2.1.

Optical System

The diagram of the optical system is shown in Fig. 1. The architecture extends standard widefield endoscopy to dual color while taking advantages of improved imaging and illumination technologies.²⁹ All components are off-the-shelf commercial items. The proximal end of the fiber imaging bundle (Alveoflex™, Mauna Kea Technologies, Paris, France) is brought to the focus of the objective by mounting it onto a 30-mm caged $Z$-axis translation mount (SM1Z, Thorlabs) enabling fine focusing and optimized coupling of the image from the distal end of the fiber to the rest of the optics. The Alveoflex™ imaging fiber has approximately 30,000 cores without any lensing on the tip (i.e., works by contact imaging). Figure 2(a) shows the fiber cores of the Alveoflex™ imaging bundle. Two LEDs at 470 and 625 nm (M470L3 and M625L3, Thorlabs) are used as illumination sources coupled to achromatic condenser lenses (ACL2520-A, Thorlabs for 470-nm LED and ACL2520-B, Thorlabs for 625 nm LED), see Fig. 2(b) for LED spectra. The collimated beams are combined using a dichroic beamsplitter (FF506-DI03-25X36, Semrock). Two-color epifluorescence is achieved using a two-color filter-set (emission, dichroic, and excitation filters in XF454, Horiba, UK) before being directed to an infinity-corrected microscope objective with NA 0.3 and working distance 10 mm (RMS10x-PF, Thorlabs). Emission and excitation filter spectra are shown in Figs. 2(b) and 2(c), respectively, along with normalized absorption and emission spectra of example fluorescent probes used in this study (PKH67 and Cellvue^® Claret; see text below).

Fig. 1

Diagram of the two-color fluorescence system. Two LEDs are combined with a dichroic mirror, and illumination is sent to the microscope objective via the emission filter and another two-band dichroic. Fluorescence from the imaging bundle is focused onto the color CMOS camera via a tube lens (200 mm focal length).

Fig. 2

Alveoflex™ fiber bundle as acquired by the camera (a) with a zoomed inset to the right showing fiber cores (bright) and cladding (dark regions between the cores). Image in (a) was acquired at full resolution ($1920\times 1200$) in monochromatic mode to improve sampling of the cores. LED sources spectra and emission filters absorption curves are shown in (b), while in (c) the fluorescence emission curves are overlaid with emission filter and Bayer filters.

Fluorescence exiting the microscope objective is imaged onto the color CMOS camera (GS3-U3-23S6C-C, Grasshopper3, Point Grey Research, Canada). A color imaging camera used in conjunction with an appropriate multiband filter set was chosen to simplify the optical design for application in a clinical environment. The camera has a universal serial bus 3 (USB3) connection to the computer allowing up to 162 frames per second (fps) at full resolution, $1900\times 1200$. The color filter is the standard Bayer pattern filter [see Fig. 2(c)] deployed in many consumer, machine vision, and microscopy cameras. The CMOS camera sensor is a Sony IMX174 which has seven photoelectrons (${\mathrm{e}}^{-}$) read noise.³⁶ We refer to the green and red channel as the green (510 to 560 nm) and red fluorescence (660 to 700 nm) as captured from the setup onto the camera and filtered by the Bayer filter on the CMOS chip. As shown on Fig. 2(c), the Bayer color filter overlaps with the green channel of the two-color filter-set used (XF454, Horiba, UK), while the overlap in the red channel is smaller. Smaller overlap in the red channel is offset by greater separation of the two channels ($\sim 100\mathrm{nm}$), which reduces spectral mixing.

2.2.

