1. Introduction

Endoscopic visualization of the internal surface of hollow organ systems is widely used in medical practice. Examples include endoscopy of the urinary tract, gastrointestinal tract, upper airway and respiratory tract. However a limitation of this method has been the inability, to date, to easily quantify internal dimensions during these examinations. Such objective measurements would be particularly valuable in understanding the behavior of the human airway. Repetitive collapse of the upper airway during sleep is the hallmark of obstructive sleep apnea, a common condition affecting 2–4% of middle-aged adults [1]. Current understanding of the pathogenesis of this condition has been constrained by the absence of techniques capable of quantitatively evaluating upper airway dimensions over significant periods of time during sleep. For example, measurement of cross-sectional area using nasopharyngoscopy requires subjective visual outlining of the airway wall and is markedly influenced by movement of the fiber-optic scope [2]. Poor transducer-air coupling prevents the use of ultrasound in the upper airway. Radiographic computed tomography (CT) scans involve potentially hazardous ionizing radiation, as do plain radiographs and fluoroscopy. Magnetic resonance imaging (MRI) is cumbersome and expensive, the environment is noisy and cramped/claustrophobic and the magnetic field interferes with measurements using metallic probes and electrodes (which, in turn, interfere with the generation of the MRI images).

For the reasons mentioned above, the routine use of CT, MRI, radiography and fluoroscopy is impractical for overnight sleep studies. Other airway parameter measurement techniques such as the forced oscillation technique (FOT) [3], measurement of pressures within the airway [4], and acoustic reflection [5] can not provide a comprehensive examination of upper airway dimensions or identification of the site of narrowing or collapse. Compared to these techniques, the advantages of optical techniques are high patient safety and long permissible patient exposure times from the use of low-intensity, non-ionizing radiation, as well as relatively low-cost and portable operation. A quantitative optical ranging technique such as optical coherence tomography (OCT) could provide continuous and dynamic measures of upper airway size and shape over lengthy periods.

Endoscopic OCT has successfully been used to capture subsurface images of segments of the mucosal tissues of internal organs in the respiratory, urinary, pulmonary, gastrointestinal, and reproductive systems [6,7]. A common feature of most reported applications of OCT, to date, is an axial scanning range of several millimeters or less. This range is sufficient to capture information about the subsurface structure and properties of tissue, in which the turbidity limits the depth to which images can be formed. However, size and shape determination in large hollow organs requires a scanning range of several centimeters. To date, neither suitable long-range OCT systems nor long-range OCT measurements have been reported.

A limiting factor on the axial scan range of OCT systems is the optical delay line used in the low-coherence interferometer. A number of techniques for long-range delay lines in low-coherence reflectometry have been reported previously. These include the use of recirculating delay lines [8], which can suffer from low SNR for high scan rates, and rotating glass prisms and cubes, which have nonlinear delay responses and dispersion which varies during a scan [9,10]. None of these schemes allow the independent control of group and phase velocity, therefore, the interference fringe frequency is set by the scan repetition rate, which can be inconvenient. The rapid-scanning frequency-domain optical delay line (FDODL) has been shown to overcome all of these disadvantages. The FDODL was developed for ultrafast laser-pulse auto-correlation [11] and was subsequently employed in OCT systems for group- and phase-delay scanning to achieve the high frame rates required for artefact-free OCT imaging [12]. Various names have been used for the FDODL, including “rapid scanning optical delay (RSOD)” [11], “phase control RSOD” [12], “Fourier domain RSOD” [13].

In this paper, we describe an FDODL configuration that, to the best of our knowledge, achieves the longest scanning range of any reported catheter-based OCT system using an FDODL. We also present modeling and measurements demonstrating the advantages conveyed by the configuration of our FDODL. We demonstrate its application by reporting, to our knowledge, the first in vivo size and shape measurements of the human upper airway using catheter-based OCT. In this context, OCT is used as a high-sensitivity internal surface-sensing technique enabling anatomical assessment of the airway.

