1438
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 4, APRIL 2008
Anatomical Optical Coherence Tomography for Long-Term, Portable, Quantitative Endoscopy Matthew S. Leigh*, Julian J. Armstrong, Alexandre Paduch, Jennifer H. Walsh, David R. Hillman, Peter R. Eastwood, and David D. Sampson
Abstract—In this paper, we report on anatomical optical coherence tomography, a catheter-based optical modality designed to provide quantitative sectional images of internal hollow organ anatomy over extended observational periods. We consider the design and performance of an instrument and its initial intended application in the human upper airway for the characterization of obstructive sleep apnea (OSA). Compared with current modalities, the technique uniquely combines quantitative imaging, bedside operation, and safety for use over extended periods of time with no cumulative dose limit. Our experiments show that the instrument is capable of imaging subjects during sleep, and that it can record dynamic changes in airway size and shape. Index Terms—Biomedical optical imaging, endoscopy, optical coherence tomography, sleep apnea, upper airway.
I. INTRODUCTION NDOSCOPIC 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. The most common tool used in these examinations is the fiber-bundle video endoscope. Although video endoscopy is extremely useful in clinical practice, it suffers from a limitation that restricts its use in research and in certain clinical applications.
E
Manuscript received May 27, 2007; revised October 10, 2007. This work was supported in part by the Medical Research Foundation of Western Australia, the Australian Health Management Group, and more recently the National Health and Medical Research Council, Australia, under Development Grant 303319. The work of P. R. Eastwood was supported by the National Health and Medical Research Council, Australia, under R. Douglas Wright Fellowship 294404. Asterisk indicates corresponding author. *M. S. Leigh is with the Optical+Biomedical Engineering Laboratory (OBEL), School of Electrical, Electronic & Computer Engineering, University of Western Australia, 35 Stirling Highway, M018, Crawley, W.A. 6009, Australia (e-mail:
[email protected]). J. J. Armstrong and D. D. Sampson are with the Optical+Biomedical Engineering Laboratory (OBEL), School of Electrical, Electronic & Computer Engineering, University of Western Australia, Crawley, W.A. 6009, Australia. A. Paduch was with the Optical+Biomedical Engineering Laboratory (OBEL), School of Electrical, Electronic & Computer Engineering, University of Western Australia, Crawley, W.A. 6009, Australia. He is now with Leica Geosystems, Heerbrugg CH-9435, Switzerland. J. H. Walsh and D. R. Hillman are with the West Australian Sleep Disorders Research Institute (WASDRI), Department of Pulmonary Physiology, Sir Charles Gairdner Hospital, Nedlands, W.A. 6009, Australia. P. R. Eastwood is with the West Australian Sleep Disorders Research Institute (WASDRI), Department of Pulmonary Physiology, Sir Charles Gairdner Hospital, Nedlands, W.A. 6009, Australia, with the School of Anatomy & Human Biology, University of Western Australia, Crawley, W.A. 6009, Australia, and also with the Curtin University of Technology, Perth, W.A. 6845, Australia. Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBME.2007.913409
This limitation is the lack of an easy method to quantify the internal dimensions of the organ under observation [1], [2]. The main techniques for quantitative imaging of large hollow organs are X-ray computed tomography (CT) and nuclear magnetic resonance imaging (MRI). Both of these techniques, while useful and accurate, have shortcomings—in particular, for longterm or repeated measurements. CT scans involve potentially hazardous ionizing radiation, which is subject to dose limitations. MRI is cumbersome and expensive, the environment is noisy and claustrophobic, and the magnetic field interferes with measurements using metallic probes and electrodes (which, in turn, interfere with the generation of the MRI images). As a consequence, CT and MRI are impractical for research into internal hollow organs that requires continuous monitoring over a long period of time. Ultrasound has found some use, particularly in vascular applications, but due to poor-transducer-air coupling it is not suitable for