2.1.

System/Design

We based our SWIR endoscope on the design of a regular rigid white light surgical endoscope. A Hawkeye Pro Slim endoscope (Gradient Lens Corporation) was optimized for the wavelength range 900 to 1700 nm using a custom coating. An off the shelf video coupler (Gradient Lens Corporation VC-35) was modified by exchanging the original lens for a SWIR compatible lens. The SWIR camera was attached to the video coupler using a c-mount extension tube. Two different SWIR cameras were used for this project: the Goldeye G-032 Cool TEC2 (Allied Vision) and the Owl 640 Mini VIS-SWIR (Raptor Photonics). Two polarizers were incorporated into the endoscope (see schematic in Fig. 1). One was positioned at the tip of the endoscope (#12-473, Edmund Optics), the other was positioned in the video coupler in front of the lens (LPNIR050, Thorlabs). Using a laser cutter, the polarizer at the tip of the endoscope was custom cut to fit exactly over the fiber ring of the endoscope without covering the lens of the endoscope. The polarizers were oriented orthogonal to each other to minimize direct reflections. The positioning of the polarizers can be seen in Fig. 1.

Fig. 1

Design of the SWIR endoscope. (a) Schematic of the SWIR endoscope, consisting of an InGaAs imager, a video coupler containing a lens and polarizer, a SWIR-optimized rigid endoscope with a second polarizer, and a liquid light guide coupled light source. The blue ellipses and orange rectangles indicate the positions of the lenses and polarizers. The red lines show the path of the illumination beam, whereas the blue line represents the light reflected off the sample and collected by the camera. (b) CAD design of the SWIR endoscope. The orange arrow shows the part of the setup that can be rotated to align the two polarizers. (c) Images of the polarizers used for the endoscope setup: the custom cut polarizer, top view; positioning of the polarizer on the tip of the endoscope; the polarizer covering only the fiber ring and not the lens, held in position by a sleeve (top to bottom and left to right).

The illumination for the endoscope consists of a narrowband 1480-nm laser (Thorlabs L1480G1), which is coupled into the endoscope using a liquid light guide (Thorlabs LLG3-4Z). A lens (Thorlabs C060TMD-C) was used to focus the laser into the liquid light guide and a computer fan with a blade broken off was used to reduce laser speckle.

2.2.

Comparison of White Light SWIR Performance for Visualizing CSF

Transmission spectra were taken of human CSF, artificial CSF (aCSF) (EcoCyte Bioscience), and water. aCSF has an electrolyte concentration closely matching CSF with final ion concentrations of NaCl 125 mM, KCl 3.0 mM, ${\mathrm{CaCl}}_2$ 2.5 mM, ${\mathrm{MgSO}}_4$ 1.3 mM, and ${\mathrm{NaH}}_2{\mathrm{PO}}_4$ 1.25 mM along with high-purity water. 700 microliter volumes of either aCSF or water were transferred to 2-mm pathlength quartz cuvettes (Thorlabs CV10Q7F) to measure attenuation spectra using Lambda 1050+ spectrophotometer (Perkin Elmer). The human CSF spectrum was taken, along with a second water spectrum for comparison, at our clinical site using a Cary 6000i (Agilent).

To investigate the potential of the SWIR absorption band seen in the spectroscopic data for fluid detection in imaging data, we imaged three Eppendorf tubes: one partially filled with water, one partially filled with aCSF, and one empty. They were placed next to each other on a white background and illuminated and imaged from above. The white light image was taken using a white light LED (Thorlabs MNWHL4), an IDS camera (UI-3880SE-C-HQ), and a Navitar lens (Navitar HR F1.4/8 mm). The SWIR image was taken using the Goldeye G-032 Cool TEC2 camera, a Navitar SWIR-35 lens, and a broadband 1450-nm LED (Thorlabs M1450L3).

The SWIR setup used for imaging the water on the chicken skin was identical to that used to image the Eppendorf tubes. For the white light imaging of the chicken skin, an IDS camera (UI-3880SE-C-HQ) was used, in combination with a Navitar lens (Navitar HR F1.4/8 mm) and a white light LED for illumination (Thorlabs MNWHL4).

2.3.

