, T_D3) to manage camera exposure delays and galvanometer repositioning. The authors provide a detailed explanation and schematic diagrams illustrating the synchronization procedure and timing parameters. A physical model-based reconstruction algorithm is used to reconstruct high-fidelity images from the low-SNR data acquired at high speed. The setup involves a Nikon Ti2-E microscope with Airy Polar-SIM, an SLM (QXGA/SXGA), and a high-speed sCMOS camera (ORCA-Flash4.0 V3). A custom-designed sub-ROI slit is used to select the regions of interest. The use of two different SLM series (SXGA and OXGA) allowed for the exploration of different trade-offs between speed and resolution.

PAR-SIM achieves an ultra-high spatio-temporal information flux of 132.9 MPixels s⁻¹ at 256 Hz with 100 nm resolution and a low SNR of -2.11 dB. This represents a nine-fold increase in speed compared to state-of-the-art SIM techniques. The system successfully visualizes microtubules, actin, and mitochondria without significant artifacts. The authors demonstrate the ability of PAR-SIM to capture dynamic mitochondrial membrane interactions, such as tubulation and fusion, in live COS-7 cells at a frame rate of 408 Hz. The high temporal resolution allows for accurate measurement of organelle movement direction, displacement, and velocity. Table 1 in the paper details the optimized synchronization parameters (T₁, Tᵢ, Tₛ, T_D1, T_D2, T_D3) and resulting frame rates for both SXGA and OXGA SLMs. The results showcase the effectiveness of PAR-SIM in capturing fast cellular dynamics with high fidelity. The parallel acquisition and readout strategy fundamentally improves the information throughput, showcasing the system's potential to overcome limitations inherent in the camera and SLM technologies.

The results demonstrate that PAR-SIM successfully addresses the key limitations of conventional high-speed SIM by achieving a significant improvement in spatio-temporal resolution without compromising image quality. The parallel acquisition-readout strategy effectively utilizes the entire sensor area and allows for a much faster acquisition rate than traditional SIM methods. The ability of PAR-SIM to capture dynamic processes in live cells, such as mitochondrial fusion and tubulation, underscores its potential for studying rapid cellular events. The development of the physical model-based reconstruction algorithm is critical for obtaining high-fidelity images from the low-SNR data acquired at high speed. PAR-SIM's performance surpasses that of existing high-speed SIM techniques, paving the way for more detailed investigations of cellular dynamics. Future work could explore further optimization of the synchronization parameters and the development of more advanced reconstruction algorithms to further enhance image quality and resolution. The technique could be adapted and applied to other types of structured illumination microscopy.

PAR-SIM presents a significant advancement in super-resolution microscopy by achieving an unprecedented level of spatio-temporal resolution. The parallel acquisition-readout approach, coupled with a robust reconstruction algorithm, enables the capture of dynamic cellular processes at speeds previously unattainable. The high frame rates and spatial resolution allow for precise measurements of organelle movement and interactions. Future directions could include exploring modifications for other types of SIM and expanding its application to various biological systems.

While PAR-SIM achieves remarkable speed improvements, there are some limitations to consider. The system's performance is dependent on precise synchronization of multiple components, requiring careful calibration and control. The effectiveness of the reconstruction algorithm is crucial for obtaining high-quality images from low-SNR data. The field of view is limited by the size of the sub-ROIs, although this can be partially addressed by tiling multiple sub-ROIs. The reliance on a rolling shutter camera may introduce artifacts related to the rolling shutter effect. Further improvements to the algorithm might reduce or correct for these artifacts. The applicability of the technique to different biological samples or imaging modalities might require further optimization and adaptations.