Optical System Characterization

The system performance was characterized by measuring the lateral spatial resolution, the spectral throughput of illumination and fluorescence collection, chromatic focus shift, verification of the camera read noise, and performance at high-frame-rate acquisition (e.g., $>30\mathrm{fps}$). The aim was to achieve the optimal contrast on targets ranging in size from bacteria (0.5 to $2\mu \mathrm{m}$) to cells, such as monocytes and neutrophils (10- to $15\mu \mathrm{m}$), and to fungi ($100\mu \mathrm{m}$ or more). The lowest detectable light intensity defines the visibility of the object against the noise floor, while chromatic aberrations potentially affect contrast levels for each spectral range detected. The required frame rate for optimal assessment in fluorescence microendoscopy is suggested as 10 to 15 fps.³⁷ While this is easily achievable with CMOS camera technologies in terms of speed and noise,³⁸ it is imperative to assess how such frame rates compromise the limit of detection. High frame rates could be applied to tissue elastography and assessing alveolar compliance. In this study, frame rates of 12–20 fps and camera resolution $960\times 600\text{pixels}$ were chosen to readily permit real-time (video rate) visualization by operators. Commercial microendoscopy systems operate at this rate.²⁰ Higher camera resolution ($1920\times 1200\text{pixels}$) was used whenever fiber core visibility was needed. All tissue imaging was done at $960\times 600$, unless stated otherwise. Higher frame rates up to 200 fps were also evaluated, which is the upper limit imposed by the hardware, notably the camera, memory transfer speeds, and hard disk space available on the computer.

Optimal spatial resolution was measured by contacting the distal end of the Alveoflex™ fiber to a standard chrome on glass USAF 1951 target (Thorlabs, part number R1DS1N). This measurement was done in transmission (through the USAF target) by allowing the fluorescence from a target (fluorescent block, Chroma) in the spectral range 510 to 560 nm (Thorlabs, MF510-42) to pass through the test target having been illuminated via the excitation delivered through the Alveoflex™ fiber and the target imaged back through the optical system in the conventional manner. An intensity line profile was taken over the image obtained from the test target to estimate contrast reduction with increasing resolution.

Spectral throughput measurements were undertaken by inserting a fiber probe connected to the spectrometer ($200\mu \mathrm{m}$ fiber connected to a USB2000 spectrometer, Ocean Optics) on the following locations: (1) distal side of Alveoflex™ (in front of the light source), (2) proximal side of Alveoflex™, and (3) the camera imaging plane. The distal end of Alveoflex™ was illuminated by a white LED lamp. Fluorescence collection efficiency was measured by illuminating the distal end of the fiber by a white LED measuring the optical power at (1) the proximal side of Alveoflex™ and (2) the camera imaging plane.

The limit of detection (LOD) for fluorescein (Sigma-Aldrich) and Cy5 (Sigma-Aldrich) solutions was determined. 1, 10, 50, and 100 nM water solutions were prepared. The Alveoflex™ was inserted into the solution, and 10 images were taken for each concentration with the exposure time set to 80 ms. The background image was obtained by illuminating the Alveoflex™ by both LEDs as in normal operation, but without any fluorescence or stray light entering the tip. Fifty background images were saved and their mean subtracted from images acquired from solutions. A $40\times 40\text{pixel}$ area was selected from the central area of the Alveoflex™ imaging fiber bundle. Mean and standard deviation were calculated for the area, for a single 80 ms exposure time image indicating signal-to-noise (SNR).

Chromatic effects are of particular importance for multicolor fluorescence microendoscopy. We assess the chromatic effects by loading the distal end of the Alveoflex™ with 0.3% relative intensity standard fluorescence microspheres (green Inspeck™ I-7219 and red Inspeck™ I-7224, Life Technologies). Fluorescence microspheres in solution readily attach to the distal tip, and this provides a good target for evaluation of contrast and contrast degradation. To optimize the focus of fluorescence onto the camera, we adjust the position of the tube lens with respect to the CMOS sensor camera and the position of the proximal fiber end with respect to the microscope objective. The tube lens was prealigned as in Ref. 26 and kept fixed. The proximal end of the Alveoflex™ was moved to three locations. Optimal focus for green fluorescence was found and moved $+/-5\mu \mathrm{m}$ around this location.

