High-fidelity microsecond-scale cellular imaging using two-axis compressed streak imaging fluorescence microscopy

TL;DR

This work addresses the bottleneck of CSI in cellular fluorescence imaging where high streak compression and motion blur limit detection of subtle, fast dynamics in continuously illuminated samples. The authors introduce TACSI, a two-axis approach that translates a conjugate image across the coded aperture to reduce compression and blur, backed by a forward model and a $CR_{2D}$ compression-ratio expression. They demonstrate substantial gains in temporal resolution (down to ~$30\,\mu$s), higher BUR, and improved fidelity in simulations and with real data from beads and CHO-K1 cells expressing voltage-sensitive dye under microsecond electric pulses, enabling visualization of membrane-potential dynamics previously inaccessible with single-axis CSI. The method does not require new cameras or lossless encoding and has broad implications for high-speed fluorescence imaging, including studies of action potentials, muscle contractions, and enzymatic reactions at microsecond timescales.

Abstract

Compressed streak imaging (CSI) is a computational imaging strategy that can acquire video at over 150 trillion frames per second. Despite this achievement, CSI faces challenges in detecting subtle intensity fluctuations in slow-moving, continuously illuminated objects. This limitation, largely attributable to high streak compression and motion blur, has curtailed the broader adoption of CSI in cellular fluorescence microscopy. To address these issues and expand the utility of CSI, we developed a two-axis compressed streak imaging (TACSI) method that results in significant improvements to the reconstructed video fidelity. TACSI introduces a second scanning axis which shuttles a conjugate image of the object with respect to the coded aperture. The moving image decreases the streak compression ratio and produces a "flash and shutter" phenomenon that reduces coded aperture motion blur, overcoming the limitations of current CSI technologies. This approach is supported with an analytical model describing the TACSI compression ratio, along with simulated and empirical measurements. We demonstrate TACSI's ability to measure rapid variations in cell membrane potentials, previously unattainable with conventional CSI. This method has broad implications for high-speed photography, including visualization of action potentials, muscle contractions, and enzymatic reactions that occur on microsecond and faster timescales using fluorescence microscopy.

Figures (16)

• Figure 1: Compressed streak imaging conceptual diagram and example images. This diagram shows the encoding scheme required to capture a digital single- or two-axis encoded streak image. Each colored tile represents a time step in the dynamic scene. A representative widefield image of a fluorescent CHO-K1 cell and a brightfield image of the CA are presented as examples. The encoded spatiotemporal streak image of the cell for the single- and two-axis modality can be observed, along with single frames from the reconstructed video. While lenses and mirrors are depicted for clarity, alternative optical elements or arrangements achieving the same conceptual outcome are equally valid.
• Figure 2: Comparison of compression ratios between single-axis and TACSI streak imaging systems. (A) Compression ratio calculated for a fixed 512 by 512 pixel scene acquired on a 2048 by 2048 pixel sensor, with varying numbers of frames. The streak trajectory is oriented vertically (y-axis), and the object trajectory horizontally (x-axis). (B) Compression ratio calculated with varying numbers of row pixels spanning the rectangular scene, while keeping the number of column pixels fixed at $N_{x}=512$ and the number of frames at $N_{t}=250$. (C) Compression ratio calculated with varying numbers of column pixels spanning the scene, while keeping the number of row pixels constant at $N_{y}=512$ and the number of frames at $N_{t}=250$. (D) Compression ratio as a function of streak ratio for different numbers of frames, calculated for a fixed 512 by 512 pixel scene. The maximum recoverable frames are limited by Equation (\ref{['eq:two-axis_frame_limit']}).
• Figure 3: Comparison between single- and two-axis compressed video reconstructions of simulated and empirically measured 10 $\mu m$ fluorescent microbeads with a 10 kHz intensity modulation. Sub-figures (A) and (B) compare several frames from the simulated two-axis and single-axis video reconstructions spanning 1 period of modulation, respectively. Sub-figure (C) shows the temporal profile of the average simulated microbead intensity within a 10 $\mu m$ circular region centered on the bead. The error bars are in standard deviations. A single frame from the high resolution video simulation can be observed in sub-figure (D), and sub-figures (E) and (F) show the single- and two-axis simulated streak images, respectively. Sub-figures (G) and (H) show several reconstructed video frames from the empirically measured fluorescent microbeads spanning 1 period of modulation for two-axis and single-axis streak, respectively. Sub-figure (I) shows the average measured intensity within a circular ROI with a 10 $\mu m$ diameter centered on the beads (N=9). Error bars are in standard deviations. Sub-figures (J-L) show a widefield, single-axis streak, and two-axis measured streak image, respectively, containing the same microbead seen in sub-figures (G) and (H). White horizontal scale bars represent 5 $\mu$m, and vertical scale bars represent 100 $\mu s$.
• Figure 4: Accuracy, temporal resolution, and sensitivity of TACSI versus single-axis CSI. (A) Relative intensity change over time for TACSI from reconstructions of fluorescent micro-beads with sinusoidal modulation amplitude varied from 10-90%. (B) Same as (A) but for single-axis CSI. (C) Temporal resolution evaluated as fluorescent micro-bead contrast versus laser power modulation frequency. TACSI: black circles ($N=3$) with red dashed fit ($R^2=0.997$, $NRMSE=0.021$); single-axis CSI: blue circles ($N=3$) with orange dashed fit ($R^2=0.98$, $NRMSE=0.077$). TACSI reaches 50% contrast at 33 $kHz$ (temporal resolution $\sim$30 $\mu s$); single-axis CSI remains below 50% over the measured range. (D) Sensitivity evaluated as fluorescence micro-bead contrast versus the relative change in the laser power modulation at 10 kHz. TACSI: black circles ($N=3$) with red dashed linear fit (slope = $2.5\times10^{-3}$, $R^2=0.96$, $NRMSE=0.082$); single-axis CSI: blue circles ($N=3$) with orange dashed fit (slope = $9.0\times10^{-4}$, $R^2=0.98$, $NRMSE=0.063$). Error bars represent the standard error of the mean.
• Figure 5: High speed video reconstructions showing cell membrane potential responses to PEFs. Comparison between (A) conventional wide-field fluorescence images and (B) averaged two-axis reconstructions of CHO-K1 cells labeled with voltage sensitive dye. Sub-figure (C) presents multiple two-axis reconstructed video frames of CHO-K1 cells during electric pulse delivery, while sub-figure (D) displays several single-axis reconstructed frames of the same process. The temporal profiles of the 4 colored ROIs in last panel of sub-figure (C) can be observed in sub-figure (E) ($n=1$) along with an oscilloscope trace showing the measured voltage at each time point. The red (cathode) and orange (anode) ROIs are positioned at regions of the cell membrane orthogonal to the electric field vector, while the magenta (left) and cyan (right) ROIs are positioned near regions of the membrane parallel to the field vector. The fluorescence response of the membrane at the cathode and anode decreases and increases, respectively. The left and right side of the cell shows no response to the field. The white rectangular scale bars represent 5 $\mu m$.