2.1.

Ex Vivo Testing

Tests exploring the spectral properties of tracheal and esophageal tissue were first conducted at the Photonics Research Center at the United States Military Academy in West Point, New York, using excised porcine tissue samples ordered from Animal Biotech Industries (Danboro, Pennsylvania). Two intact larynx, trachea, and esophagus samples were harvested and shipped overnight in a styrofoam cooler stored over ice. All tests were conducted upon receipt within 30 h of harvesting. Both the trachea and esophagus were dissected from the larynx, separated from each other, and sliced longitudinally to expose the inner tissue lining (lumen). Each tissue sample was then dissected into three horizontal sections to represent the upper, mid, and lower portions of the trachea and esophagus, respectively. Additional slices of the epiglottis and vocal folds were also taken from larynx, the region just above both the separation of the trachea and esophagus. All slices were individually mounted onto cork such that they lay flat for imaging with the hyperspectral camera.

2.2.

Hyperspectral Camera Instrumentation

Images of the tissue were taken using an OKSI Hyperscan VNIR imager and HyperVision software (Opto-Knowledge Systems, Inc., Torrance, California). The camera consisted of a $696\times 1314\text{pixel}$ array and was paired with a 70-mm Apochromat lens yielding an approximate $80\text{-}\mu \mathrm{m}$ spatial resolution. Spectral data were collected in a sweep mode with the resulting data cube yielded one column of spatial data with the full 520 wavelength elements for each CCD exposure, resulting in a total image spectrum from 396 to 1050 nm at each pixel with a 1.25-nm spectral resolution. All samples were illuminated with four white lights (4700 K, D50) and one UV LED (396 nm, Edmund Optics, Barrington, New Jersey) within a fully enclosed blackout box to isolate the camera and sample from external light. The lights were manually positioned to provide uniform illumination with the UV LED positioned to provide maxima illumination at the center of the target.

To obtain a background noise offset, a dark reference spectrum was collected with all the lights off and the camera shutter closed. A white reference spectrum was also collected by imaging a 99% reflectivity standard (5# White Balance Target, #58-610, Edmund Optics, Barrington, New Jersey) exposed to all five lights. The integration time was set to avoid saturation of the white reference. The data collected from each image were then normalized, accounting for both the white and dark spectra of the camera, using the following equation:

2.3.

Ex Vivo Data Analysis

After imaging, the hyperspectral data were processed with the Hypervision software to generate spectral curves for multiple points on each section of the trachea and esophagus along with corresponding color images of the complete imaged section. Each sampled point consisted of a $10\times 10\text{pixel}$ square, with each pixel comprising an approximate $81\times 81\mu {\mathrm{m}}^2$ of the imaged section. Choosing to analyze the spectra within a $10\times 10\text{pixel}$ square helped average out any noise and also resulted in each spectral curve being the average spectral reflectance over $\sim 0.65\text{-}{\mathrm{mm}}^2$ area on the imaged section. From a practical sense, with our future investigations targeting fiber optic and discrete sensors, we felt that analyzing the reflectance spectra with this perspective would more accurately reflect the sensing area of a probe in close proximity of the tissue than by analyzing pixels individually.

2.4.

In Situ Testing

In situ tests were conducted at the United States Army Institute of Surgical Research in Fort Sam Houston, Texas. Pig models were obtained from a previously approved Institutional Animal Care and Use Committee protocol within the Institute involving a procedure that ended in humane euthanasia and did not compromise the properties of the upper respiratory system. In order to remain as close as practical to in vivo representation, spectral data were collected from each pig within 2 min of the euthanasia before any postmortem changes or degradation could affect the local tracheal and esophageal tissue properties.

For in situ testing, reflectance spectra were captured using a fiber optic reflection probe connected to a compact spectrometer and a halogen white light source, as shown in Figs. 1 and 2. The fiber optic probe was a $200\text{-}\mu \mathrm{m}$ core, 2-m long bifurcated reflection probe (R200-7-UV-VIS, Ocean Optics, Dunedin, Florida) with a central light emitting fiber with six collection fibers surrounding it to capture the light reflected from the luminal tissue, as diagrammed in Fig. 1 and displayed in Fig. 2.

Fig. 1

System block diagram of experimental setup with compact spectrometer and halogen white light source connected to the bifurcated reflection probe with a cross-sectional arrangement of six collecting fibers around one emitting fiber.

Fig. 2

Fiber optic instrumentation setup, as systematically diagrammed in Fig. 1.