2. Delay line description

The FDODL consists of a grating, lens and galvanometer mirror in a folded, double-pass arrangement. To demonstrate the importance of the configuration of the FDODL in long-range scanning, we first describe two possible variations, which we denote as “on-axis” and “off-axis”. Figures 1(a) and (b) show schematic diagrams of the top and side views of an off-axis configuration of the FDODL. The “off-axis” qualifier refers to the path of the beam, which is offset from the optical axis (the axis of the focusing lens L2) and passes above and below the axis in a round trip. Figure 1(c) shows an on-axis configuration of the FDODL that we have developed for long-range optical-delay scanning [14]. The combination of the polarizing beam-splitter (PBS), quarterwave plate (QWP), and static mirror (M) allows the beam in the delay line to scan through the axis of the focusing lens L2, which conveys advantages for long-delay scanning that are discussed in Section 3. The configuration requires the input beam to be linearly polarized at the correct angle to pass through the PBS without reflection, and then through the QWP, which changes the polarization to circular. On the return path, after traversing the grating-lens-mirror combination and QWP, the polarization at the PBS is again linear, but orthogonal to the input beam, so the PBS reflects it to the mirror M. The mirror is aligned to be perpendicular to the beam, so the reflected beam travels back along the same path in reverse, and couples into the single-mode fiber. The effect of unequal grating reflectance for s- and p-polarizations is to cause insertion loss, but the delay properties are unaffected.

 

Fig. 1. Schematic diagrams of the off-axis and on-axis configurations of the FDODL. SMF, single-mode fiber; L1 collimating lens; L2, focusing lens; PBS, polarizing beamsplitter; QWP, quarterwave plate; GRA, grating; GAL, galvanometer mirror; M, mirror. For simplicity, dispersion of the beam by the grating is not shown.

Download Full Size | PDF

The other design choice we consider is the lateral position of the galvanometer mirror pivot in the focal plane of the lens. Lateral displacement of the pivot point from the lens focal point imparts a phase modulation to the optical field, which causes modulation of the interference signal at the detector [12]. This allows bandpass detection of the signal at some frequency above the 1/f noise regime, thus, improving the signal-to-noise ratio of the system. Without the “off-pivot” setting, the system requires some other source of modulation, such as an integrated-optic phase modulator, to achieve high signal-to-noise ratio.

3. Modeling

To investigate the FDODL coupling efficiency, we performed computer modeling using ZEMAX^™ (Focus Software). The model consisted of an arrangement of single-mode fiber, collimator (ZEMAX lens prescription for the Shäfter+Kirchhoff M20 collimator lens), grating (400 lines per mm), achromatic doublet lens (190-mm focal length, ZEMAX lens prescription for the Melles Griot 06-LAI-015), and mirror, with layout as in Figure 1 and dimensions that closely match the experimental FDODL to be described in Section 4. We considered the effect of using an off-axis or on-axis configuration. For the off-axis configuration, the input beam and beam reflected from the galvanometer mirror are separated by 20% of the lens aperture and are equidistant above and below the optical axis. We also examined the effect of placing the pivot point of the galvanometer mirror either on or 10 mm away from the focal point but still in the focal plane, i.e., either “on-pivot” or “off-pivot”. The offset was chosen to provide the minimum phase modulation necessary to avoid the problem of phase fading, i.e., slightly in excess of one fringe per envelope. This choice represents the best case for the “off-pivot” configuration. The fringe frequency is proportional to the pivot offset [12], so for ease of extraction of the envelope a greater offset would usually be used to achieve perhaps 5–10 fringes per envelope. Three equally spaced wavelengths were used in the model, weighted according to a Gaussian spectrum centered on 1325 nm with a width (FWHM) of 54 nm. The model calculations included all aberrations and vignetting, and used the built-in ZEMAX fiber-coupling model, which is a hybrid overlap-integral approach following Wagner and Tomlinson [15]. We modeled the four possible combinations of the off-axis or on-axis configuration, and off-pivot or on-pivot settings. Figure 2 shows the results of calculations of the coupling efficiency versus scan position, plus a measured coupling efficiency for an FDODL using the same parameters as above, configured with the beam on-axis and the focal point at the galvanometer pivot. Relative coupling efficiency is defined as the ratio between the input and output powers of the delay line. There is good agreement between the measured and theoretical curves. The maximum range for the FDODL, determined by locating the mean wavelength at the lens edge, is given by Aλ₀/p (26.6 mm in this case), where A is the available aperture, λ₀ is the center wavelength, and p is the grating pitch [14]. Practical considerations such as the apertures of the lens, grating and mirrors, and the pitch of the grating, will ultimately set the limit on maximum achievable range. The plot shows the importance of placing the galvanometer pivot at the lens focal point. The relative loss in this case is less than 1.5 dB over the predicted range for either on-axis or off-axis configuration (the curves are not shown separately on the plot because they overlap almost exactly). The maximum relative loss over the same range for the measured curve is 1.9 dB. The results presented in Fig. 2 are consistent with previous experimental results [14] (approximately 20-mm FWHM range), recognizing that for a given aperture, the maximum theoretical range for the configuration in [14] exceeds the present configurations by a factor of 1.5 through the proportionately higher value of λ₀/p, and the pivot offset used was half of the value used here.

