1. Introduction

Phosphorescence lifetime imaging plays an important role in material sciences and life sciences [1–9]. In recent years, the development of phosphorescence based oxygen sensors has enabled in vivo studies of the microenvironment of biological systems [4,5,10–15]. As the partial pressure of gases in blood holds great significance in many biomedical research disciplines such as neuroscience, stem cell, eye diseases and cancer, phosphorescence lifetime imaging in live biological tissue is becoming a widely employed research tool.

The most common form of fluorescence lifetime imaging is based on point scanning and time domain detection with a fast sensor (e.g., single photon counting APD, PMT). Point scanning methods such as two-photon excited fluorescence imaging have the advances for imaging thick scattering tissue [16]. Following the 3D confined point excitation, all the emitted photon (ballistic and scattered) can be effectively collected by high etendue collection optics and detectors (e.g. PMT with large area and access solid angle). A major problem of using point scanning for phosphorescence lifetime imaging is that the emission lifetime is very long (often tens to hundreds of microseconds) and therefore on average each molecule’s emission flux is very low. Bluntly increasing the laser focal intensity is unable to speed up the signal accumulation and is instead likely to cause photobleaching and measurement artifacts. For imaging less turbid and thinner tissue (e.g., cells and tissue culture, zebrafish larvae, eye), the parallel recording approach is the preferred solution [17,18]. The simplest adaption is to instead use a PMT array detector. In addition to its high cost and low array count (much less than 100), such time-domain detectors are not suitable for accumulating weak signals due to the inherent detector noise. For example, with a few photons detected following each laser pulse excitation, it will take hundreds of reading to accumulate over 1,000 photons, which will aggregate the detector noise through hundreds of read out processes. Intensified cameras can offer a large number of pixels (∼ 1 million) for parallel detection. However, each frame is gated for a defined time point following the pulsed laser excitation. Therefore, it is highly lossy in the time domain.

We propose to combine sCMOS cameras with mechanical streaking [19] and line laser excitation to form a low-cost high-throughput low-noise phosphorescence lifetime imaging system. Mechanical streaking is to utilize a rotating reflective surface to spread an optical beam to different angles. Its speed is less than that of the optoelectronic streak camera but is more than enough for phosphorescence lifetime imaging. Let’s consider a common 5 mm, 26 optical degrees, 8 kHz resonant galvo scanner (e.g., CRS 8 kHz, Cambridge Technology). At the wavelength of 500 nm, the angular optical mode is λ/d (0.1 mrad) where λ is the optical wavelength and d is the galvo aperture. Overall a scan range of 26 degree, we have 4,538 optical modes. At 8 kHz resonance, we will have an average line rate (bidirectional scan) of 16 kHz. So the average scan rate is 72.6 M mode/sec. With the sinusoidal scanning, the ratio of the peak speed to the average speed is π/2. Therefore, the peak scanning speed is 114 M mode/sec near the center of the scanning range. To satisfy the sampling limit, we need at least 228 M data points/sec. Inversely, the sampling step size is ∼4.4 ns for this single galvo single pass configuration, which is a bit slower for typical fluorescence lifetime measurement but more than sufficient for phosphorescence lifetime imaging [3].

For imaging, we will utilize a pulsed line excitation. The mechanical streaking will spread the subsequent phosphorescence emission onto a 2D sCMOS camera. Compared to PMT arrays, sCMOS sensors offer high QE (∼80%), low read noise (∼1 electron) and large format (simultaneous recording over 2,000 lines). More importantly, such a configuration allows the accumulation of weak emission signals over many pulsed excitations followed by single low-noise readout. Each readout of the camera frame completes the lifetime measurement for one line of the 2D image. We will translate the laser line position with respect to the sample to form 2D or even 3D lifetime images.

2. Methods

We implemented the mechanical streaking based lifetime imaging with a line confocal imaging configuration (Fig. 1). A collimated continuous wave (cw) laser beam was focused by a cylindrical lens. A dichroic mirror directed the focused beam onto a single-axis galvo scanner. The horizontal line shaped focus (on yz plane) was positioned precisely on the galvo mirror surface. L2 converted the reflected beam into a vertical line along xz plane, which traveled through the adjustable vertical slit. L3 and the objective lens imaged the slit onto the sample which was supported by a 3-axis piezo stage. The emitted phosphorescence signal traveled backward through the slit which was subsequently imaged by L2 and L1 onto the sCMOS camera (C11440-22CU, Hamamatsu). As the galvo was in constant motion, the phosphorescence emission through the slit was spread onto a 2d image (mechanical streaking). Following the camera readout, we translated the sample along the y axis using the piezo stage and start another exposure. The recording over many lines formed a 2D image.