The compact spectrometer was sensitized for 420 to 1070 nm with a $2048\times 14\text{pixel}$ Hamamatsu S9840 back-thinned CCD (SPM-002-DT, Photon-Control, Vancouver, British Columbia, Canada) and was connected through USB to a laptop with SpecSoft software to display the spectra in real time and capture them for postanalysis. The exposure time was set to 5 ms in order to prevent saturating the detector from the output power of 8.8 mW from the halogen lamp white light source (HL-2000-HP, Ocean Optics). To help mitigate some of the noise, five subsequent exposures were averaged to amount to 25 ms of total exposure time. To account for background interference and the imperfections of the white light source, dark and white reference spectra were collected and stored. The dark spectrum was stored with all ambient lights and the white light source off while the probe was covered with black felt. The white light source was then turned on and the distal end of the probe placed in a white integration sphere (FIOS-1, Ocean Optics) to capture the reference white spectrum. The same reference spectra were loaded before each data collection.

Figure 3 shows the instrumentation of the ET tube in the trachea and the relative placement of the fiber optic probe into the trachea [Fig. 3(a)] and the esophagus [Fig. 3(b)]. The ET tube remained in the trachea from the procedure, and the fiber optic probe was passed through the tube into the tracheal region. The tube was then retracted to expose the distal end to the luminal tissue and the reflectance spectrum was displayed in real-time using SpecSoft and was captured. The probe was repositioned to capture the reflectance profile from six locations within the trachea. After saving the tracheal reflectance spectra, the probe was removed from the ET tube and cleaned with an isopropanol surface disinfectant wipe (CaviWipe™, Metrex Research, LLC, Orange, California). The probe was then inserted into the esophagus and the reflectance spectrum was displayed, captured, and stored at six different locations. The ET tube was left in place to ensure that the probe did not re-enter the trachea.

Fig. 3

(a) Fiber optic probe instrumentation when inserted into the tracheal lumen through the endotracheal tube. (b) Fiber optic probe instrumentation for sensing the esophageal lumen by remaining outside and posterior to the ET tube.

2.5.

In Situ Data Analysis

The reflectance, absorbance, and amplitude spectra collected through the SpecSoft Software were stored as text files and imported into MATLAB for spectral analysis. Each pig was numerically identified as 1 through 8, and the respective tracheal and esophageal spectra were organized accordingly. The tracheal and esophageal spectra were plotted separately and jointly for each pig to isolate as well as compare the unique features. To compare spectra among the pigs, the reflectance spectra for each pig’s trachea and esophagus were averaged by summing the value at each data point across the captured profiles and dividing by the number of profiles in order to arrive at one reflectance spectrum representing the trachea and one reflectance spectrum representing the esophagus for each pig. The averaged reflectance spectra were then compared among the pigs by plotting them simultaneously.

To quantitatively characterize and compare the unique features of the spectra, the data were first smoothed by applying a second-order, 15-point framelength Savitsky–Golay filter. Local minima and maxima were then identified in each spectrum, with the mean and standard deviation of the wavelength occurrence of each maximum and minimum determined for each pig. Furthermore, the relative amplitudes, wavelength distances, and slopes were calculated between each maximum and minimum for each spectrum and averaged for one mean per pig. These quantitative characteristics of the minima and maxima were then compared within tissue types and among all pigs for relative similarities as well as between the two tissue types for comparative differences. We chose to assess the relative minima and maxima for the purpose of identifying unique characteristics that could be detected and recognized using discrete components, such as filters, that would be realistic for a prehospital implementation.

3.1.

Ex Vivo Results

Over 200 distinct points were sampled across the two tracheal and esophageal samples. Qualitative examination of the spectral curves from the samples showed no difference among the different sections (upper, middle, and lower) within each respective tissue type; however, the spectral curves of the tracheal tissue compared to the esophageal did show a significant difference in the profiles, as seen in Fig. 4. Figure 4 shows normalized representative samples of the spectral curves found for both the esophagus [Fig. 4(a)] and the trachea [Fig. 4(b)]. The key spectral difference is a distinct peak followed by two local minima seen in the trachea between 530 and 580 nm wherein lies a peak at $\sim 560\mathrm{nm}$ and the minimum lies near 540 and 577 nm. With the peak qualitatively observed, the data were then normalized to the reflected intensity at the peak (560.9 nm) to both assist in evaluating the peaks qualitatively and determine the strength of the local peak to the relative minima. The insets in Figs. 4(a) and 4(b) show the spectral curve around the peak (535 to 585 nm), clearly showing the distinctive peak seen in the tracheal tissue that is not present in the esophageal tissue.

