2.1.

Opt-ETT Device

A schematic of the upgraded Opt-ETT system is illustrated in Fig. 1. The device consists of an illumination component, a detection component, and a laptop computer loaded with custom software. In the illumination component, a side-firing optical fiber with a $200\mu \mathrm{m}$ core diameter is attached to an uncuffed ETT, and an 810 nm NIR LED (M810F2, Thorlabs, Newton, New Jersey, United States) is used as the light source. The NIR illumination power and power density on the tissue in the current study are significantly lower than those used in the published in vivo studies.^29–34 The side-firing tip of the optical fiber is approximately 1 inch from the distal end of the ETT. In the detection component, the NIR light that diffuses through the tracheal tissue and skin is captured by a detector board attached to the chest skin using a dark medical tape. The board includes five surface-mounted phototransistors P1–P5 (TEMT7000X01, Vishay Semiconductor, Malvern, Pennsylvania, United States) mounted on the front surface (skin side) of the printed circuit board, as shown in Fig. 1(b). Detectors P1, P2, and P3 are positioned with a 10 mm separation along the ETT to monitor the linear displacement of the tube, while Detectors P4 and P5 are placed on each side of P2 with a 3.5 mm separation to detect rotations of the tube. The five phototransistors are connected to a +5V power supply in a common-collector configuration. The emitter resistors (430 kΩ) determine the gain for signal amplification. A resistor-capacitor low-pass filter with an 82 nF capacitor and a 200 kΩ resistor is included in each phototransistor output, resulting in a cutoff frequency of 9.7 Hz. The output voltage signals are recorded by a data acquisition card (USB-6001, National Instruments, Austin, TX) connected to a laptop computer. The data acquisition card has a maximum sampling rate of 20 kHz; nonetheless, the sampling rate was adjusted to 250 Hz, which is sufficient for ETT displacement monitoring. A custom MATLAB (MathWorks, Natick, MA) program has been developed for real-time data acquisition and displacement calculation. During in vivo experiments, the connector of the ETT was attached to a resistive linear position sensor (SP1-4, TE Connectivity, Switzerland) through a thin inelastic wire. The displacement measured by the position sensor served as a reference for the Opt-ETT method.

Fig. 1

(a) Schematic of the experimental setup with an Opt-ETT device. (b) An enlarged front view of the board facing the chest skin. (c) A sketch indicating the relative positions of the sensors to the ETT. The sensors, represented by dots on the top surface of the board in the illustration, are actually located on the opposite side, facing the skin. Sensors P1–P3, placed along the ETT, are used to measure linear (longitudinal) movements of the ETT, and sensors P4 and P5 are placed perpendicular to the ETT to detect ETT rotation.

2.2.

Estimation of ETT Displacement

The displacement of the ETT is estimated using voltages acquired from three sensors, P1, P2, and P3, denoted by $V_{P1}$, $V_{P2}$, and $V_{P3}$, respectively. Two voltage ratios, $r_{12}=V_{P1}/V_{P2}$ and $r_{32}=V_{P3}/V_{P2}$, are calculated as input parameters for an estimation model. Prior to real-time monitoring, calibration of the Opt-ETT is essential to account for the diverse thickness and properties of tracheal tissue across individuals. This calibration process involves measurements performed at five ETT locations, $-10$, $-5$, 0, $+5$, and $+10\mathrm{mm}$, measured by the reference position sensor. The “0” position is established by vertically aligning the tip of the side-firing fiber with the P2 sensor, as confirmed by the attainment of the maximum reading in P2. A positive displacement is induced by pushing the ETT inward, while pulling out the tube results in a negative displacement. Second-order polynomial functions are employed for calibration, which defines the relationship between voltage ratios and displacement. The visualization of the calibration procedure is illustrated in Fig. 2. In the negative displacement range, a polynomial model, $f_{12}^{\text{High}}$ (red solid line), is used to establish the relationship between $r_{12}$ and displacement $d$, based on calibration measurements at $-10$, $-5$, and 0 mm. Similarly, a low signal intensity model $f_{32}^{\text{Low}}$ (blue dotted line) is obtained using $r_{32}$ values measured at the three positions. In the positive displacement range, a high signal intensity model $f_{32}^{\text{High}}$ (blue solid line) and a low signal intensity model $f_{12}^{\text{Low}}$ (red dotted line) are fitted using calibration measurements $r_{32}$ and $r_{12}$ at 0, $+5$, and $+10\mathrm{mm}$, respectively.

