1.

Introduction

Obstructive sleep apnea (OSA) affects 9% to 38% of the adult population,¹ and the corresponding economic impact can be up to $150 billion annually.² If inadequately diagnosed or treated, OSA can lead to severe health complications such as cardiovascular events, neurocognitive effects, and depression.³ The gold standard diagnostic tool for OSA is polysomnograms, which measure important parameters such as the apnea–hypopnea index, percutaneous oxygen content, rapid eye movement sleep, and limb movements.⁴ However, polysomnograms cannot provide information regarding the structure and anatomical collapse of the airway. The upper airway is a complex anatomical structure that undergoes dynamic geometric changes over the respiratory cycle. These structural changes can cause airway compromise at many locations, including the palate, base of the tongue, pharyngeal wall, epiglottis, and larynx. The accurate identification of the sites of airway obstruction during sleep is essential for determining the ideal medical or surgical management. Drug-induced sleep endoscopy (DISE)⁵ is an emerging method to identify the obstruction sites during sleep and is now widely implemented in surgical practice. However, the interpretation of DISE can be subjective with interobserver disagreement in the location of airway obstruction.⁶ Despite best efforts utilizing multiple diagnostic approaches, a large number of sleep patients (55% to 57%)⁷ still fail to adequately respond to surgical interventions. As such, the fields of sleep medicine and sleep surgery are in need of better diagnostic tools for the evaluation of upper airway collapse during sleep.

Optical coherence tomography (OCT) is a well-established mesoscopic imaging modality that can obtain non-invasive, high-speed, and high-resolution images of biological systems.⁸ OCT can couple into a sectional miniature imaging catheter and be used as a range-finding modality to determine the size and shape of upper airway anatomical structures.^9–15 In a 2012 study by Jing et al.,¹⁶ a long-range OCT system was used to rapidly scan the upper airways of three awake human subjects. This system generated a volumetric dataset that reproduced the airway lumen with an accuracy within the order of tens of microns. The present study aims to utilize a similar OCT setup to acquire high-resolution three-dimensional (3D) reconstructions of the upper airway lumen in an apneic patient during sleep. By accounting for the curvature of the dynamic airway in the volumetric reconstruction, we theorize that the respiratory airflow modeling will provide a rigorous method for identifying sites of obstruction in OSA.

2.

Methods

2.1.

Long-Range OCT System

As depicted in Fig. 1, the OCT system utilized in this study incorporates an acousto-optic modulator (Brimrose Corp. of America, Sparks Glencoe, Massachusetts, United States) to generate a frequency shift of 150 MHz in the interferometer signal. A full imaging range of 12.8 mm at 6 dB was achieved by removing signal processing artifacts.¹⁷ The swept-source laser (Axsun Technologies, Billerica, Massachusetts, United States) had a center wavelength of 1310 nm with a bandwidth of 109 nm, allowing for an axial resolution of $10\mu \mathrm{m}$ in tissue (assuming a refractive index of 1.4). The fiberoptic-based imaging catheter consisted of a gradient-index rod lens for focusing and a 45-deg mirror at its distal tip to achieve an endoscopic 90-deg-angle view (Fig. 2). The optical assembly at the tip was protected within a stainless-steel housing with a lateral opening for imaging beam propagation. The metal housing had an outer diameter of 1.5 mm and was firmly soldered to a stainless-steel torque coil, which contained the single-mode optical fiber that connected the light source to the imaging optics. The torque coil had an outer diameter of 0.97 mm. Driven by a brushless direct current (DC) motor at the proximal end, the entire catheter was rotated through the torque coil during imaging acquisition. Each complete rotation of the catheter provided a two-dimensional cross-sectional image, or a B-scan. Volumetric imaging was achieved by pulling back the catheter using a motorized translation stage during rotation to acquire a series of B-scans with equal spacing. The DC motor and the translational stage together comprised the scanning apparatus. Imaging was performed at a rotational rate of $25\text{frames}/\mathrm{s}$ with 2000 axial scans per image and a pullback rate of $12.5\mathrm{mm}/\mathrm{s}$.

Fig. 1

Schematic of the OCT system for upper airway imaging.