Camera read noise was measured by taking 100 bias frames (0 ms exposure time) in dark room conditions with a metal C-mount cap on the sensor.³⁹ Camera gain was set to 0 dB and the camera video mode to 7 (lowest noise mode for this particular camera⁴⁰). The mean of all bias frames was subtracted from a sample bias frame and the standard deviation provides the read noise in analog-to-digital units (ADU). We used $0.51e^{-}/\mathrm{ADU}$ as the system gain⁴¹ to derive the read noise in ${\mathrm{e}}^{-}$. We also used the monochrome version of the sensor for read noise measurements (part number GS3-U3-23S6M-C, Point Grey Research, Canada). Dark noise was measured by taking a 3 s exposure time image in dark room conditions as above. The average bias frame was subtracted from the 3 s exposure frame, and the standard deviation of the resulting image was divided by 3 to obtain the dark noise for a 1 s exposure.

For all experiments, a background image was acquired prior to imaging sessions and subtracted from the acquired images. The resulting image was contrast enhanced to allow for optimal illustration of image SNR. No frame averaging or image processing was performed, although the potential improvement of image processing is discussed further below.

2.3.

Microsphere and Tissue Experiments

Current methods to determine the bacterial burden in suspected ventilator-associated pneumonia (VAP) rely on the culture of BALF. The growth of colony forming units (CFU) above ${10}^4\mathrm{CFU}/\mathrm{ml}$^42,43 or ${10}^5\mathrm{CFU}/\mathrm{ml}$^44,45 is considered diagnostic of VAP. These are the accepted clinically relevant limits, and we have based our system on being able to perform at this limit of detection. As broncholavolar lavage involves the instillation and retrieval of fluid, a dilution effect is seen. The true bacterial burden in the infected distal lung is likely to be ${10}^5$ to ${10}^6\mathrm{CFU}/\mathrm{ml}$.⁹ Therefore, to establish the suitability of the instrument for the lower limit of bacterial detection in the distal lung, we initially performed experiments in two scenarios whereby we used microspheres as surrogates for bacteria: (a) imaging of microspheres alone in black Eppendorf tubes (to avoid tube fluorescence) and (b) imaging of microspheres in ex vivo lung tissue placed in a well plate. 0.3% Inspeck™ $2.5\mu \mathrm{m}$ microspheres (green Inspeck™ I-7219 and red Inspeck™ I-7224) were used as a fluorescence emission standard indicative of labeled bacteria. Green 0.3% Inspeck™ microspheres have equivalent emission to green OMI smartprobe⁷ labeled bacteria (Pseudomonas aeruginosa), and red 0.3% Inspeck™ microspheres are 1.5 times weaker than PKH red (Cellvue^® Claret, Sigma-Aldrich) labeled bacteria (Pseudomonas aeruginosa). Blackened microtubes were filled with concentrations ${10}^3/\mathrm{ml}$, ${10}^4/\mathrm{ml}$, ${10}^5/\mathrm{ml}$, and ${10}^6/\mathrm{ml}$ to derive the detection limit of the system without the lung tissue present. Tubes were centrifuged for 10 s before each experiment. Excised lung tissue was placed in well plates with 0.3% Inspeck™ $2.5\mu \mathrm{m}$ microspheres deposited at the following concentrations: ${10}^3/\mathrm{ml}$, ${10}^4/\mathrm{ml}$, ${10}^5/\mathrm{ml}$, and ${10}^7/\mathrm{ml}$. Controls included excised lung tissue alone, ${10}^6$ empty microspheres (unstained control beads supplied with green Inspeck™ I-7219, Life Technologies) per milliliter in lung tissue, distilled water alone and lastly, ${10}^6$ empty microspheres per milliliter in distilled water. The controls were used to minimize false positive detections, especially in the green channel, where small parts of tissue could be mistaken for bacteria.

Each sample was imaged for 30 s. The fiber was moved every 2 to 3 s to image as much of the volume as possible in the limited observation time. Detection of weakly fluorescent microspheres was assessed visually through user interpretation of the real-time images acquired during the 30 s experiment. A successful detection was defined as visual identification of at least a single target with pattern recognition similar to those in the controls. The camera settings were set as follows: 8 bit image acquisition, gain 24 dB, exposure time 80 ms (12 fps).