Fig. 2

Visualization of the calibration process, which employs second-order polynomial functions, utilizing data captured at three separate displacements for each curve. The calibration curves for $r_{12}$ measurements are indicated by red lines, while the navy blue lines are the calibration curves for $r_{32}$ measurements.

For displacement estimation, the displacement direction (positive or negative) is determined by comparing $r_{12}$ (or $r_{32}$) to its values at $d=0\mathrm{mm}$ $r_{12}(0)$ [or $r_{32}(0)$]. A ratio $r_{12}(d)/r_{12}(0)>1$ indicates an ETT movement towards the negative displacement (outward), whereas $r_{12}(d)/r_{12}(0)<1$ suggests an inward movement or positive direction displacement, or the opposite way if $r_{32}$ is used. As the fitted calibration curves are monotonic within their applicable ranges, displacement estimation is performed by obtaining an estimated displacement $d_{\text{High}}$ from the high signal model and $d_{\text{Low}}$ from the low signal model. The larger one of the two ratios, $r_{12}$ and $r_{32}$, is considered more reliable due to better signal-to-noise ratio (SNR) and is thus weighted more for displacement estimation. The final displacement estimation is a weighted averaging of these two estimations, formulated as $d_{\text{Final}}=w·d_{\text{High}}+(1-w)·d_{\text{Low}}$, where the weight $w$ was empirically set to 0.95. The optimal weight may be further investigated in future research.

2.3.

Monte Carlo Simulation of NIR Light Propagation in Tracheal Tissue

The operation of the Opt-ETT device relies on analyzing changes in the intensity of NIR light emitted from the skin when the tube is displaced. To assess the feasibility of the approach, Monte Carlo simulations have been performed to investigate the light distribution on the chest skin and the relationship between changes in light intensity and ETT displacement. The tissue was modeled as a five-layer structure consisting of mucosal/submucosal/glandular tissue, trachea cartridge, adipose-rich soft tissue, muscle-rich tissue, and skin, arranged from the lumen of trachea to the anterior direction, respectively. Each layer is assumed to be uniform. The optical properties and thicknesses of each layer were determined using a frequency-domain NIR spectroscopy instrument and other relevant studies.^31,35–37 A summary of tissue properties used for the simulation is provided in Table 1. The ValoMC, an open-source MATLAB toolbox for light transport simulation developed by Aleksi Leino et al., was employed to conduct the simulations.³⁸ A two-dimensional (2D) simulation was implemented to study the light intensity distribution along the direction of the sensors P1–P3. A total of 30 million photons were launched into the tissue as a point source following a cosine intensity distribution. Based on the results from the 2D simulation, voltage ratios at various tube displacements (from the 0 position with the fiber tip right below sensor P2) were calculated. The efficacy of the calibration using a second-order polynomial fitting with the five data points was evaluated using the simulated data. Furthermore, three-dimensional (3D) simulations were also performed to investigate the effect of tube rotation. The light source, i.e., the tip of side-firing fiber, was modeled with three launching angles, including 0 deg, $-40\mathrm{deg}$, and $+40\mathrm{deg}$. The light beam contained 50 million photons. The calculated voltage ratios were compared among the three launching angles.

Table 1

Summary of optical properties and thickness of tissue layers used for simulation.

2.4.

Ex Vivo Experiment

Ex vivo experiment was conducted on porcine tissue to assess the accuracy of the Opt-ETT device within a meticulously controlled laboratory environment. The tissue, consisting of both adipose and muscle, was approximately 16 mm thick. To prevent dehydration, the tissue surfaces were covered in clear plastic food wrap, which permits transmission of NIR light. The experimental setup for ex vivo experiment is shown in Fig. 3. A stainless-steel tube was inserted into the ETT, which was then attached to a linear stage, which restricts the movement of the ETT exclusively to the longitudinal axis. In a configuration similar to that illustrated in Fig. 1(a), a reference position sensor was attached to the proximal end of the ETT. The ETT was placed beneath the tissue, with its side-firing tip facing up, while the detector board, with sensors downward-facing, was attached atop the tissue. This setup was designed to simulate the scenario where an ETT is inserted into the trachea. It should be noted that, to prevent cable twisting during the experiment, the detector board was oriented such that sensors P1 and P3 were reversed along the longitudinal axis in contrast to that in Fig. 1(a). This necessitated minor modifications, specifically the interchange of data between these two sensors, of the displacement calibration and estimation procedures. The ETT was repeatedly maneuvered within a range of $\pm 15\mathrm{mm}$ for a total of six cycles using a manual linear stage. The estimated displacement from the Opt-ETT was then compared with that of the reference sensor through scatter plot.