Cadaver Study

To evaluate ease of use and proof of principle for visualization of CSF on the skull base, we performed a cadaver study. A trephine was performed through an infrabrow incision to access the frontal sinus and a communication was achieved with the intracranial compartment by puncturing the posterior table of the frontal sinus. A catheter was then inserted into the defect. An intranasal injury to the skull base was performed allowing communication for the artificial CSF. The aCSF pumped from the frontal sinus could be assessed from the nasal cavity with our endoscope. For imaging, the SWIR endoscope was used, with a 1450-nm LED light source, the Owl 640 Mini, and no polarizers.

2.4.

Use of a Narrowband Illumination Improves CSF Contrast

The initial test to assess the effect narrowing the illumination band has on our imaging data was done using the macrosetup described above. As a sample, we used a Petri dish filled with a small amount of water and placed above a test target. To compare relatively broadband to narrowband illumination, we performed imaging using a 1450-nm LED (Thorlabs M1450L3) to the same LED with an excitation filter (1480 nm center, 12 nm width; Thorlabs FB1480-12). The depth of the water was roughly 1 mm (slightly less in the middle of the Petri dish).

Using the macrosetup and the 1450-nm LED (Thorlabs M1450L3) without a filter versus the same 1450-nm LED with the excitation filter (1480 nm center, 12 nm width; Thorlabs FB1480-12), we also imaged a piece of chicken skin, both without water present and with a droplet of water running across it.

For the assessment in the endoscopic setup, we again used a partially water filled Petri dish above a test target. This time the Petri dish was slightly angled to create a gradient of water thickness. A very small amount of dish soap was added to the water to reduce the surface tension and create a thinner film of water. The maximum thickness of water in the images is around 1 mm, and the thinnest part is $\ll 1\mathrm{mm}$. These images were taken without the use of polarizers. For the broadband illumination, we used a 1450-nm LED (Thorlabs M1450L3), for the narrowband illumination we used a 1480-nm laser (Thorlabs L1480G1). For the macro and the endoscopic images, the Goldeye G-032 was used.

2.5.

Reducing Reflections Through the Use of Polarizers

One polarizer was positioned inside the video coupler (between endoscope and camera), and a second polarizer was attached to the tip of the endoscope. The polarizer at the tip of the endoscope was custom cut using a laser cutter to only cover the fiber ring of the endoscope and not the lens. The final dimensions of the polarizers were 4.3 mm (outer diameter) and 2.5 mm (inner diameter). The polarizers were oriented orthogonal to each other to bring the reflections to a minimum. Images were taken using the endoscopic setup (1480-nm laser, Goldeye G-032 camera) (first without polarizers, then with polarizers added in), and the sample was a piece of chicken skin. Water was dripped onto the chicken skin during imaging.

2.6.

Comparison of Calculated and Measured Values

A piece of nasal turbinate was taken from a porcine nasal cavity and a reflection spectrum was measured using a Lambda 1050+ spectrophotometer (Perkin Elmer). A transmission spectrum of artificial CSF (aCSF) (EcoCyte Bioscience) was also taken using the same instrument. The resulting data were used to calculate the contrast we would expect to see between the porcine nasal tissue and the aCSF if using an illumination source centered on the absorption peak of aCSF.

To be able to compare the calculated value with an actual measurement, we imaged a small pool of aCSF inside a porcine nasal cavity. We used our endoscope setup (1480-nm laser illumination, Goldeye G-032 camera), and the pool of aCSF was about 2 mm deep. The resulting image was used to calculate the actual contrast between the porcine nasal tissue and the aCSF.

2.7.

Imaging the Nasal Cavity

To assess our system in a realistic nasal environment, we imaged a porcine nose using the improved endoscope system. The endoscope system consisted of the narrowband laser light source and the polarizers, as described above (Sec. 2.1), and the Goldeye G-032 camera was used. Our sample was a food grade half pig head. The pig head had been cut in half along the sagittal plane, making it possible to drip water directly into the nasal canal. The endoscope was inserted into the nasal canal and water was dripped into the nasal canal so that it drained toward the endoscope and the tip of the nose.