Fig. 2

Upper airway OCT catheter. (a) The schematic of the imaging procedure for upper airway imaging. The OCT catheter is housed in a transparent FEP sheath that is inserted through the nasal cavity and positioned in the hypopharynx. The 90-deg imaging catheter rotates and retracts within the FEP sheath during data acquisition. (b) Photograph of the OCT catheter. (c) Close-up photograph of the catheter tip.

2.3.

Image Processing and 3D Reconstruction

Image sets were assessed for overall quality prior to 3D reconstruction. Image quality was evaluated by considering signal intensity, distortion caused by non-uniform rotation of the imaging catheter, presence of artifacts/noise, identifiable anatomical structures, and limitations of imaging range. A detailed description of the data selection process has been previously reported in a 2015 study by Lazarow et al.¹⁸ Only image sets that demonstrated high signal-to-noise ratio, minimal to no distortion, and identifiable anatomical structures were rendered into 3D models.

Raw images were processed and analyzed using IrfanView (IrfanView, Irfan Skiljan, Austria), MATLAB (MathWorks, Natick, Massachusetts, United States), and Amira (FEI Visualization Science Group, Burlington, Massachusetts, United States). The technique used to create 3D reconstructions of the upper airway from OCT images has been detailed in a 2014 study by Nguyen et al.¹⁹ Dispersion correction was not further performed in this system. The images were post-processed for correct anatomical scaling, sequential image alignment, image downscaling, and converting from Cartesian to polar coordinates. The processed images were imported into Mimics (Materialise, Leuven, Belgium) for manual segmentation to define the contour of the airway lumen at each B-scan. The B-scans of the contour were assembled into a “cylindrical” representation of the airway. To recreate the anatomical curvature of the native airway, the spatial coordinates of each B-scan were adjusted based on the curvature obtained using the 3D electromagnetic tracker using a custom morphing algorithm in Visual Basic (Microsoft, Redmond, Washington, United States). If important anatomical structures (e.g., tracheal wall) were outside or at the periphery of the imaging range, often the pre- and post-frame images could be used to manually extrapolate the structures. The anatomically accurate 3D model was created in Mimics, and the excess proximal and distal ends of the model were trimmed off using Netfabb Basic (Netfabb, Parsberg, Germany) (Fig. 3).

Fig. 3

Flowchart describing the process of 3D reconstruction from OCT images.

3.

Results

A 28-year-old male patient with a body mass index of 27.6 who reported nighttime snoring and had an Epworth Sleepiness score of nine was enrolled in the study. Figure 4 represents the awake (4a-c) and asleep (4d-f) results of the upper airway reconstruction. As the most intuitive parameter in describing airflow, the measurements of airspace cross-sectional area at each B-scan were calculated using ImageJ,²⁶ and the results for the nasopharynx [Figs. 4(a) and 4(d)], base of tongue [Figs. 4(b) and 4(e)], and epiglottic region [Figs. 4(c) and 4(f)] are shown in each subfigure. More than 85% of the image volume data were analyzable with a high degree of measurement confidence. The cross-sectional areas are reduced by 8%, 56%, and 32% in the nasopharynx, base of the tongue, and epiglottic regions, respectively.

Fig. 4

Upper airway OCT images of the studied subject (sleepiness score of nine and a BMI of 27·6). (a)–(c) Images taken during wakefulness, and (d)–(f) images taken during wakefulness. The corresponding cross-sectional area measurements are noted (Video 1, MP4, 4.25 MB [URL: https://doi.org/10.1117/1.BIOS.1.3.035002.s1]).

During the experiment, OCT was displayed in real time during the rotation and retraction of the imaging catheter. This is demonstrated by Video 1, which shows the OCT video streams during wakefulness and sleep. The 3D reconstructions of the corresponding airway lumens are presented in Fig. 5 in the frontal, oblique, and lateral view, redemonstrating the severe degree of oropharyngeal obstruction during sleep. Video 2 presents a paired virtual fly-through of the airway models reconstructed from Fig. 4 to graphically illustrate the differences between wakefulness and sleep.

Fig. 5

Frontal, oblique, and lateral views of volumetric reconstruction of the upper airway lumen space. (a)–(c) Data acquired during wakefulness and (d)–(f) during native sleep (Video 2, MP4, 16.32 MB [URL: https://doi.org/10.1117/1.BIOS.1.3.035002.s2]).