use in primarily air-filled hollow organs. Compared to CT and MRI, the advantages of optical techniques are high patient safety and long permissible exposure times from the use of low-intensity, nonionizing radiation, as well as relatively low-cost and portable operation. A quantitative optical measurement technique such as optical coherence tomography (OCT) [3], [4] could provide continuous and dynamic measures of hollow organ size and shape over lengthy periods. Endoscopic OCT has been used to capture subsurface images of segments of the mucosal tissues of internal organs in the respiratory, urinary, pulmonary, gastrointestinal, and reproductive systems [5], [6]. A common feature of most reported applications of OCT, to date, is an axial scanning range of several millimeters or less. The objective of such applications is to capture information about the subsurface structure and properties of tissue, in which the turbidity limits the depth to which images can be formed. Size and shape determination in large hollow organs, such as the airway or bowel, does not require subsurface imaging and requires an extended scanning range of several centimeters. To date, our long-range OCT system [7] is the only such reported system designed for taking suitably long-range measurements, with initial intended application to the upper airway [8]. An OCT system capable of long-range OCT measurements using adaptive ranging has been reported [9]; however, such an approach is unlikely to be suitable for the measurements we describe. It continuously scans over a short range (1.5 mm), positioned anywhere in a longer range (7 mm). Correct functioning of its adaptive ranging depends on uninterrupted tracking of the tissue surface. Any real or apparent discontinuities in the tissue surface (for example, caused by the shadowing present in parts
0018-9294/$25.00 © 2008 IEEE
LEIGH et al.: ANATOMICAL OPTICAL COHERENCE TOMOGRAPHY FOR LONG-TERM, PORTABLE, QUANTITATIVE ENDOSCOPY
1439
Fig. 1. Schematic diagram of the optical coherence tomography system used for ranging to airway tissue. PM: phase modulator; PC: polarization controller; BBS: broadband source.
Fig. 2. (a) Schematic diagram of the distal end of the fiber-optic probe used for the experiments, in position inside the PVC catheter. (b) Photograph of the probe with the protective metal tube removed.
of the upper airway) would likely result in missing information while tracking was regained. In this paper, we describe technical developments and methods regarding our hollow organ profiling system, which we have dubbed anatomical optical coherence tomography (aOCT). We present improved validation results, demonstrating our system’s accuracy and resolution compared against X-ray CT of a phantom. We present representative results, showing the system applied to the human upper airway. We show the potential for dynamic measurements by demonstrating the first breath-gated aOCT upper airway imaging in awake patients undergoing continuous positive applied pressure (CPAP) treatment. We highlight the unique ability of aOCT to monitor the size and shape of the upper airway in sleeping patients by reporting an obstructive apnea event. We also discuss the issues involved in converting the data acquired from a 2-D to a 3-D presentation, and report the first 3-D aOCT reconstruction of an upper airway. II. MATERIALS AND METHODS A. Instrument Design 1) Optical Hardware: The aOCT system is based on an optical coherence tomography imaging system with a fiber-optic endoscopic probe. The long range required for hollow organ
imaging was achieved by making several modifications to a standard Mach–Zehnder-type interferometer, configured for OCT operation. These modifications and associated design issues are described in [7], and will not be discussed in detail here. A schematic diagram of the aOCT system is presented in Fig. 1. The interferometer incorporates an integrated-optic phase modulator to produce an amplitude-modulated interference signal. Depth scanning is performed using a frequency-domain optical delay line (FD-ODL) [10], [11]. The delay line is driven by a triangle wave at 250 axial scans per second. The system is capable of scanning up to a radius of 26 mm, giving a 52-mm diameter image slice. The optical source is a superluminescent diode producing 21 mW of polarized light at 1301 nm, with a bandwidth of 34.2 nm. The measured axial resolution is 33 m, and the resolution calculated from the measured spectrum is 22 m. The difference is due to uncompensated dispersion in the system. The sample arm of the interferometer ends in a rotating optical fiber probe. The probe is similar to that used in endoscopic OCT imaging, but optimized for long range. A diagram and microscope photograph of the distal end of the probe are presented in Fig. 2. The fiber probe consists of 2.4 m of optical fiber encased in a biplex torque transmission coil made from stainless steel wire. The probe terminates in a cylindrical graded-index