To compare the SWIR images to white light images, we also imaged the porcine nasal cavity using a white light endoscope (Schindler Endoskope). The camera was the same as previously used for white light imaging (IDS UI-3880SE-C-HQ), and we used a broadband white light source for illumination (Thorlabs OSL2).

3.1.

Comparison of White Light SWIR Performance for Visualizing CSF

To test our hypothesis that the absorption properties of CSF are similar to water and to predict whether a SWIR endoscope could be beneficial for CSF detection, we investigated the optical properties of CSF. The spectroscopic analysis confirmed that the properties of human CSF and aCSF match those of water for the range of 400 to 1800 nm. The suspected absorption band of human CSF and aCSF at around 1450 nm was clearly visible in the spectroscopic data [Fig. 2(a) and 2(b)].

Fig. 2

Optical properties of water, aCSF, and human CSF: comparison of the absorption spectra of (a) water and human CSF and (b) water and aCSF. In both (a) and (b), there are no significant differences in the spectra in the range of 400 to 1800 nm. The differences in OD between human CSF and aCSF are due to a difference in path length for the human CSF spectrum and the aCSF spectrum and the use of different instruments. A water spectrum was taken along with each of the (a) CSF samples to show that the difference in OD is due to different path lengths/instruments and not due to different absorption properties of human CSF and aCSF. (c) Eppendorf tubes from left to right: empty, water, and aCSF. Both water and aCSF show strong absorption at $\sim 1450\mathrm{nm}$, which causes them to appear black, whereas they are translucent in the visible wavelength range. (d) Comparison of tissue without and with the presence in water. Although in the visible spectrum, the presence of water cannot clearly be defined, when using the broadband 1450-nm SWIR imaging, the water produces a much clear contrast to the surrounding tissue and is clearly visible as a dark area. The orange arrows indicate the area where water is present. The background colors correspond to the highlighted wavelength ranges in the spectra.

The imaging of the Eppendorf tubes confirmed that while being transparent in the visible there was strong attenuation when moving the 1450-nm SWIR imaging window, for both water and aCSF; whereas in the visible the fluid in the Eppendorf tubes is translucent, it appears dark and opaque in the SWIR and has a stark contrast to the air-filled Eppendorf tube and the background [Fig. 2(c)]. Having confirmed that the optical properties of water, human CSF, and aCSF in the SWIR are near identical, we will use water for all experiments from this point on, due to water being readily available.

Since skin and other biological tissues have a high water content, we were expecting a reduction in contrast between water and background when moving to imaging the water on top of tissue. We tested whether moving to the SWIR would give us an advantage in the detection of water on top of tissue compared to white light imaging using a piece of chicken skin (food grade; inner side of skin facing upward/toward camera). Using a piece of plastic tubing inserted through the chicken skin, water could be pooled onto the surface of the chicken skin. The resulting images showed that in the visible there is no significant difference to be seen between chicken skin with and without water. In the SWIR, however, the water droplet on top of the chicken skin can clearly be depicted against the surrounding skin [Fig. 2(d)].

3.2.

Cadaver Study

The cadaver imaging confirmed that we can detect an accumulation of artificial CSF inside the human nasal cavity using the SWIR endoscope. The SWIR endoscope system is also able to show anatomical landmarks comparable to scopes currently used in sinus surgery. Detecting thin films of artificial CSF proved difficult, highlighting the need to make modifications to the system to be able to detect smaller amounts of liquid (Fig. 3).

Fig. 3

Imaging the skull base of a cadaver using the SWIR endoscope. (a) The endoscopic SWIR images show that larger accumulations of aCSF can be detected due to the increased amount of absorption. The yellow outline shows the cavity in which aCSF accumulated, the left and right SWIR images are otherwise identical. (b) Detection of thin films of aCSF is difficult as there is not enough contrast between the thin films and the surrounding and underlying tissue. In the right image, we have highlighted the area where water is present. For comparison, the left image is identical to the right, other than the water not being outlined. This shows that accurate detection of thin films of aCSF was not possible with the endoscope design used for the cadaver study, which incorporated a 1450-nm LED without an excitation filter.

3.3.