 

Fig. 2. The relative coupling efficiency of the FDODL versus delay for the configurations: (o) off-axis and off-pivot, (+) on-axis and off-pivot, and (Δ) both the off-axis and on-pivot, and on-axis and on-pivot. The heavy solid line is an experimentally measured response for the on-axis, on-pivot configuration.

Download Full Size | PDF

The modeling of the FDODL provides several insights into its performance. Firstly, the range for the off-pivot configurations is much worse than the on-pivot configurations because the beam focussed onto the galvanometer mirror is not reflected from the focal plane of the lens except at zero tilt angle, thus, it is not properly recollimated by the lens. This causes the return beam at the fiber to have a scan-dependent error in location and focus, which could be described as a “descan” error, and which causes a reduction in coupling efficiency. Secondly, comparing the on-axis and off-axis configurations, a slight degradation in the power efficiency of the off-axis configuration arises from increased aberration caused by the utilization of a larger fraction of the lens aperture. However, for a lens of long focal length, the NA is low and so we find that performance of the on-axis and off-axis configurations are very similar. Thirdly, after dispersion by the grating the different wavelengths of the beam pass through the lens at different lateral positions. Therefore, for a symmetric alvanometer scan the individual wavelengths are asymmetrically scanned across the lens. As a result, chromatic aberration of the lens causes an asymmetric curve of coupling efficiency versus delay. These three effects cannot be eliminated completely, but their impact may be reduced somewhat by the use of a well-designed lens. All three effects become much more pronounced if a singlet lens is used in the FDODL.

There are other interesting areas that could be examined via the same type of modeling and experimental measurement of the FDODL response, in particular, the effect of the spectral filtering of the FDODL, and its variation versus delay scanning for different configurations. Such modeling would particularly apply if one were considering ultra-high resolution OCT with broad-spectrum sources.