2.4.

Labeled Targets, Tissue, and Bacterial Imaging Experiments

Excised human lung tissue was placed in well plates with five experimental conditions: (1) freshly isolated human monocytes labeled with PKH red (Cellvue^® Claret, Sigma-Aldrich); (2) freshly isolated human neutrophils labeled with PKH green (PKH67, Sigma-Aldrich); (3) Aspergillus fumigatus labeled with PKH red; (4) labeled monocytes, neutrophils, and fungus (Aspergillus fumigatus) together in single well; and (5) bacteria (Pseudomonas aeruginosa) at ${10}^7\mathrm{CFU}/\mathrm{ml}$ labeled in PKH green (PKH67, Sigma-Aldrich).

Neutrophils and monocytes were isolated by discontinuous percoll gradients from the peripheral blood of human volunteers using methods previously described.⁴⁶ As mentioned above, bacteria used were Pseudomonas aeruginosa (ATCC 47085), and fungi used were Aspergillus fumigatus (clinical isolate, Royal Infirmary of Edinburgh, Scotland). Bacteria or fungi were counterstained with PHK67 or Cellvue Claret dyes (Sigma-Aldrich, St Louis, MO) in accordance to the manufacturer’s instructions. Briefly, cells were washed three times in PBS and re-suspended in $500\mu \mathrm{L}$ of Diluent C, then added to $500\mu \mathrm{L}$ of Diluent C containing $2\mu \mathrm{L}$ of stock dye. Following 3 min, bovine serum albumin was added to terminate the reaction, and cells were washed a further three times and re-suspended as required for experiments. Human tissue was used with regional ethics committee (REC: 13/ES/0126) approval and was retrieved from the periphery of specimens taken from lung cancer resections. Monocytes and neutrophils were isolated and purified from peripheral blood of healthy volunteers (REC: 08/S1103/38). Each sample was imaged for 30 s, and the camera settings were as follows: 8 bit image acquisition, gain 24 dB, exposure time 80 ms (12 fps). PKH red and green were used as per manufacturer’s instructions.

Well plate 5 with PKH green labeled Pseudomonas aeruginosa at ${10}^7\mathrm{CFU}/\mathrm{ml}$ was additionally imaged at 5 ms exposure time (200 fps) to study the dynamics of fiber core blinking events which are observed when imaging objects smaller than the fiber core. Furthermore, to study the effect further, we dissolved $5\mu \mathrm{L}$ of Constellation™ microspheres (C-14837, Life Technologies) in $500\mu \mathrm{L}$ of distilled water and imaged the solution in (1) free space—fiber bundle was removed from the setup in Fig. 1 and $10\mu \mathrm{L}$ solution was placed on the slide and covered by a coverslip—and (2) with fiber bundle, that is, conventional microendoscopy imaging. Constellation™ microspheres have a range of sizes and colors and are a good candidate for cross comparison of imaging effects on small (less than fiber core size $\sim 3\mu \mathrm{m}$) and large objects (greater than several fiber core sizes $\sim 10\mu \mathrm{m}$). Fiber bundle imaging of Constellation™ microspheres was performed in a microtube, as contact with the glass slide could potentially damage the fiber bundle tip.

3.1.

Optical System Characterization

Below, we present the measurement results for system spatial resolution, camera read noise, fluorescence collection, fluorescence spectral throughput, chromatic focus shift, and an evaluation of high-speed performance on ex vivo lung tissue.