Fig. 3

Photograph of the ex vivo experimental setup.

2.5.

In Vivo Experiments

The feasibility and performance of the device were evaluated through in vivo experiments on eight piglets. The study was approved by the Institutional Animal Care and Use Committee of the Medical College of Wisconsin (MCW). Preparation procedures, including anesthetic injection and ETT intubation, and vital signs monitoring were conducted by MCW Biomedical Resource Center veterinary staff (Vet). Animals were measured on three different days at a mean weight of $7.8\pm 1.3$, $10.2\pm 1.0$, and $11.0\pm 1.4\mathrm{kg}$, respectively. The second experiment (day 14) was performed 2 weeks after the first experiment (day 1), and the third experiment (day 21) was 1 week after the second experiment, resulting in a total of 24 experiments. Prior to each experiment, the ETT tip was placed near the middle of the trachea of the animal determined by the Vet, which was set as the 0 position. The Opt-ETT detector board was taped on the chest with sensors P1–P3 aligned along the tube and P2 right above the tip of the side-firing fiber and at 0 fiber rotation (confirmed by maximum signal for P2). At the beginning of each experiment, calibration was performed using data captured at five tube locations ($-10$, $-5$, 0, $+5$, and $+10\mathrm{mm}$). After that, the ETT was manually displaced within a $\pm 15\mathrm{mm}$ displacement range from the 0 position, with a step size of around 1 mm measured by the reference position sensor, and the voltages of the sensors P1–P3 were recorded by the laptop. The illumination fiber was kept facing the detector board (0 rotation position) during these procedures. Each back-and-forth movement of ETT in the full range from $-15$ to 15 mm is called one cycle. A summary of cycles for the experiments is presented in Table 2, with an asterisk (*) indicating that tube displacement was obtained by directly manually pulling/pushing the tube, while in the remaining experiments displacement was achieved manually but through a linear stage. To evaluate the effects of ETT rotations on the ratios $r_{12}$ and $r_{32}$, measurements were repeated with the tube rotated to two specific angles, $\pm 40\mathrm{deg}$. Rotation of the ETT was facilitated by affixing two sets of water-resistant alphabetical labels onto its connector, illustrated in Fig. 4. The first series of labels (“a,” “b,” “c,” etc.) corresponds to 20-deg increments, while the second series (“I” and “i”) indicates 10-deg intervals. Aligning the ETT with two specific labels from the first series enabled rotational adjustments of approximately $\pm 40\mathrm{deg}$.

Table 2

Summary of movement cycles for linear displacement. The symbol * indicates that the tube was manually pulled/pushed without using the linear stage.

Fig. 4

Photograph of the water-resistant alphabetical labels for rotation control during the in vivo experiments.

During the first eight experiments, the tube was directly pushed or pulled by the Vet or an otolaryngology resident. A picture of the experimental setup is shown in Fig. 5. Although care was taken during tube movement, tube rotations, tilting, and hand tremors might have occurred due to arm tiredness. Therefore, during the remaining 16 experiments, the connector of ETT was fixed to a linear stage for easy and stable adjustment of the tube position. In all experiments, displacements were estimated from the Opt-ETT readings using the calibrated polynomial model and compared to readings of the reference sensor using linear regression, and the degree of agreement between the two methods was investigated using the Bland-Altman method.³⁹

Fig. 5

Photograph of the experimental setup for manual pull/push during the first eight experiments.

3.1.

Monte Carlo Simulation

Simulation results for the propagation of NIR light in tracheal tissue are presented in Fig. 6. The 2D distribution of photon power density when the fiber tip is placed at the 0-horizontal position on the inner surface of the trachea, or $(0,-16)$, is shown in Fig. 6(a). The white dashed lines indicate the boundaries of each tissue layer, and the tissue types are indicated in the color bar on the right. The tissue model is 16 mm thick, and the output power of the light source is normalized to 1 W in simulations for simplicity. The distribution of power density is displayed in colors with the scale based on natural logarithm. The light intensity distribution on the skin is plotted along the horizontal direction (P1–P3) in Fig. 6(b), with the smoothed result obtained using averaging filtering with 6 data points. The intensity decreases to half from the peak value at approximately $\pm 7.5\mathrm{mm}$. The simulated voltage ratios in Fig. 6(c) indicate that $r_{12}$ (dotted blue line) is larger than $r_{32}$ (dotted green line) in the negative displacement (ETT pulled out) range, and the trend is reversed in the positive displacement (ETT pushed in) range. The second-order polynomial model fitted ratio $r_{12}=f_{12}(d)$ (dashed red line) provides a good approximation to the Monte Carlo simulated ratio $r_{12}$ in the displacement range from $-12.5$ to 0 mm. Similarly, the model fitted $r_{32}=f_{32}(d)$, shown as a dashed black line, matches well with the simulated ratio $r_{32}$ in the displacement range from 0 mm to $+12.5\mathrm{mm}$, which confirms the choice of the larger voltage ratio for displacement estimation. This observation has demonstrated the feasibility of applying second-order polynomial function and five-point measurement for calibration. The voltage ratios obtained from the 3D simulations are shown in Fig. 6(d). No apparent difference was observed among the ratios between the three different rotation angles. Therefore, it is reasonable to assume that the fitted models are relatively insensitive to rotations of the ETT within $\pm 40\mathrm{deg}$. If the ratios P4/P2 and P5/P2 go beyond a preset value, a warning signal can be generated so that a correction of excessive rotation can be performed timely for better accuracy in displacement estimation.