4. Experimental setup and procedure

The catheter-based long-range OCT system is shown in Fig. 3. The interferometer is in a Mach-Zender configuration, with three-port polarization-insensitive circulators in both reference and sample arms for maximum power efficiency [16]. An advantage of the dual-circulator Mach-Zender configuration is that light in the reference arm will pass only once through a device with a relatively high insertion loss, such as an integrated-optic phase modulator. Power efficiency is important in attaining a sufficient signal-to-noise ratio over a long-range scan. A phase modulator with a 4-dB insertion loss was used to serrodyne modulate the interference signal, allowing bandpass detection [17]. Double-balanced detection was used to eliminate excess intensity noise [18]. The FDODL with the same parameters as above and in the on-axis, on-pivot configuration was used in the reference arm. As the modeling has shown, the use of an on-pivot configuration is more important in achieving maximum scanning range than the choice between on- or off-axis configuration. The return loss of the on-axis configuration is polarization dependent (it only passes a single linear polarization), so to ensure minimum power loss the FDODL requires careful polarization control so that the input light is linearly polarized along the transmission axis of the PBS. The FDODL employs a galvanometer mirror driven by a triangular waveform at 100 Hz, which results in a constant delay scan speed of 5.8 m/s. The galvanometer scan rate was limited by the maximum possible data acquisition rate available at the time of the experiment (1.25 Msamples/s). The use of the on-pivot FDODL configuration allows the use of a small galvanometer mirror, with a low rotational inertia, so it could potentially be scanned much faster. The axial scan range achieved was 26 mm (+/- 3.6° galvanometer tilt), limited by the 50-mm aperture of the optical elements in the FDODL. The 9-mW polarized broadband source (λ₀=1325 nm, Δλ=54 nm) used gave a calculated axial resolution of 14.3 µm [19], and a measured value of 17.4 µm. This resolution did not noticeably vary over the FWHM range of the FDODL depth scan. The detected interference signal is bandpass filtered and logarithmically demodulated before digitisation by a data acquisition card. The sample arm of the interferometer consists of a transverse-scanning catheter-mounted probe, which is connected to the system by a fiber-pigtailed optical rotary joint (FORJ). The distal end of the probe consists of a 0.276-pitch GRIN lens of 1.0-mm diameter attached to the optical fiber, with a 0.7-mm right-angle prism attached to the lens to deflect the beam transverse to the probe. The waist of the probe beam is 40 mm from the distal end, and the beam spot size at the waist is 180 µm. There is an obvious tradeoff in the use of a low-numerical aperture probe that extends the depth of field without requiring dynamic focussing but reduces collection efficiency. However, size and shape measurement requires only surface location, therefore, collection efficiency is traded off for long range. The point-spread function of the system is highly asymmetric; the axial resolution is greatly exceeded by the transverse resolution. However, this is not a problem for large hollow organ size and shape determination, which does not require micrometer-scale high resolution or subsurface imaging. The probe launches 1.4 mW of optical power, and the system is sensitive to reflected power corresponding to a return loss of as low as -96 dB, which is 6 dB worse than the theoretical value of -102 dB for the system [16]. This reduction in dynamic range is due to imperfect matching of the electronic filter bandwidth to the signal bandwidth and the logarithmic demodulation electronics, which are not completely optimized. For in vivo profiling of the upper airway a sealed, transparent naso-esophageal catheter (external diameter 3.8 mm) was passed through the external nares traversing the pharynx until the tip was positioned in the middle third of the esophagus. The probe was passed through the catheter with the distance of the probe tip to the external nares monitored. The probe was then fixed at various levels within the pharynx, and cross-sectional images obtained. These were reconciled with anatomical landmarks and with cross-sectional images obtained from CT scans at the same levels in the subject. A single rotation of the probe took 3.75 seconds. During a measurement, the volunteer breathed shallowly for several seconds to minimize motion artefacts. The FORJ is mounted on a motorized translation stage, which allows the probe to be automatically moved to various locations along the catheter. This has the benefit that, in addition to capturing cross-sectional images at fixed locations, it is also possible to move the probe along the catheter while simultaneously capturing images. In a second procedure, the probe was moved at a constant velocity of 0.2 mm per second away from the distal end of the catheter in order to capture a “pullback” sequence.

 

Fig. 3. Schematic diagram of the endoscopic long-range OCT system. BBS, broadband source; PM, phase modulator; PC, polarization controller; M, motor.

Download Full Size | PDF

5. Experimental results

Figure 4 shows six in vivo measurements on the volunteer, five taken at different locations along the airway from the nasal cavity to the hypopharynx, and a sixth taken in the esophagus. Ranging up to 26 mm from the probe was possible, however, the signal from the air-tissue reflection fell below the noise level at some locations during a scan. The system sensitivity is -96 dB, the dynamic range at the output of the logarithmic demodulator is 42 dB, and the images show the lowest 25 dB of that range. Each Cartesian image comprises 400×400 pixels, down-sampled and transformed from the raw polar-form data set (375×5600). The transformation process does not lose any significant information. Apart from some sections of the airway walls that are “shadowed” by the airway structure, or show signal fading due to highly oblique surfaces, the airway profiles are complete. A number of anatomical features in the airway are noted. The image shows the inner and outer surfaces of the catheter, which are useful calibration aids. These images are expected to be clinically very useful, as they represent a method of determining location and dynamic behavior of the airway walls. Such information cannot presently be routinely gathered over long periods of time.

 

Fig. 4. Six in vivo measurements taken of the airway (and esophagus) of a human volunteer, arranged by distance into the airway: (a) nasal cavity, (b) nasopharynx, (c) velopharynx, (d) oropharynx, (e) hypopharynx, (f) esophagus. Some anatomical features are noted: nasal septum (N), middle turbinate (MT), inferior turbinate (IT), posterior nasal spine (P), base of uvula (BU), base of tongue (BT), arytenoid cartilage (AC). The two circles at the center of the images are the reflections from the inner and outer surfaces of the catheter.