Fig. 4

(a) Reflectance spectra for multiple samples of the esophageal tissue (inset). Reflectance spectra over the area of interest (535 to 585 nm) for the esophageal tissue (image). Colored image taken from the hyperspectral camera with the sample points highlighted by the circles. (b) Reflectance spectra for multiple samples of the tracheal tissue (inset). Reflectance spectra over the area of interest (535 to 585 nm) for the tracheal tissue (image). Colored image taken from the hyperspectral camera with the sample points highlighted by the circles.

In order to further characterize the results, the local maxima and minima were compared among upper, middle, and lower sections, between the tissue types and between the two pigs. For the first pig, the location of the maxima and minima was compared between the top and center sections to determine if the location in the trachea had any difference in the spectral characteristics. An independent $t$-test was conducted on both samples for the two local minima and the local peak. For the top section, the peak was located at 560.9 nm for both sections, resulting in a $p$-value of 1, showing no statistical difference. The $p$-values for the two minima of 0.65 and 0.32 also showed no significant difference. Given the lack of significant difference and no known anatomical or physiological differences that would cause the sections to have different spectral responses, the spectral responses from the different points of each section were combined for the trachea and esophagus, respectively. For the trachea, the averaging of the spectral curves across the points and sections resulted in a local maximum of $560.9\pm 1.109\mathrm{nm}$ and local minima locations of $543.242\pm 0.828\mathrm{nm}$ and $577.917\pm 0.506$ for pig one, and $560.377\pm 1.217\mathrm{nm}$ and local minima locations of $542.398\pm 0.746\mathrm{nm}$ and $577.082\pm 0.579$ for pig two. Analyzing across pigs showed no significant difference in the location of the peaks; but, there was a significant difference in the locations of the minima, with $p$ values of 0.0048 and $<0.0001$, respectively. Overall, the spectral shape of the “tracheal peak” was the primary characteristic that allowed us to distinguish between the trachea and esophagus subjectively and was enough of a basis to continue research in situ.

In addition to just examining the spectral data, we also examined single-band images (each band being 1.25 nm in spectral width) focusing on local minima and maximum seen in the spectral data plot. In order to emphasize the contrast between bands, the same common background darkness was subtracted from all images and the resulting differences scaled to 256 level grayscale. In Figs. 5(a)–5(c), it is seen that the tracheal tissue at the 560.9-nm band in Fig. 5(b) is brighter than the same tissue in Figs. 5(a) and 5(c) at the 541.6- and 577.7-nm bands, as would be expected based on the spectral data. It can also be seen that Fig. 5(e), corresponding to the 560.9-nm band in the esophageal tissue, is less intense than Figs. 5(d) and 5(f), corresponding to the same tissue at the 541.6-nm band and 577.7-nm band, respectively.

Fig. 5

(a–c) Single-band images of tracheal tissue for wavelength bands centered on (a) 541.6 nm, (b) 560.9 nm, and (c) 577.7 nm, corresponding to the local maxima and local minima seen in Fig. 4(b). (d–f) Single-band images for the esophageal tissue for wavelength bands centered on (d) 541.6 nm, (e) 560.9 nm, and (f) 577.7 nm. The same common background signal was subtracted from each image to highlight the contrast in intensities.

3.2.

In Situ Results

Spectral data were collected on eight pigs, with three to nine spectral captures in both the trachea and esophagus for each pig. In order to collect the data as close to the time of death as possible, the same calibration spectrum was loaded in SpecSoft before each test.

Figures 6(a) and 6(b) show the averaged esophageal and tracheal spectra for all the pigs, with each spectrum representing the average spectral profile for each pig. Initial qualitative analysis confirmed the existence of the same features as identified by the previous ex vivo investigation, with two minima and one maximum within the region of 530 to 580 nm. Conversely, the esophageal spectrum exhibited a gradual negative slope over the same region of interest. The data were then normalized similar to the ex vivo results, where the maximum reflectance at the peak occurring around 560 nm was normalized to 1, and then was smoothed using a second-order, 15-point framelength Savitzky–Golay filter.

Fig. 6

(a) Normalized esophageal reflectance spectra averaged across all captures for each pig to yield one spectrum per pig. (b) Normalized tracheal reflectance spectra averaged across all captures for each pig to yield one spectrum per pig.