Fig. 6

The results of Monte Carlo simulations. (a) 2D distribution of power densities in a five-layer tissue model, with “M,” “C,” “A,” “M,” and “S” representing mucosal/submucosal/glandular tissue, trachea cartridge, adipose-rich soft tissue, muscle-rich soft tissue, and skin, respectively. The red dots indicate the positions of sensors P1–P3 at the zero position. (b) Light intensity distribution on the skin in the horizontal direction (along P1–P3) from the 2D simulation. (c) Simulated voltage ratios $r_{12}$ and $r_{32}$. Sensor P2 is placed at (0, 0) position as the starting point. The dashed lines indicate the calibration function fitted by the calibration measurements denoted by the color dots. (d) Voltage ratios obtained from 3D simulations with the light source being rotated by 40 deg in both directions.

3.2.

Ex Vivo Experiment

The results of the ex vivo experiment are shown in Fig. 7. Voltage readings from sensors P1–P3, plotted against the displacement recorded by the reference position sensor [Fig. 7(a)], exhibit a similar shape to the simulated curves depicted in Fig. 6(b). The observable difference in peak voltages of P1–P3 may be attributed to variations in the thickness and composition of the adipose and muscle tissues within the porcine sample as well as sensitivity of the three sensors. Correspondingly, the voltage ratios, as plotted in Fig. 7(b), show a trend akin to the simulations illustrated in Fig. 6(c), which suggests the feasibility of a second-order polynomial fitting for effective calibration. A total of 134 measurements were obtained. The scatter plot in Fig. 7(c) shows a high correlation ($R^2=0.994$) between the displacements estimated via the Opt-ETT and readings measured by the reference position sensor. This linear regression trajectory closely aligns with the line of perfect agreement. The 95% limits of agreement span from $-2.5\mathrm{mm}$ to 2.8 mm across the full displacement range from $-15\mathrm{mm}$ to $+15\mathrm{mm}$. The measurement data predominantly resides within these bounds, with only a few outliers occurring at displacement ranges exceeding $\pm 11\mathrm{mm}$, which are outside of the calibration range of $\pm 10\mathrm{mm}$. These ex vivo experiment results demonstrate a substantial level of agreement between the measurements obtained from the Opt-ETT and reference position sensor, thereby confirming the efficacy of the Opt-ETT device in a controlled experimental situation.

Fig. 7

Results of the ex vivo experiment. (a) Voltage readings from sensors P1–P5 against displacement by the reference sensor. (b) Calculated voltage ratios $r_{12}$ and $r_{32}$ against displacement. (c) Scatter plot depicting the correlation between the estimated displacements from Opt-ETT and readings from the reference sensor. The shaded areas represent 1.96*SD or a 95% percentile with a normal distribution.

3.3.