1440
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 4, APRIL 2008
lens (NSG America SLW-1.0, Tokyo, Japan) of 1.0-mm diameter, capped with a right-angle prism of 0.7-mm width for beam deflection at 90 . The GRIN lens has a focal length of approximately 13 mm. The microoptic components are encased in a 1.3-mm steel tube for protection (not shown in the photograph). The proximal end of the probe is connected to a fiber-optic rotary joint to allow full 360 radial scanning, rotating at 1.25 r/s. The rotary joint and probe coupling are mounted on a movable carriage to allow -axis scanning. The probe is designed to rotate within a transparent PVC catheter of 3.0-mm outer diameter. This catheter provides protection and waterproofing for the optical components, and allows the probe to rotate and move axially without being sensed by the patient. The probe launches 4.0 mW of optical power. The beam spot size at the waist (located at 8.4 mm from the probe) is 180 m. The beam waist varies from 180 to 240 m over the 26-mm range. The system can detect reflections as low as 98 dB. Signal-to-noise ratio (SNR) is limited by beat noise from parasitic reflections within the sample arm of approximately 40 dB. 2) Signal Processing: The optical signal is detected using a balanced photoreceiver. The amplitude modulation of the OCT signal is at a frequency of 1.3 MHz. The signal is preamplified and passed through a bandpass filter with a bandwidth of 600 kHz, centered at 1.3 MHz. The bandwidth is chosen to maximize the SNR, rather than preserve axial resolution, and also to account for motion of the sample, which causes a Doppler shift in the modulated signal. The filtered signal is fed into a logarithmic amplifier, which log-compresses and demodulates the signal. A 300 kHz, low-pass filter at the output completes the demodulator. This signal is acquired directly by a personal computer which controls the aOCT system. The galvanometer-limited axial scan rate of 250 Hz is currently the limiting factor on the acquisition speed of the system. The axial scan range (26 mm) and the spot size at the end of that range (240 m) determine a maximum number of 680 resolution elements around the circumference of the scan area. However, the rotational scan rate of 1.25 Hz corresponds to only 200 axial lines per revolution. This undersampling at the outer edge of the scan range is tolerated because the rotation rate is close to the minimum necessary to avoid motion artifacts due to the breathing of the subject. Note that the sampling rate required is proportional to the distance from the axis, and therefore, images with data near the center of the frame suffer less from this effect. The axial sampling rate is dependent on the axial resolution of the aOCT system (33 m), giving 790 resolution elements across the full range. We operate the data acquisition card (see the following for details) at 2250 pixels per line, above the Nyquist limit. 3) Electromechanical Hardware: A full 3-D image from the aOCT system scans through three axes: , , and . The scan is performed by the FD-ODL, and is the fastest scan (analogous to the A-scan of ultrasound and other OCT systems). The axis is scanned by rotating the fiber probe, using a direct current (dc) motor driven by a 12-V pulsewidth modulated signal. The slowest scan, in the -axis, uses a stepper motor driving a linear ball screw to move the probe from site to site within the airway (either as a discrete seeking movement to a particular location or as a continuous scan). An assembly language
Fig. 3. Schematic diagram of the signal processing path. The supervisory thread monitors the other threads and controls interthread communication.
