1. Introduction

Labeling biological samples with long lifetime luminescent probes such as lanthanides [1–6] or platinum [7] compounds has some distinct advantages over using fluorescence dyes as it allows for easy discrimination from cellular autofluorescence. [2,5,6] In addition, long lifetime Ruthenium dyes are used as oxygen sensors, which provide an important read-out of the metabolic state of cells. [8]

Time-resolved imaging in this context is advantageous because it allows probe concentration and quenching effects to be separated, since the decay is generally independent of the probe concentration. [9, 10] It also allows imaging multi-exponential decay kinetics which are inaccessible via intensity-based imaging. [9] Scanning approaches are impractical for time-resolved imaging of long lifetime probes due to pixel dwell-time issues, so time-resolved wide field imaging is carried out in the time domain with gated image intensifiers [11, 12] or directly gated CCD-cameras. [13] However, the light level must be high enough so it can be gated and the signal outside the gate is lost, compromising the available photon budget.

We present a novel solution to this dilemma: after one excitation pulse, we simultaneously record the arrival time of hundreds of individual photons in a wide-field imaging system. This approach combines ultra-fast wide-field imaging with single photon sensitivity and parallel arrival timing in each pixel with microsecond time resolution.

Photon counting imaging is a well-established low-light-level imaging technique, particularly in astronomy, both on the ground [16, 17] and in space. [18, 19] The Hubble Space Telescope’s Faint Object Camera, and, more recently, the European Space Agency’s X-ray Multi-Mirror satellite were fitted with a photon counting imaging optical monitor. [19] Linearity, high dynamic range, high sensitivity, zero read-out noise, large area, well-defined Poissonian statistics and good spectral response in the UV are particular strengths of the photon counting imaging approach. [20] Applications of photon counting imaging to diverse fields such as autoradiography [14], bioluminescence [15] and fluorescence imaging [21] have also been reported. Photon counting imaging has another distinct advantage over CCD-based imaging: the ability to time the arrival of individual photons in each pixel. The time resolution is given by the frame rate of the camera, typically video frame rates (60 Hz), which yields a millisecond time resolution. [24] However, we show here that frame rates of 40 kHz (and up to 500 kHz) can also be used, allowing, firstly, 1000 times faster acquisition and, secondly, parallel photon arrival timing with microsecond resolution and reduced photon pile-up.

2. Experimental set-up

Our system incorporates an image intensifier, a 40 mm diameter dual proximity-focused, three-microchannel plate device (Photek) operating in photon counting mode, as described previously. [21–23] The phosphor screen of the intensifier was imaged using a 50 mm focal length photographic lens (Canon, F=1/1.2), and the P20 phosphor decay time (to 1/10 of its peak value) is quoted as 250 μs by the manufacturer.

A Photron Fastcam SA 1.1 CMOS camera acquired up to 40,000 frames per second (fps) with an image size of 256×256 pixels. Higher frame rates of up to 500,000 fps are achievable with a reduced image size. Due to the high frame rate, the images cannot be transferred to the PC as they are acquired: they are instead recorded in the camera’s 6 GB RAM, and downloaded to the computer once the acquisition is finished. However, a live display for field of view adjustment and focusing is possible. Software written in-house was used to control the camera, transfer, process and save the frames on the PC in parallel. As the downloading time is around 20 ms per frame and the processing time 5 ms, this approach saves a considerable amount of time compared with the downloading-then-processing method. To maximize transfer speed, we used a home-made compressed file format, which drastically decreased the file size, as up to 90% of the pixels can be black, e.g. at very low light levels in the tail end of a decay. The software also allows a real-time preview of the resulting images, i.e. a sample of the raw data saved on the camera can be inspected before they are downloaded to the PC. The processing itself is done by thresholding the frames, and identifying every single photon event using a recursive algorithm: for every pixel above the threshold, the surrounding pixels are checked against the threshold, [22,23] and the total size and intensity of the photon event is determined; overlapping multiple photon events are discarded according to the number of pixels covered. Only the pixel address of the central peak of the photon events is recorded.