 

Fig. 1. Optical configuration for mechanical streaking based lifetime imaging. CL, cylindrical lens of 500 mm focal length; L1-3, telecentric relay lenses of focal length 100, 100, and 300 mm, respectively; LP, 600 nm long-pass filter; DM, 580 nm long-pass dichroic mirror; Slit, adjustable mechanical slit (VA100, Thorlabs); Piezo, 3-axis piezo stage (Nano-OP100, MCL).

Download Full Size | PDF

Such a configuration greatly simplifies the lifetime measurement. First, we have the flexibility of using either pulsed lasers or cw lasers. In this work, we only utilized cw lasers. With the galvo scanner in motion, the slit naturally performed the task of temporal illumination gating. Second, there is no need to precisely synchronize the laser illumination with the starting time of the detector. We only need to ensure that the camera exposure time is an integer number of the galvo’s half cycle time (two scans per cycle). Third, averaging repeated exposure becomes a trivial task. All we need is to increase the camera exposure time to capture more scanning cycles followed by a single readout.

The slit was positioned at the center of the galvo scanning FOV, which was also imaged to the center of the sCMOS sensor. Therefore, a scanning cycle produced two streaking measurements on one camera image, one half from center to left and the other half from center to right. Both images contain the same information. For better SNR, we divided the image to two halves, flipped one half from left to right and combined the two half images. For each line of the image, we fit the data with a single exponential function and extracted the decay lifetime.

3. Results

To quantify the system response time, we utilized 1 µm diameter fluorescence beads (FluoSpheres F8821, ThermoFisher) as the sample (Fig. 2). In this measurement, we utilized a 5 mm aperture resonant galvo running at 7.9 kHz with a scan range of 26 optical degrees. With the 100 mm lens (L1 in Fig. 1) and 6.5 µm pixel size considered, the streaking time across each pixel was 5.77 ns near the center of the scan range. In comparison, the synthetic dye typically has a lifetime of a few nanoseconds [20], which surely is faster than our system response. So measuring such fast decaying beads allowed us to quantify the system response. Although our laser (532 nm 200 mW CW fiber laser, MPB Communications) wavelength is far from the peak excitation wavelength (580 nm), the beads had strong emission. The averaged laser excitation power for each pixel was 0.43 nW and the exposure time for each camera frame was 10 ms. The total image [Fig. 2(a)] recording time was 2.56 sec (256 vertical lines). Exponential fitting [Fig. 2(b)] showed a lifetime of ∼14 ns. This suggested that our system was fast enough for measuring the typical phosphorescence lifetime (tens of nanoseconds to hundreds of microseconds).

 

Fig. 2. System response test with 1 µm fluorescence beads. (a) Fluorescence lifetime image of the beads. FOV, 55 × 110 µm². (b) Emission lifetime trace from one pixel of the mage (marked by the white arrow). Exponential fitting shows a lifetime of ∼14 ns.

Download Full Size | PDF

Next we utilized the same configuration for imaging platinum-octaethyl-porphyrin (PtOEP, Exciton) which had been explored as an organic light emitting diode material [2]. We dissolved the PtOEP powder in toluene and dried the solution on glass slide. We imaged the clusters of PtOEP (Fig. 3) using an excitation wavelength of 532 nm. The average excitation power was 16 nW per pixel and the camera exposure time was 100 ms. The total recording time for the image with 256 vertical lines (Fig. 3) was 25.6 seconds.

 

Fig. 3. (a) Lifetime imaging of PtOEP cluster on a glass slide. FOV, 55 × 110 µm². (b) Emission lifetime trace from one pixel (marked by the white arrow). Exponential fitting shows a lifetime of ∼28 ns.