program running on an Atmel AT90S8515 microcontroller controls both the dc and stepper motors. Position information is obtained by step counting for the stepper motor, and by a rotary optical quadrature encoder for the dc motor. The microcontroller acquires the position information and stores it for batch transmission to the controlling personal computer (PC). The acquisition clock is locked to the depth scan waveform to minimize sampling jitter. The microcontroller communicates with the PC through an RS-232 serial port using a simple challenge-response protocol, with ASCII encoding. All motion parameters can be set from the PC software, and a useful subset can also be set from the instrument front panel. The microcontroller circuit includes an expansion port to connect a 4-channel analog-to-digital converter (ADC) operating at a comparatively low sample rate, for acquiring related experimental data alongside the images. Alternatively, the data is collected by an external lab data acquisition system (Powerlab, ADInstruments, Sydney, Australia) and integrated into the aOCT data in postprocessing. The two recording systems are synchronized using a set of clock, synchronization, and event signals output from the aOCT system, which are recorded by the Powerlab system. 4) Software: The aOCT system is controlled by a custom multithreaded application written in Borland C++, running on a Windows 2000 PC. The application is designed to perform three main tasks simultaneously—digitize and save to disk the demodulated OCT signal, scan-convert the OCT signal to a Cartesian image and display it, and control the electromechanical aspects of the apparatus. All the acquisition and control tasks requiring accurate timing are carried out in real-time hardware by either the microcontroller or the DAQ card. The ADC of the demodulated OCT, signal is performed by a 12-bit DAQ card (National Instruments, PCI-6110E, Austin, TX) allowing sample rates up to 5 MS/s. The software allows the sample rate to be reduced if less axial resolution is required, in order to lower the disk space requirements of the saved OCT data. The application comprises six main threads (see Fig. 3 for a block diagram). The data acquisition thread runs continuously, copying data from the DAQ card in large blocks and distributing it to the other threads. A large circular buffer is used to improve stability in the event of unexpected system
LEIGH et al.: ANATOMICAL OPTICAL COHERENCE TOMOGRAPHY FOR LONG-TERM, PORTABLE, QUANTITATIVE ENDOSCOPY
1441
Fig. 4. The aOCT and X-ray CT comparison. The target is a paraffin wax block with 23 holes of various shapes and sizes ranging from 10 to 50 mm in size. The images are arranged in columns with the CT and aOCT scans of each hole placed together for visual comparison. The CT scans show each hole in black as an absence of the wax. The aOCT scans show an outline of the holes, as most of the signal arises from the surface backscattering of the periphery of the hole.
load. The “motor position” thread maintains communications with the microcontroller and handles requests for changes in the probe motion parameters. A worker thread saves the data to disk in a continuous binary stream. This thread is also responsible for recording motor position data and timestamps. The display thread performs polar-to-Cartesian scan conversion using nearest-neighbor interpolation, and displays the resulting image for the user. The display is updated in real time, and provides instant feedback for clinical users. The supervisory thread runs error logging and cleanup tasks, and ensures the rest of the application remains synchronized and responsive. The user interface is in a separate thread to allow user interaction during data recording. A suite of utilities was developed along with the main data acquisition application to organize and postprocess the data. 5) Upper Airway Clinical Imaging Requirements: The main parameters of interest are the maximum cross-sectional dimensions of the upper airway lumen, the dynamics and scale of lumen motion, and the required imaging resolution. In addition, patient tolerance of the catheter is important. The 3.0-mm diameter catheter we used is similar to naso–pharyngeal catheters used in other research and clinical applications, and we have found it to be well tolerated by our subjects. Previous studies have demonstrated that placement of such a catheter in the upper airway and esophagus does not induce additional sleep disturbance in patients with sleep-related breathing disorders [12]. The size of the human airway varies greatly between locations and individuals, but an initial data set of 18 patients gave an average size of approximately 25 mm in the lateral dimension (the lateral size is typically the largest dimension in the upper airway), and up to an average maximum extent of 40 mm. Dimensions are typically smaller in patients with airway-related pathologies. Catheters such as those used in aOCT typically rest against the posterior wall of the airway, roughly centered laterally. This implies a maximum imaging radius of 26 mm would cover most individuals of clinical significance. If the catheter were to be positioned very eccentrically, which occurs in prac-
tice occasionally, a maximum airway size of only 26 mm could be measured. The primary source of motion in the airway is respiration. Quiet breathing is typically at a rate of approximately 15 breaths/min. The system has initially been designed for a rotation rate of 75 rotations/min (1.25 Hz), giving approximately five full image frames per breath. This rate is adequate for capturing quiet breathing with good fidelity, but improving the rotation speed to better account for uneven respiration or sudden events is an ongoing task. In medical imaging modalities, imaging gating is a standard technique to reduce the impact of subject motion during acquisition, in this case respiratory motion [13]. We demonstrate its effectiveness for aOCT in Section III. The requirement on resolution for aOCT is greatly relaxed compared with subsurface endoscopic OCT applications. The aOCT modality is intended for detecting the location of the tissue surface and imaging the gross anatomical features of the hollow organs. Existing gold standards for hollow organ measurements, X-ray CT and MRI, both typically have resolutions in the half-millimeter range, and so the resolution of aOCT comfortably exceeds these. B. Experimental Method The aOCT system can be operated in the so-called “pullback” mode, in which a 3-D helical scan of the airway region of interest is taken. The probe, located within the stationary catheter, is positioned at the start point of the scan, usually in the esophagus. The position is determined by measuring the distance between the probe head and an externally visible anatomical landmark, typically the external nares. It is then pulled back slowly (at 0.2 mm/s or less) until the desired region has been imaged. The resulting data is a collection of voxels located in 3-D space, which can be sliced using either commercial or custom software (in this case, a custom Matlab script). Currently, no correction is applied for the curve of the airway, and the images presented
1442
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 4, APRIL 2008
Fig. 5. Images taken from an aOCT pullback scan through the human upper airway. Images (a)–(e) are slices extracted at the marked locations. The anterior/ posterior and lateral views are longitudinal slices through the entire scan (120 mm long). Scale bars are 10 mm.