Figure 3(a) shows the USAF test target with an even fluorescent illumination, and Fig. 3(b) shows the line profile across group 6 elements where the line thickness of the smallest resolvable line (group 6, element 6) is $4.38\mu \mathrm{m}$. Individual test chart lines are clearly distinguished at this resolution. Fluorescence ranging from 510 to 570 nm is imaged and focused in Fig. 3. Fiber core size, cladding size, and core to core coupling⁴⁷ limit the system’s spatial resolution and hence image quality; see Fig. 2(a) for a demonstration of fiber core packing density. Read noise was measured to be $8e^{-}$ and dark noise $1e^{-}/\mathrm{s}$. As the frame rates used are at least 12 fps, the influence of dark noise is minimal. The signal is read noise dominated for the detection of up to about $400e^{-}$, where read noise is 35% of the Poisson noise. The uniformity of the field of view (FOV) is as follows: for the green channel, edge intensity is 54% of the central intensity, and for the red channel, edge intensity is 76% of central intensity. While future work will need to address the uniformity of the green channel, in the current systems this did not seriously affect the imaging quality. The optical power exiting the distal fiber and fluorescence collection efficiency are summarized in Table 1. The maximum optical power coupling to Alveoflex™ for 470 nm LED is 2.4 mW, but we limit this to 1.8 mW due to the intrinsic red autofluorescence of the Alveoflex™; see fiber background discussion below. For 625 nm LED, the optical power measured at the distal end of the fiber is 3.4 mW. Total optical power is within acceptable limits of maximum exposure as defined in Ref. 48. For fluorescence collection, we measured 53% efficiency for both channels.

Table 1

Illumination and fluorescence collection efficiency.

Fig. 3

(a) Image of a USAF 1951 test target illuminated with an even fluorescent signal and (b) the line profile across group 6 elements. Experiment performed at 510 to 560 nm spectral range, 8 bit image, 80 ms exposure time (12 fps), gain 15 dB, $1920\times 1200$ camera resolution.

Spectral throughput is analyzed in Fig. 4. Fiber fluorescence throughput is above 85% across the spectral range studied, 500 to 750 nm. System fluorescence throughput was measured at the camera image plane and includes attenuation by the fiber bundle, objective, dichroic, emission filter, and the mirror (see Fig. 1). It reflects what would be expected from the dichroic and emission filters used. Figure 4 shows Bayer filter spectra superimposed to illustrate expected reduction of the signal onto the camera.

Fig. 4

Spectral throughput analysis. Although spectral attenuation is not significant to the camera plane, the Bayer filter does attenuate the red channel.

SNR analysis of the uniform solutions of fluorescein and Cy5 are given in Table 2. Fluorescein has a LOD concentration of 10 nM, and Cy5 has a LOD concentration of 50 nM. Other solutions, such as in Ref. 21, obtain lower detection limits, predominantly due to deploying several EMCCD cameras with optimal spectral ranges on each camera, whereas we have a single color camera with a higher inherent read noise. We verified this by using a scientific CMOS camera (PCO.edge mono 4.2LT, PCO AG, Germany) and LOD improved to 1 nM for fluorescein and 10 nM for Cy5.

Table 2

SNR for FITC and Cy5 solutions.

Figure 5(a) shows the contrast degradation with defocus for green and red fluorescence. Defocus is obtained by moving the proximal fiber side. Figure 5(b) shows the line profile across the green microspheres on the green channel. It indicates that the $0\mu \mathrm{m}$ position is the optimal focus for green microspheres, judging by contrast. Figure 5(c) shows the line profile across the red microspheres on the red channel. It demonstrates that $5\mu \mathrm{m}$ is the optimal focus for the red microspheres. This axial displacement for the red channel over the green channel is likely due to chromatic aberration. As indicated previously,⁴⁹ some imaging fiber bundles have detectable fiber background autofluorescence. The modulation on Fig. 5(c) for the $5\mu \mathrm{m}$ position is due to red fluorescence from our fiber cores and provides further evidence that this is the optimal focus for the red channel. The compromise is to place the proximal end of the Alveoflex™ between the two optimal focal locations, minimizing the chromatic effect with our 0.3 NA microscope objective. The presence of chromatic aberration is well understood and is, in part, a consequence of a single channel detection platform. However, by placing the proximal end of the Alveoflex™ between the two optimal focal locations, we minimize the chromatic effect with our 0.3 NA microscope objective, and image reconstruction is not compromised significantly for either of the imaging channels.