Use of a Narrowband SWIR Illumination Improves CSF Contrast

Previous data had shown that droplets of water and thick films of water gave good contrast to surrounding tissue; thin films on the other hand were much harder to depict as they showed too little absorption to contrast clearly against the background. We aimed to improve this by moving the illumination window to a narrower range and focusing it more on the attenuation peak we had seen in the spectroscopic data [Fig. 4(a)]. The resulting images showed that narrowing the spectrum had a strong effect on the transparency of the water. Although it was clearly possible to depict the test target under the thin layer of water when using the broadband LED, this was no longer possible when adding the bandpass filter [Fig. 4(b)]. This increase in opacity when moving to the narrower illumination wavelength range meant that a trickle of water on top of tissue (chicken skin) could be seen much clearer than when using the 1450-nm wideband illumination (Fig. S1 in the Supplementary Material).

Fig. 4

The effect of a narrower wavelength range on the absorption (and thus detection) of water. (a) The colored bands in the spectra show the wavelength range for the illumination of the images. (b) When narrowing the wavelength range the absorption of the water clearly increases, making the water appear darker and less translucent. This becomes apparent when viewing the test target with the different wavelength ranges, and the effect is present both in a macroimaging setup as well as in the endoscopic imaging setup. The yellow line on the endoscopic images shows the edge of the water.

We adapted these findings to our endoscopic setup and exchanged the previously used broadband LED (M1450L4, Thorlabs) for a 1480-nm narrowband laser (L1480G1, Thorlabs). The laser allows for improved optical efficiency and provides additional power at the narrowband of 1480 nm, which we would lose with a broadband LED and filter setup. We analyzed the effect that had on the endoscopic image in a similar experiment to the one described above for the macrosetup. We again used a partially water-filled Petri dish laid on top of a test target. To be able to assess the effect the depth of the water had on its transparency, we slanted the Petri dish to create a gradient of water thickness. We added a very small amount of dish soap to the water to reduce the surface tension of the water and thus create a thinner film of water. The resulting images showed that the laser gave comparable images to those where the LED spectrum had been narrowed using a bandpass filter, and that the laser spectrum was sufficiently narrow for our purposes without requiring an additional bandpass filter. Although the test target was clearly visible throughout the entire image with the broadband illumination, it was only clearly visible for very thin films ($\ll 1\mathrm{mm}$) when using the narrowband illumination. With an increasing thickness of the film of water, it was no longer possible to perceive the test target [Fig. 4(b)].

As a final test, we then imaged water on top of tissue using the endoscopic setup with the narrowband laser illumination. We used the same endoscope setup as had been previously used to image the Petri dish with water, only swapping the Petri dish out for a piece of chicken skin. We imaged the chicken skin without water and while running a droplet of water across it. The resulting images showed that narrowing the illumination helped to depict where water was present on the sample (Fig. 5). This result reflected the result we had seen when comparing the 1450-nm LED wideband illumination to the 1450 nm + 1480-12 bandpass filter narrowband illumination in the macrosetup. Narrowing the illumination spectrum around the absorption peak of water increases the contrast between tissue and water and thus the chances of being able to pick up smaller amounts of water against the surrounding tissue.

Fig. 5

The effect of a narrower wavelength range when imaging tissue. (a) The narrowing of the wavelength range increases the overall contrast of tissue structures. It also makes it easier to pick out water on top of tissue, and it has an especially large effect on the edges and thinner areas of water. These images were taken without the use of polarizers. (b) The plot of the blue line profiles seen on the images above. The line profile cuts the edge of the water area. In line profiles over the water, we see that the narrowband image has higher contrast than the broadband image.

3.4.

Reducing Reflections Through the Use of Polarizers

In the previous data, we observe that some areas of the images are overexposed, due to the presence of direct reflections of the illumination source of the sample. We aimed to reduce these bright reflections by adding polarizers to our endoscope design. One polarizer was positioned inside the video coupler (between endoscope and camera) and a second polarizer was attached to the tip of the endoscope. The polarizer at the tip of the endoscope needed to be custom cut to only cover the fiber ring of the endoscope and not the lens. The polarizers were oriented in such a way to each other as to bring the reflections to a minimum. Using this modified endoscope setup, we again imaged a piece of chicken skin, first without water and then while running water across it (Fig. S2 in the Supplementary Material). Comparing the images we obtained with and without polarizers, we can show that adding polarizers to the design gives us two positive effects: First, reducing the bright reflections, we are able to analyze more of the image accurately. Before, the regions displaying bright reflections could not be analyzed, as it was impossible to tell whether the tissue was causing the reflections, or whether the surface of the water was causing the reflections. Adding the polarizers reduced the size of overexposed areas on the images, thereby increasing the amount of information that could be obtained from one image. Second, reducing the bright reflections allowed us to increase the overall illumination; this in turn enhanced the contrast between tissue and water.