In Vivo Experiments

The results from one of the swine experiments (the day 21 experiment for Pig # 4) are presented in Fig. 8. The voltages from sensors P1–P5 are plotted against the ETT displacement measured by the reference sensor in Fig. 8(a). The tube was moved back and forth for five cycles and linear interpolation was applied to the consecutive measurements at 1.0 mm interval in displacement. The solid lines in the figure represent the mean of all measurement points, while the shaded areas indicate 1.96 times the standard deviation (i.e., 1.96*SD). The maximum voltage varies among the sensors, with P1 and P3 exhibiting peak voltages reaching 3 V, while P2 has a value below 2.5 V. This variation in maximum voltage may be attributed to the nonuniformity of the tissue, sensor responses, and the varying contact pressure between the fiber tip and the tracheal wall. The peak values for P4 and P5 are lower because the two detectors were placed 3.5 mm away from path of the fiber tip. The shape of the curves appears wider (or broader) compared to the simulated data [Fig. 6(b)], and the peak voltages may not occur exactly at 0 mm for sensors P2, P4, and P5, or at $\pm 10\mathrm{mm}$ for P1 and P3, again likely due to tissue heterogeneity. The calculated voltages ratios $r_{12}$ and $r_{32}$ are plotted in Fig. 8(b), which follow the similar trend as the simulated data [Fig. 6(c)]. The shaded area is narrow in the displacement range of $-10$ to $+10\mathrm{mm}$, indicating that ratios $r_{12}$ and $r_{32}$ can be used to obtain more accurate estimate of the tube position within this displacement range than using the individual readings. The voltage ratios obtained at ETT rotation angles of $\pm 40\mathrm{deg}$ and 0 deg are plotted in Fig. 8(c). The differences in voltage ratios among the three rotation angles are small between $-10$ and $+10\mathrm{mm}$, suggesting that the model calibrated at 0 deg may be directly used for measuring ETT displacement even at small to moderate angles of rotation.

Fig. 8

Example measurements from a single swine experiment. (a) Readings from sensors P1–P5. (b) Calculated voltage ratios $r_{12}$ and $r_{32}$. (c) Voltage ratios obtained at ETT rotations of 0 deg and $\pm 40\mathrm{deg}$. The shaded areas represent 1.96*SD or a 95% percentile with a normal distribution.

3.4.

Comparison with the Reference Sensor

The comparison of displacements estimated by the Opt-ETT device with readings from the reference position sensor is shown in Fig. 9. Specifically, results obtained from experiments with manual push/pull of the ETT are shown in Figs. 9(a) and 9(b), while results obtained from experiments through the linear stage for positioning are shown in Figs. 9(c) and 9(d). In the first eight experiments with manual push/pull, the estimated displacements are plotted against those obtained with the reference sensor in Fig. 9(a). The linear regression line closely aligns with the perfect agreement line, with a coefficient of determination of 0.959, indicating a high correlation between the two methods. In Fig. 9(b), the result from the Bland–Altman analysis is presented, with the mean and difference of displacements measured by the two methods plotted on the $x$-axis and $y$-axis, respectively. The distribution of measurement points is non-uniform and shows a decreasing trend, suggesting that the measurement variances are not constant. Therefore, a linear regression is applied to estimate the bias, as shown by the red solid line. A positive bias of less than 1.25 mm is observed in the displacement range of $-15$ to $+15\mathrm{mm}$, indicating that the Opt-ETT approach tends to overestimate displacements slightly to the positive direction (ETT moving in).

Fig. 9

Estimated Opt-ETT displacements versus readings from the reference position sensor for all pigs. For experimental results obtained with manual maneuver of the tube, the scatter plot and Bland–Altman analysis are shown in panels (a) and (b). For experimental results obtained with a linear stage for tube positioning, the scatter plot and Bland–Altman analysis are shown in panels (c) and (d). The shaded area represents the 95% confidence interval bounds of estimation.

In the sixteen experiments using the linear stage, data from one experiment was discarded due to obvious errors in the calibration measurements. The scatter plot in Fig. 9(c) shows a similar regression trend as in Fig. 9(a), but with a higher coefficient of determination of 0.986 and a tighter bound for a 95% confidence interval. The Bland–Altman analysis in Fig. 9(d) shows a decreasing bias with the mean displacement, $\sim 1.5\mathrm{mm}$ at $-15\mathrm{mm}$ and $-1.0\mathrm{mm}$ at $+15\mathrm{mm}$. The estimated bias is close to zero near 3 mm mean displacement. The 95% limit of agreement is narrower compared to those in Fig. 9(b) and exhibits a shrinking trend, resulting in less variance and uncertainty in the positive displacement range.

Overall, a high level of association was observed between the displacement obtained by the reference position sensor and the Opt-ETT device, regardless of the ETT maneuvering technique. Nevertheless, less variance and uncertainty were achieved using a linear stage for ETT positioning compared to direct manipulation by hand, which is reasonable as a linear stage is more stable and less prone to unintentional movements. It is important to note that the Bland-Altman analysis has some strict assumptions that may not always be fully fulfilled in practice, and therefore, the results sometimes can be misleading and should be interpreted with caution.⁴⁰ Patrick Taffé has proposed a method to address this issue, but it requires repeated measurements from at least one method, which were not available in this study.⁴⁰