CFD analysis of airflow was used to understand the physics of respiration in wakefulness and sleep. Currently, practical approaches for direct measurement of instantaneous velocity and pressure within the complex geometry of the human upper airway are lacking, and therefore, numerical simulations are of great value. DNS-LBM was used to calculate changes in velocity and pressure in fluid flow. The resulting velocity magnitude contours from the DNS-LBM are illustrated in Fig. 6. They demonstrate the acceleration of the airflow through anatomical regions with reduced cross-sectional area, where a high-velocity flow was most notable at the base of the tongue region indicated by the arrows in Fig. 6. As the tongue retruded, the flow velocity increased profoundly at the base of tongue of this patient.

Fig. 6

Velocity contours from simulations based on a flow rate of $9.45\mathrm{L}/\min$. (a) Simulation based on wakefulness data. (b) Simulation based on native sleep data. Arrows indicate the most severe sites of obstruction.

Figure 7(a) is a structural depiction of the patient’s airway with seven anatomical points of interest and corresponding airflow obstruction levels, an approach that was previously validated by Wang and Elghobashi.²³ The pressure differences shown in the false color scale in Fig. 7(b) demonstrate a local minimum at point F, namely, at the patient’s base of the tongue. The calculation of the time-averaged second spatial derivative of the pressure difference ${\partial }^{2}p/\partial z^2$ could accurately pinpoint the major obstruction sites, as shown in Fig. 7(c).

Fig. 7

In accordance with previously published methods,²³ an upper airway model indicating the locations at which time-averaged $\partial \mathrm{p}/\partial \mathrm{z}$ and $\partial 2\mathrm{p}/\partial \mathrm{z}2$ were calculated. (a), (b) Pressure differences within the airway model. (c) Calculated time-averaged $\partial \mathrm{p}/\partial \mathrm{z}$ and $\partial 2\mathrm{p}/\partial \mathrm{z}2$ at each location. (NP, nasopharynx; VP, velopharynx; OP, oropharynx; HP, hypopharynx).

4.

Discussion

Accurately targeting sites of airway obstruction leads to improved surgical decision-making and outcomes in OSA patients. In this study, high-fidelity structural information and airflow data of a sleeping patient’s upper airway were obtained through OCT imaging and 3D reconstruction, allowing for precise identification of various anatomical sites with airflow obstruction. The resultant models demonstrate a severe degree of oropharyngeal obstruction during sleep by depicting the locations of significant airway lumen collapse as determined by cross-sectional measurements. Although information on the site of airway occlusion during sleep may be inferred from the structural images alone, a more comprehensive understanding of its physics is gained from CFD analysis of the airflow. With the 3D reconstruction from the OCT scans, we performed airflow simulations through a sophisticated algorithm to calculate pressure and velocity profiles. By examining local pressure differences, the dominant site of airway obstruction could be accurately identified. This is contrary to current diagnostic tools such as polysomnograms and DISE, with major limitations lending to their subjective nature. The proposed OCT upper airway imaging through a fiberoptic catheter approach combined with CFD analysis is an improved diagnostic and surgical decision-making tool for sleep surgeons and may increase the treatment success rate for OSA patients.

The reported methodology for OCT catheter imaging is minimally invasive and can be performed during native sleep without the need for anesthetics or an operating room. This provides a novel, accurate, and low-cost pathway to quantitatively assess upper airway geometry and obstruction of patients suffering from OSA. Unlike polysomnography, OCT directly visualizes the upper airway lumen geometry, which can be critical in identifying the anatomical changes that occur during sleep. OCT imaging is also superior to other imaging modalities that have been historically utilized in OSA evaluation such as computed tomography (CT) or magnetic resonance imaging (MRI).^27,28 Ultrafast (Cine) CT and MRI are available, but Cine CT exposes patients to significant ionizing radiation, whereas Cine MRI is loud (disrupting sleep) and lacks adequate temporal and spatial resolution for comprehensive dynamic imaging. The instrumentation and imaging catheters utilized in this study are physically similar to commercial in vivo pH monitors and nasogastric feeding, making them low-profile and unobtrusive. The imaging catheter operates within a protective sheath that has a 1.8-mm diameter, but it can be miniaturized to further reduce discomfort, improving the ease of imaging especially during native sleep. Upper airway OCT catheters with a diameter of $<1\mathrm{mm}$ have been previously reported.^29,30 As the diameter of the airway lumen is typically up to 25 mm, the imaging optics should possess a focal length that is approximately similar to that. Typical focusing optical components such as convex lenses or gradient-index rod lenses are the limiting factors for the miniaturization of the imaging catheter. Alternatively, a graded-index fiber can be fused with the transmitting single-mode optical fiber to reduce the catheter diameter. All in all, the minimally invasive, highly precise, and low-cost nature of OCT positions it as a superior imaging modality to characterize OSA in the future.