Fig. 5

Reduction in contrast for individual red and green microspheres is shown. (a) Visual appearance of dots at different focus shifts (proximal fiber side). (b) The line profile of the green channel across the green microsphere. (c) The line profile of the red channel across the red microsphere. Exposure time was set to 40 ms (25 fps), gain 24 dB, 8 bit image, $960\times 600$ camera resolution.

Figure 6 demonstrates fast image acquisition of ex vivo lung autofluorescence. While 3% Inspeck™ microspheres are visible at both 20 and 200 fps (see arrows in Fig. 6), the weaker fluorescing 0.3% microspheres are barely visible at both 20 and 200 fps. Despite the fact that faster frame rates result in scaling down the fluorescence captured on the sensor, the autofluorescence and tissue structure are still visible at 200 fps. Faster frame rates require faster display hardware (display rate on monitors is currently limited to 60 to 120 Hz). One interesting observation from the 200 fps video sequence (see accompanying Video 1) is the possibility to extend microendoscopic imaging to tissue elastography.⁵⁰ Also, contrast often increases with increasing frame rate despite lower fluorescence signal. This is due to better localization of the object. Slower frame rates can “smear” the contrast of an object under study due to the fast moving endoscopic environment.

Fig. 6

(a, c) 20 fps and 200 fps imaging with weakly fluorescing micropsheres, 0.3% Inspeck™ and (b, d) more fluorescing microspheres, 3% Inspeck™, in ex vivo lung tissue. Lung autofluorescence can be imaged at 200 fps and possibly more. See the accompanying Video 1 illustrating visibility of lung tissue and red 3% Inspeck™ microspheres (d) at 200 fps. Camera gain and resolution were set to 24 dB and $960\times 600$, respectively (Video 1, MOV, 10.5 MB [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.1]).

3.2.

Microsphere and Tissue Experiments

For both red and green 0.3% Inspeck™ beads, the LOD was determined to be 10⁴ microspheres per ml. Tables 3 and 4 show that LODs for green and red 0.3% Inspeck™ microspheres embedded in the excised human lung tissue are equivalent to ${10}^5/\mathrm{ml}$ and ${10}^4/\mathrm{ml}$, respectively. The controls show no detected microspheres in both cases. The higher LOD for green microspheres is due to high level of lung tissue autofluorescence in the green. Autofluorescence of the human lung elastin and collagen enables anatomical navigation between the airways and alveoli. However, once the distal lung is reached, it would be desirable to remove the background autofluoresence if the object we are trying to image has broadly equivalent excitation wavelengths. In fact, this is the major reason to resort to designing molecular probes in red and NIR in the lung coupled with microendoscopy as the background fluorescence drops dramatically with higher wavelengths.

Table 3

Well plate experiment for green 0.3% Inspeck™ microspheres.

Table 4

Well plate experiment for red 0.3% Inspeck™ microspheres.

3.3.

Labeled Targets, Tissue Experiments, and Observations on Bacterial Blinking

Figure 7 illustrates the capability of the two-color widefield system to perform fluorescence imaging of multiple targets of immediate relevance to pulmonary inflammation and infection. Figure 7(a) shows PKH red labeled monocytes in ex vivo lung tissue. Figure 7(b) shows PKH green labeled neutrophils in ex vivo lung tissue. Figure 7(c) shows the strong signal from Aspergillus fumigatus. Due to its large size, Aspergillus fumigatus is a relatively easy target to image. Simultaneous detection of monocytes, neutrophils, and Aspergillus fumigatus is demonstrated in the accompanying Video 2. As tissue autofluorescence has a strong green component, MI of targets in green is more difficult. However, for cellular targets $15\mu \mathrm{m}$ in size or more, this is entirely feasible providing signal to background is greater than 10%.⁵¹

Fig. 7

Images of multiple targets in ex vivo lung tissue including (a) monocytes, (b) neutrophils, and (c) fungus (Aspergillus fumigatus). Exposure time was set to 80 ms (12 fps), gain 24 dB, 8 bit image, $960\times 600$ camera resolution (Video 2, MOV, 3.12 MB [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.2]).