Download Full Size | PDF

Figure 5 shows an in vivo OCT measurement of the hypopharynx of the volunteer, Fig. 5(a), and a CT scan, Fig. 5(b), taken on a separate occasion at approximately the same location. The OCT image shows the reflection from the hypopharynx epithelial tissue. The OCT image and the CT image show similar shape and dimensions of the airway, although precise comparison between the images is not intended. The images are not expected to be quantitatively comparable because they were acquired under different breathing protocols and correlation of location is only approximate (+/- 10 mm). Airway shape and size can vary by a significant amount during the breathing cycle and for different head and neck postures. In the CT scan, the radiographer has indicated the airway surface with a solid line.

 

Fig. 5. (a) Measured cross-section of the in vivo hypopharynx of a human volunteer. (b) CT scan of the volunteer’s airway at a location close to where the OCT scan was performed.

Download Full Size | PDF

Figure 6 contains three video recordings of in vivo upper airway pullback sequences, reduced to 280×280 pixels to conform to presentation guidelines. The videos improve visualization of the airway structure, because human perception of faint features is better when faced with dynamic images rather than static images. The pullback sequences may be used to extend quantitative measurements of airway structure to three dimensions.

 

Fig. 6. Three video recordings of in vivo pullback measurements in various sections of the upper airway. They are: (a) velopharynx to nasopharynx (2.41 MB), (b) nasopharynx to nasal cavity (1.67 MB), (c) nasal cavity (1.75 MB).

Download Full Size | PDF

6. Discussion and conclusions

For shape determination in the upper airway, the rotation rate would ideally be fast enough to avoid breathing-related motion artefacts; 0.5 seconds per rotation is estimated to be sufficient, and such rotation rates have previously been demonstrated [7]. The probe rotation rate of our current system is too low to avoid motion artefacts during normal breathing. The rate is currently limited by the probe rotation components of the system, which are being redesigned for increased speed. The rotation rate for constant transverse resolution is limited by the requirement to ensure at least one axial scan per transverse resolution cell. Thus, a higher rotation rate will require a higher axial scan rate which, in turn, is limited by the FDODL galvanometer. A resonant galvanometer has been operated in an FDODL at a rate of 4 kHz [13]. As an example, if a rotating endoscopic OCT system with radial range of 25 mm were required to maintain a transverse resolution of 200 µm over that entire range, a 4 kHz axial scan rate would allow a maximum rotation rate of 5 Hz. As noted above, this rate is sufficient for the upper airway. Any increase in scanning rate is accompanied by a proportionate decrease in sensitivity, which would degrade image quality. An alternative means to avoid motion artefacts is to separately record the breathing cycle and correlate it with the scan. Similar co-registration techniques have been used in intravascular ultrasound [20].

Hollow organs such as the airway can be highly non-circular in cross-section and the linear scan range required can exceed 50 millimeters, depending on the location of the catheter. It is also possible that, at times, the airway wall can be sufficiently concave or irregular so that part of the airway wall is obscured from the probe. The detection of reflections at long range from the oblique surfaces that may arise is challenging and remains to be thoroughly investigated. Our results demonstrate, however, that substantial segments of organ circumferences can be detected, sufficient to provide medically useful information. Optical detection has the advantage that it may be used repetitively over long periods of time, for example, to routinely monitor sleeping patients with obstructive sleep apnea. For reasons documented earlier (see Introduction), CT, MRI and ultrasound cannot feasibly be used in this way. Previous studies have demonstrated that placement of a catheter in the upper airway and esophagus does not induce additional sleep disturbance in patients with sleep-related breathing disorders [21].

Measurement of upper airway size and shape is central to defining the pathophysiology of airway collapse during sleep (obstructive sleep apnea). Thus, the measurements obtained from this optical device have the potential to substantially advance our understanding of the pathophysiology and management of this condition. Long-range OCT has the distinct advantage of allowing dynamic measurements to be made during breathing. Such measurements during sleep will help determine factors contributing to airway collapse and also allow the effect of treatments on airway caliber to be directly measured. As obstructive sleep apnea is a common condition and associated with substantial morbidity, the measurements provided by this device are potentially very important.

In conclusion, we have identified that the on-pivot configuration of the double-pass FDODL allows the maximum achievable scan range and explained the reasons for this. Our long-range FDODL has enabled an axial scan range of 26 mm, the longest reported range of an endoscopic FDODL-based OCT system. The long-range scanning capability was demonstrated by in vivo measurements showing profiles from various locations in the human upper airway and esophagus. Long-range OCT shows clear promise as a new modality for quantitative imaging of the human upper airway and any other hollow organ system accessible by a catheter or endoscope.