The time-resolved acquisition was performed by triggering a pulsed diode laser (Hamamatsu PLP-10, 470 nm, 90ps optical pulse width) from a pulse generator (Thandar TG105) and synchronizing the laser pulses with the camera acquisition. The signal generator was used to synchronizes the camera so as to start the acquisition some tens of frames before the pulse, and to record hundreds of frames after it. A schematic of the experimental set-up is shown in Fig. 1. The pulse width and repetition rate of the diode laser, typically 10–20 Hz, were fixed by a function generator (FG600 Feedback Instruments Ltd) coupled to the pulse generator. The detection system was mounted on the side port of an inverted microscope (Nikon Eclipse TE2000-E), and the samples imaged through a 610 nm long pass filter (Semrock). The laser was illuminating the samples with an average power of less than 1 μW.

 

Fig. 1 Schematic of the experimental set up. The pulsed laser is synchronized with the frame acquisition at a rate of up to 250 kHz. [26]

Download Full Size | PDF

3. Results

3.1. Steady state photon counting imaging

The capabilities of our detector were evaluated by imaging a flat field at a frame rate of 30,000 fps, with the microscope’s tungsten lamp for transmitted light imaging illuminating the focal plane. A single frame acquired at this speed is shown in Fig. 2(a). The photon events (bright spots) can clearly be seen on a dark background. Since there is no interlaced read-out in the CMOS camera, the photon events on the intensifier’s phosphor screen are rendered faithfully onto the camera. Moreover, we do not observe photon event afterglow in successive frames - long phosphor decay components are presumably not bright enough to be detected by the camera. The sum of 12,288 frames results in a flat field as shown in Fig. 2(c). The brighter bottom half on the image is due to non-uniform illumination.

 

Fig. 2 Typical characteristics of the image intensifier performance, obtained by imaging a flat field at 30 kHz frame rate and 256×256 pixels. (a) Single photons on 30 μs exposure time frame. (b) The photon events have different intensities as illustrated by a photon event pulse height distribution. (c) A flat field, the sum of all frames. (d) A photon arrival time plot shows a mean of 174 photons detected per frame. (e) Photon arrival time plot with a logarithmic time axis over five orders of magnitude. (f) Distribution of the number of photons per frame, and the corresponding Poissonian fit.

Download Full Size | PDF

The photon event pulse-height distribution, Fig. 2(b), conforms to the expected one for such a saturated gain device. [14–17, 22–25] However, there may be some loss of photons due to non-optimal optical coupling of the intensifier to the camera. The average number of photons per frame versus time over 20 ms is shown in Fig. 2(d), and a plot on a logarithmic time axis shows that photon arrival timing can be performed over five orders of magnitude, Fig. 2(e). The overall distribution of the numbers of photons per frame is shown in Fig. 2(f). It obeys Poissonian statistics as expected for photon counting. The average count rate of 174 photons per frame results in an overall count rate of around 5 MHz. The effective acquisition time of the image shown in Fig. 2 was about three seconds. These results are in good agreement with conventional video-rate photon counting imaging systems [14–17, 22–25] except that our system runs 1000 times faster. The key point is, however, that it allows the timing of photon arrival in hundreds of pixels in parallel with microsecond time resolution. We note here that the camera was sensitive and accurate enough to detect single photon events at frame rates of 250,000 fps but with a resolution reduced to 128×80 pixels, and phosphor persistence in up to six successive frames. Although this limits the local count rate, the phosphor persistence is invariant and this feature can be exploited to perform photon arrival timing with sub-exposure time resolution. [26]