here are a representation of the airway as if it were straight along its entire length. Pullback imaging is used to preview the airway and locate anatomical landmarks for future probe location. However, many studies require a single location to be monitored over time. The system operates in this mode by keeping the probe rotating at the same location and recording continuously. The resulting data can be compiled into a movie for observing changes, or particular sections can be extracted for more detailed quantitative analysis. The aOCT system is capable of continuously recording for more than a day, if required. The respiratory cycle is continuously monitored by respiratory inductance pneumography (Respitrace, Ambulatory Monitoring, Ardsley, NY) with the transducers placed at the level of the nipples and umbilicus to provide a measure of volume displacement. The signals were recorded at 1 kHz on the Powerlab data acquisition system, simultaneously with the aOCT images, allowing a posteriori gating of the aOCT data to accurately match it to a particular phase of respiration. III. RESULTS A validation study was carried out by comparing aOCT against a gold standard imaging modality: X-ray CT. A numerical summary of this validation study has been previously published [8]. A phantom was constructed from cast paraffin wax with holes of varying dimensions and shapes. The hole sizes spanned the range 10–50 mm, corresponding to the size range of the human airway. A preliminary measurement was
Fig. 6. Breath-gated aOCT image showing the effect of CPAP on the pharynx. Measurements were taken in the oropharynx, and each shade is an averaged image of 4–6 breathing cycles over a specific phase of respiration (the time from peak pressure from 20% to 30% of the cycle). The inner trace shows an applied pressure of 3-cm H O, center shows 9-cm H O and outer shows 12-cm H O. The small black circle in the center is the PVC catheter. Horizontal FOV is 52 mm.
conducted with simultaneous aOCT and CT imaging of the same phantom (image not shown). This measurement showed agreement between the two modalities to within the resolution limit of the CT scanner ( 0.5 mm). A larger scale comparison was then performed, with the resulting images shown in Fig. 4. Note the aOCT images of paraffin wax are fainter than those of airway tissue because of the low backscattering from wax at this wavelength.
LEIGH et al.: ANATOMICAL OPTICAL COHERENCE TOMOGRAPHY FOR LONG-TERM, PORTABLE, QUANTITATIVE ENDOSCOPY
1443
Fig. 7. Recording of a subject during an obstructive apnea event. The top graph shows mask pressure, with a higher pressure corresponding to expiration. The lower graph shows airway cross-sectional area as measured by aOCT. The subject is asleep at t 0 s, and from t 11 s to t 21 s, the airway has closed involuntarily. At t 22 s, the subject aroused from sleep and resumed breathing. The aOCT images show the airway state at labeled points.