Smaller objects such as bacteria are also visible at high signal to background; however the fast moving microendoscopy environment requires a large number of dots for them to be immediately observable by a clinician. Figure 8 shows two images of ex vivo lung, one without bacteria, Fig. 8(a), and one with bacteria, Fig. 8(b). Bacteria are visible as blinking fluorescing dots; see Videos 3 and 4. The number of blinking dots is less in Video 3 (no bacteria) when compared to Video 4 (with bacteria). Blinking can be explained as follows: If a fluorescing object is smaller than the fiber core and cladding, then the inherent relative motion of the distal end of the fiber bundle with respect to the tissue will induce the object to move in front of the core (visible) and cladding (not visible). In order to explore the phenomenon of blinking of dynamic targets over the fiber cores, we also evaluated frame rates up to 200 fps in imaging ex vivo human lung tissue with PKH green labeled Pseudomonas aeruginosa. Figure 9 demonstrates this mechanism in three sequential images, each 5 ms exposure time, from Video 5. A core lights up in the left image of Fig. 9, then goes :off: in the middle image, and then lights up the neighboring core in the right image. Figure 10 illustrates the effect further. Constellation™ microspheres were utilized to demonstrate that in free space imaging the blinking effect is absent in both large and small fluorescing objects. However, when imaged through the fiber bundle, blinking dominates the imaging of smaller objects. Objects larger than the core and cladding do not blink. Top and bottom figures in Fig. 10 are not on the same FOV. Videos 6 and 7 demonstrate the effect clearly.

Fig. 8

Images of ex vivo lung tissue (a) without bacteria and (b) with PKH green labeled bacteria. Exposure time was set to 80 ms (12 fps), gain 24 dB, 8 bit image, $960\times 600$ camera resolution. See accompanying Videos 3 and 4 where bacterial detection via higher presence of green dots and their blinking is more obvious (Video 3, MOV, 2.37 MB [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.3]; Video 4, MOV, 1.50 MB [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.4]).

Fig. 9

A sequence of three images demonstrating one mechanism of blinking of bacteria moving between the fiber cores. Exposure time was set to 5 ms (200 fps), gain 24 dB, 8 bit image, $1920\times 1200$ camera resolution (Video 5, MOV, 1.07 MB) [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.5].

Fig. 10

No blinking dots are observed in free space imaging of the Constellation™ microspheres (top) while fiber bundle imaging (bottom) illustrates dots coming in and out of the view as on Fig. 9. Dots which appear in one image, but not in the whole sequence of three images, are circled. Large dots spanning several cores do not blink. Three images from both experiments are consecutive images from a sequence with exposure time 80 ms (12 fps), gain 24 dB, 8 bit image, $1920\times 1200$ camera resolution. Video sequences for both cases are included (Video 6, MOV, 851 KB [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.6]; Video 7, MOV, 871 KB [URL: http://dx.doi.org/10.1117/1.JBO.21.4.046009.7]).

This mechanism can be put to good use providing the bacterial colonies are small enough and bright enough. One can even think of engineering fiber cores and cladding (or grids for ex vivo study of assays) to match the size of the object that needs detecting, thus maximizing blinking. The key attraction is that, to our experience, blinking is readily detectable by observers at SNRs even less than 1. One also has to be cautious; any object smaller than the fiber cladding will eventually blink, whether bacteria or not, so false positives are likely. The same applies to scenarios where ultrafast imaging is performed, which will accentuate blinking due to motion effects and low SNR (blinking due to shot noise). Additionally, many bacterial colonies are large in size ($>20\mu \mathrm{m}$), and these will not blink. Despite limitations, we believe the technique could be useful if appropriately integrated into MI algorithms involving microendoscopy.