Although this study focuses on the utilization of OCT as a range-finding device to acquire airway lumen volume, it is worth noting that OCT can provide cross-sectional tissue information and can be used to study morphological abnormality within the upper airway at the cellular level.³¹ OCT systems powered by longer wavelength swept-source lasers have been demonstrated for enhanced imaging depth^32,33 and may be adapted in upper airway imaging where abundant cartilage tissue exists. Furthermore, OCT imaging catheters can be integrated with supplementary modalities, such as near-infrared fluorescence imaging that identifies inflammatory activity at the molecular level,^34–36 which can provide additional information regarding the disease pathology of OSA. OCT-based elastography techniques have also been proposed to study tissue mechanical response in the upper airway.^37–39 These techniques can bring further insights into the mechanisms of OSA at the cellular level by studying in vivo tissue morphological and mechanical changes.

This proposed methodology for characterizing OSA with OCT has some limitations. First, this study had only one subject, and there were no healthy controls. As such, future studies with multiple OSA and healthy subjects for thorough comparisons are warranted. Next, high-speed swept-source lasers have become commercially available and can provide an approximately eightfold improvement in imaging speed while maintaining the imaging range when adapted into our current imaging system. However, the normal respiration rate is about seven breaths per minute,⁴⁰ or $\sim 8.6\mathrm{s}$ per breath, and our current OCT setup takes $\sim 20\mathrm{s}$ to complete each pullback (with the spacing of $500\mu \mathrm{m}$ between B-scans). Therefore, we cannot finish data acquisition within a respiratory cycle. A higher imaging speed that enables faster pullback would be especially advantageous because the upper airway geometry varies over the course of the respiratory cycle. Quicker airway lumen imaging decreases the opportunity for geometric change, thus increasing the accuracy of CFD analysis. Another limitation is that manual extrapolation was done for several images where the airway would fall in proximity or outside of the imaging range, which could affect the accuracy of the results, and as such future studies should have extended imaging range to combat this limitation. Furthermore, the OCT raw images have a variable degree of artifacts (as is also shown in Fig. 4), which can theoretically become indistinctive from airway walls or other important structures on occasion. These artifacts can arise from photodetector saturation due to a large degree of specular reflection off the protective sheath. In this paper, user-defined spatial filters delineated artifacts away from the airway wall when they were in contact or indistinctive, and thus, future studies need to make this process both more accurate and expedited. In the future, the artifact phenomenon can also be reduced by several methods such as subtracting superficial plexuses/complexes after weighting,⁴¹ software/hardware advancement,⁴² tracking assisted scanning,⁴³ and machine learning automation,⁴⁴ among others. Next, the OCT catheter and spatial coordinate catheter were passed through the same sheath and as such some movement or instability may have confounded the results. In addition, the current post-processing method that constructs the 3D models for CFD analysis requires tedious manual labor due to the enormous amount of images generated in each OCT acquisition. Computer vision and other machine learning techniques have been proposed for automatic tissue layer segmentation and could reduce the necessary manpower in the future.^29,45–47

5.

Conclusions

Given the lack of objective diagnostic tools and low rates of success in sleep surgery, a method to precisely and objectively determine sites of upper airway obstruction during sleep in OSA patients is warranted. This study demonstrated accurate anatomical structure, 3D remodeling, and calculations of airflow obstruction of a sleeping patient’s upper airway using OCT images. This OCT methodology provides valuable information regarding locations and degrees of upper airway obstruction during sleep, which can benefit patient decision-making and surgical outcomes in OSA. A comprehensive understanding of the physics of upper airway geometry will open new realms of research in diagnostic and therapeutic approaches to OSA that may rapidly transform the clinical management of these patients.