3.2. Time-resolved photon counting imaging

Europium-containing polyoxometalate (POM) nanoparticles [27, 28] on a glass slide were imaged through a 40× objective (NA 0.6) which should yield a spatial resolution of around 600 nm. The MCPs have 12 μm pores on 15 μm centers [23], so the optical resolution of this system is limited by the NA of the microscope objective. The camera operated at 40,000 fps and an image size of 256×256 pixels. This set-up was used to fill the camera memory with 87296 frames. A lifetime image of POMs is shown in Fig. 3(a), with a good monoexponential decay and an average lifetime of 2.9 ms, as indicated by the lifetime histogram. Spatially averaged results of dry POMs are shown in Fig. 3(b), where the number of photons in each frame has been counted, the 256 excitation cycles added and plotted versus the time, with a time channel width of 25 μs. Quantitative analysis by fitting the data with an exponential decay yields a decay time of 2.98 ± 0.20 ms, in excellent agreement with previous work using an established method. [29]

 

Fig. 3 Using a pulsed laser, the arrival times of photons after an excitation pulse can be determined with microsecond resolution in all pixels in parallel. (a) A Eu-POM lifetime image and lifetime histogram. The scalebar is 100 μm. (b) A decay of Europium-containing polyoxometalate (POM) nanoparticles [28] after 256 excitation pulses averaged over the whole field of view. (c) Distribution of photons detected after a single laser excitation pulse, showing up to 5 detected photons on a single pixel in an image. (d) Semi-logarithmic bar graph of the overall distribution of all detected photons in (c).

Download Full Size | PDF

Apart from timing hundreds of photons in different pixels after one excitation pulse, a further distinct advantage over conventional picosecond time-correlated single photon counting (TCSPC) [30] is that our method can time the arrival of several photons per pixel after one excitation pulse (as long as they do not arrive in the same frame). This is shown in Fig. 3(c), with up to 5 photons detected on a single pixel in an image after one excitation pulse.

A representative example of the overall detected photon distribution, after 256 excitation pulses, is shown in the semi-logarithmic bar graph Fig. 3(d). In 10⁷ pixels no photon is detected, and in 714,022 pixels exactly one photon is detected. Moreover, in 86,015 pixels 2 photons are detected. In conventional picosecond TCSPC, only the first of these photons would be timed. Our system, however, is capable of timing both. In addition, in 13,310 pixels we detect 3 photons, and up to 7 photons in 11 pixels, as shown in Fig. 3(d). Altogether, we detect 222, 757 photons that would have suffered pile-up in conventional TCSPC. [30] Thus, our approach prevents the loss of 20% of the total number of photons compared to conventional TCSPC, and consequently also prevents pile-up skewed decays.

4. Discussion

The demonstration that a photon counting imaging system can be run at a frame rate of 40,000 fps opens up the possibility for ultra-fast imaging with single photon sensitivity. This is not possible with conventional CCD or CMOS cameras alone, although sub-exposure time resolution could be achieved, to some degree, by temporal pixel multiplexing, i.e. trading spatial resolution for time resolution. [31] Using a pulsed excitation source, it enables novel two-dimensional, wide-field TCSPC imaging on the microsecond timescale with a typical acquisition time of only a few seconds. Moreover, our method can detect more than one photon per pulse per pixel (but no more than one photon per pixel per frame), i.e. it is a method between TCSPC [30] and multichannel scaling techniques [32]. No photons are lost as in gating approaches, which is particularly beneficial for live cell imaging where phototoxicity through prolonged excitation may compromise the cells. Whilst our time-resolution was initially for luminescence decay times >50 μs, note that the frame rate of this ultra-fast Photron SA 1.1 camera is up to 500,000 fps (at a reduced number of pixels), and at this speed it is still possible to detect individual photon events, if a suitable phosphor, e.g. P46, is used. In this regime it would be possible to detect the decay of Ruthenium, an oxygen sensor. [8] Moreover, it would also allow higher local and global count rates. Thus the combination of the sensitivity and precision advantages of TCSPC with the advantage of high speed in wide-field imaging, as demonstrated in Fig. 3, is a very promising prospect. The method would also allow imaging Fluorescence Correlation Spectroscopy, where a photon correlation curve could be generated in each pixel, in analogy with Image Correlation Spectroscopy. [33] For a more practical implementation, we would envisage almost real-time data processing by implementing thresholding and photon event detection on the camera, before transferring the data to a computer.