To further evaluate the relative uniqueness between the spectra, we combined all the individual captures into one joint dataset, including both esophageal ($n=54$) and tracheal ($n=57$) spectral data. With each spectrum representing an observation and each wavelength data point representing a variable ($n=1529$, 450 to 950 nm), we computed the principal components (PCs) of the collective tracheal and esophageal dataset. Figure 7 shows the first two PCs extracted from the combined esophageal and tracheal data plotted alongside the mean spectrum for each of the trachea and esophagus. The individual means of the tracheal and esophageal spectra were calculated by averaging all the tracheal spectra and all of the esophageal spectra (Fig. 6) separately. Since the normalized data from the averaged tissue spectra were normalized, such that the intensity at 560 nm was set to 1, the data were renormalized to set 560 nm to 0 such that it matched up with the PCs.

Fig. 7

PC1 and PC2 from the combined tracheal and esophageal data set plotted versus the averaged spectrum from all the tracheal captures (trachea) and all the esophageal captures (esophagus).

The first principal component (PC1) and the esophageal mean spectrum resulted in a positive correlation coefficient of 0.9145 ($p<0.0001$), whereas the second principal component (PC2) and the tracheal mean spectrum also showed a positive correlation coefficient of 0.6724 ($p<0.0001$). Conversely, when the esophageal mean was compared with PC2 and the tracheal mean compared with PC1, both yielded significant negative correlation coefficients of $-0.5143$ and $-0.5902$, respectively.

As highlighted from observing both the normalized spectra and the PC comparisons, the 530- to 580-nm region still exhibited the most unique characteristic that was not present in the esophageal spectra. To focus in on this region of interest, we identified the maxima and minima, and their wavelength occurrences within the 530 to 580 nm region for both esophageal and tracheal tissue. The wavelength occurrences and amplitudes were subsequently plotted using open circles on top of the spectra plots, as shown in Fig. 8, to superimpose the detected maxima and minima on the spectra. Figures 8(a) and 8(b) show examples of the smoothed spectral data zoomed in on the region of interest with the minima and maxima circled for both the esophagus and trachea, respectively.

Fig. 8

(a) Smoothed esophageal spectra from pigs 1 and 2. (b) Smoothed tracheal spectra from pigs 1 and 2 with detected minima highlighted by the open boxes and the local maxima circled.

For the purpose of clear viewing of the comparative points and individual spectra, the data shown in Fig. 8 were not normalized. As seen from the superimposed results of the maxima and minima detection (Fig. 8), the three points identified on each spectrum represent the detected minima or maxima within the 530- to 580-nm region. Table 1 shows the mean wavelength occurrences and standard deviations for each local minima or maxima for the respective tissue types, termed point 1, point 2, and point 3 corresponding to the first, second, and third points detected within the region. Both the numerical values and the previous spectral graphs highlight that the tracheal tissue exhibited the consistent pattern of two local minima, points 1 and 3, with one local maximum, point 2. On the other hand, the esophageal tissue most often exhibited a pattern of two local maxima, points 1 and 3, with one minimum, point 2. Due to the nature of the gradually decreasing reflectance over the 530- to 580-nm region, 13% of the individual esophageal spectra did not yield a third significant point as the interval only showed one local maximum at the beginning, near 530 nm, and one local minimum at the end of the interval, near 580 nm.

Table 1

Mean wavelength occurrence of minima and maxima of tracheal and esophageal data. For tracheal tissue: point 1 and point 3 correspond to minima, whereas point 2 represents the maximum. For esophageal tissue: point 1 and 3 represent maxima, whereas point 2 represents the minimum.

Using the identified local minima and maxima, we further characterized the respective patterns by quantifying the amplitude, distance, and slope between each point for each individual capture for each tissue. The calculated values were then averaged for each pig and averaged across all pigs to calculate the overall mean amplitude, mean distance, and mean slope, as reported in Table 2. Negative values indicate a decrease in relative reflection.

Table 2

Amplitudes, distances, and slopes between minima and maxima for each tissue type. For tracheal tissue: point 1 and point 3 correspond to minima, whereas point 2 represents the maximum. For esophageal tissue: point 1 and 3 represent maxima, whereas point 2 represents the minimum.

An independent $t$-test was also performed on the group of mean amplitudes, distances, and slope values for each pig ($n=8$) to compare between the two tissue types. For the purpose of comparison, absolute values were used in the calculation. As shown in Table 2, there was a significant difference in the amplitudes and distances for the mean amplitude and distances between both pairs of points. On the contrary, the absolute values of the average slopes did not differ significantly for either pairs of points.