Download Full Size | PDF

Next, we explored the imaging of long emission lifetime molecules (Fig. 4). Experimentally, we air dried 0.2 µm europium luminescent microspheres (FluoSpheres F20881, ThermoFisher) on a glass slide. To accommodate the hundreds of microseconds decay time, we switched the galvo to a 5 mm non-resonant galvo scanner (8315k, Cambridge Technology). We drove the galvo with a 100 Hz symmetric saw tooth signal with a scan range of 4 optical degrees and achieved a streaking time of 4.66 µs per pixel. The absorption peak of europium luminescent microspheres was at 365 nm. We utilized a 405 nm cw diode laser (50 mW CUBE 405, Coherent) as the excitation source. Due to the wavelength mismatch, the emission was rather weak. The average excitation power was 40 nW per pixel and the camera exposure time was 500 ms which effectively accumulated 100 decay traces within a single exposure. The overall recording time for an image with 256 vertical lines was 128 seconds.

4. Discussion

The tens of nW per pixel average power level shows that the line scanning mechanical streak camera based lifetime imaging is particularly useful for using low level illumination to accumulate weak emission signals over many excitation cycles. As the sCMOS camera’s read noise is only ∼1 electron and the dark noise is much less than 1 electron per second, the image recording was very clean, dominated by the shot noise. This is a key advantage over PMT based time domain measurement. For example, the signal at ∼100 ns delay in Fig. 3(b) contained ∼30 photons, which was accumulated over ∼1600 excitation cycles (100 ms). If PMT arrays were used, the signal would contain substantial detector noise aggregated over∼1600 readout and the signal would also be weaker due to PMT’s lower QE. Moreover, the common PMT array detectors only contain tens of detector channels. In comparison, sCMOS offers above 2,000 lines for parallel recording.

The maximum throughput of the mechanical streak design is limited by the sCMOS camera (∼0.5-1.5 G pixels/sec). Modern camera architecture allows great flexibility in increasing frame rate at the cost of column number without affecting the overall data throughput. For phosphorescence measurement, this feature allows using the minimum number of time sampling pixels to achieve the maximum lifetime imaging line rate.

The single galvo in our design scanned both the excitation beam and the emission beam. With the slit aperture, we were able to use cw lasers for lifetime recording although its power usage efficiency was lower than that of pulsed laser sources as the excitation beam was blocked by the slit during the phosphorescence recording. However, as cw laser sources are widely available (e.g., in common confocal microscopy and light sheet imaging systems) and often offer greater flexibility in power and wavelength selection, such a design can greatly facilitate the adoption of phosphorescence lifetime imaging.

We used a single slit in the experiment. Scanning the beam across in one direction only produced a single illumination and lifetime trace. Towards greater throughput and efficiency, multiple parallel slits (e.g., created by photolithography) and multiple laser lines (e.g., generated by diffractive optics) can be used. Even if we use 100 sampling points for a lifetime trace, we will be able to record at least 20 spatially separated lines on a single sCMOS frame.

For resonant galvo based measurement, the scanning path across the entire FOV is a sinusoidal function of time. In this work, we only utilized the tens of pixels in the middle, which was highly linear. For applications that will use larger portion of the FOV (e.g., the multi-slit multi-line excitation configuration), we will need to convert the pixel to time based on the sinusoidal function before executing the exponential curve fitting.

 

Fig. 4. (a) Lifetime imaging of europium luminescent microsphere cluster on glass slide. FOV, 55 × 110 µm². (b) Emission lifetime trace from one pixel (marked by the white arrow). Exponential fitting shows a lifetime of ∼340 µs.

Download Full Size | PDF

5. Conclusion

Towards high-throughput parallel detection, we have developed a line scanning mechanical streak camera based phosphorescence lifetime imaging system. Using a confocal line detection configuration, we imaged samples with emission lifetime ranging from tens of nanoseconds to hundreds of microseconds. A key advantage over common time domain detector based recording is the capability of accumulating weak signals through many excitation cycles followed by a single readout, which is particularly important for imaging long lifetime phosphors (with inherently low emission photon flux). The capability to work with low excitation laser power will be highly valuable for life science applications [17,18]. In addition to confocal line imaging, this method can potentially be adapted for line scanning light sheet imaging. As the new generations of sCMOS camera technologies are offering even better QE, noise level, format, throughput and cost, the line scanning mechanical streak design will be highly valuable to a broad range of applications.