3.5.

Comparison of Calculated and Measured Values

To check how well our improved SWIR endoscope performed we estimated the contrast, we would expect to see between porcine nasal tissue and aCSF. We compared a reflection spectrum of porcine nasal tissue to the transmission spectrum of aCSF. The absorption value for aCSF at 1480 nm was at around 2.5 OD for a path length of 2 mm. At this wavelength, the porcine nasal tissue had an OD of around 1 [Fig. 6(a)]. We would, therefore, expect to see $2.5\times$ less signal from the aCSF compared to the tissue if we are looking at an aCSF depth of 1 mm.

Fig. 6

(a) Absorption spectrum of aCSF (path length of 2 mm) versus reflectance spectrum of a piece of porcine nasal mucosa (including underlying structures). From these data, we would expect to get roughly two and a half times as much signal from the nasal mucosa compared to a 1-mm film of aCSF. (b) Image of pig nasal mucosa with a pool of ca. 1-mm deep aCSF.

We imaged a thin pool of aCSF on top of porcine nasal tissue using our endoscopic SWIR device. Region of interest (ROI) 1 shows porcine nasal tissue, and ROI 2 shows aCSF of roughly 1 mm of depth [Fig. 6(b)]. The ROIs have the following counts (see Table 1).

Table 1

Contrast between porcine nasal mucosa and aCSF.

Our measured ratio lies very close to the calculated value, showing that the modifications we made to the endoscope have led to a very efficient imaging of fluid.

3.6.

Imaging the Nasal Cavity

For the final assessment of the improved SWIR endoscope design, we used a pig head that was halved along the saggital plane (food grade) and imaged the nasal cavity. The endoscope setup was identical to the setup described in Sec. 2.1. This setup combines the use of a narrowband illumination with the polarizers. Water was dripped into the nasal cavity near the skull base and trickled toward the tip of the nose. The nasal cavity was imaged without water present and while the water was trickling along it (Fig. 7). The same experiment was repeated using the white light endoscope.

Fig. 7

Images of the nasal cavity showing: no water, droplet appearing, droplet running across area of view, a trickle of water following in the path that the droplet left, and residue of water. Images from (a) and (b) the SWIR endoscope; (c) and (d) a white light endoscope. (b), (d) The water highlighted and otherwise identical to the images above them. With the revised endoscope design (including the polarizers and the narrow wavelength illumination), we significantly improved upon the quality of the data compared to our initial cadaver imaging. There are now fewer distracting reflections present, enabling us to clearly analyze almost the entire image and needing to omit very little areas in the image from analysis. Not only large droplets but also thin films and even small residues of water can be picked up against the tissue. In contrast, in the visible it is extremely difficult, if not almost impossible, to accurately say whether water is present or not. When viewing a video rather than still images, the differences between the imaging channels become even clearer (Videos 1 and 2). An issue in the SWIR imaging that needs to be addressed in further work is differentiating between shadows and water. In all images, it is not possible to accurately diagnose water in the areas that are not well illuminated. (Video 1, mp4, 10.8 MB [URL: https://doi.org/10.1117/1.JBO.28.9.094803.s1]; Video 2, mp4, 8 MB https://doi.org/10.1117/1.JBO.28.9.094803.s2).

Using the pig head, we were able to show that we were able to detect both droplets of water and thin films of water inside a nasal cavity. Even water residue left behind on the tissue could be detected, showing that we were able to greatly improve upon our initial endoscope design. In contrast, when using the white light endoscope, it was much harder to say whether water was present or not. The difference between the SWIR images and the white light images become even more evident when viewed as video (see Videos 1 and 2).