=
A pullback scan of the upper airway of a volunteer is shown in Fig. 5. Two views of the data are presented. The anterior/posterior and lateral views are slices along the -axis of the data. The scan was taken between the esophagus and the nasal septum. The – slices are taken in the plane orthogonal to the catheter axis, and show the shape of the airway at the marked locations. In Fig. 5(b) (hypopharynx, behind the epiglottis), the epiglottis is visible at the top of the image. Fig. 5(c) (oropharynx, behind the tongue) shows the base of the tongue (also at the top of the image). Fig. 5(d) (nasopharynx, behind the soft palate) shows the airway near its narrowest point in this individual, and Fig. 5(e) (also nasopharynx) shows the airway becoming wider near the nasal cavity. Typically, the pullback scan is compiled into a video that can be immediately reviewed in order to locate anatomical sites of interest. Fig. 6 demonstrates an application of respiratory motion gating to enable accurate assessment of the effect of increasing pressure within the pharynx, using continuous positive airway pressure (CPAP), at a single location. A subject with obstruc-
=
=
=
tive sleep apnea (OSA) was scanned for several minutes while CPAP was applied through a nasal mask. The mask pressure was changed once per minute. The figure shows a composite averaged representation of this data, showing the degree of change in airway size and shape due to the applied pressure. Each trace on the image was compiled from one minute of data. Data was gated to compile an averaged image from a section of the breathing cycle from 20% to 30% after peak inspiration. The result is a representation of the airway at a particular applied mask pressure. By comparing the three traces, it can be seen that the airway wall for this subject was more compliant in the lateral than anterior/posterior dimension, with a larger lateral size corresponding to increased airway pressure. Measurements such as these may be a valuable indicator of the mechanism of collapse, and possibly the efficacy of treatment, in individuals with OSA. The aOCT system is also capable of continuous airway imaging during sleep, something not previously feasible in a bedside setting. In Fig. 7, we present a recording of an obstruc-
1444
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 4, APRIL 2008
Fig. 8. The aOCT image with the GVF algorithm applied. The thick line shows the starting position of the snake, and the thin lines show intermediate positions. The thick inset line shows the final position of the snake, tracing the contour of the aOCT data closely.
tive apnea event in a subject with severe OSA. In addition to the aOCT system, a face mask was connected to deliver CPAP and measure respiration pressure. The subject was asleep at the start of the displayed section of data and, after two breaths, the airway collapsed and airflow ceased (an apnea). The aOCT slices show the airway state at significant times. Of particular 14, 17, and 21 s, which show the airway note are slices at collapsed during the time of no change in mask pressure (i.e., no airflow). The violent motion of the airway wall at the first 26 s slice. The airway postarousal breath is evident in the remained closed for approximately 11 s. The aOCT cross-sectional area data shows the time at which the airway reopens, which agrees closely with the mask pressure data in the plot. The area data shows that the airflow ceases slightly before the airway closes fully—this is likely due to the site of the collapse being slightly above or below the imaging location. The airway 25 s is slightly undersampled due to the area data after rapid breathing of the subject after the apnea. The task of image analysis is made difficult by the noise characteristics of the aOCT images, as well as the overall moderate image contrast. Most processing to date has been done by manual image tracing, but we have investigated the performance of various contour extraction algorithms. A suitable contour extraction algorithm would allow rapid and accurate extraction of physiologically relevant data, such as airway cross-sectional area and dimensions. The most promising algorithm so far is
the gradient vector flow algorithm, which uses deforming contours informally known as “snakes.” This algorithm, described by Xu et al. [14], calculates a map of the gradients of the image and iterates the “snakes” towards the lowest energy state, which should correspond to the boundaries of the object. We present a preliminary test in Fig. 8. From an arbitrary starting position, the snake finds the contour of the aOCT image effectively. The technique is not yet advanced enough to apply automatically to a large data set, and the results require close checking by a human operator. However, it shows promise in reducing the time required to analyze a typical data set. Recording a 3-D data set by performing a pullback scan leads naturally to displaying it as a 3-D reconstruction. This is a technically challenging task, as the aOCT data does not currently have sufficient contrast to be directly used for volume visualization. The technique of extracting a vector representation from a 2-D slice (either manually or by using an algorithm such as that described previously) appears promising. This method has been used to generate Fig. 9. A 20-mm pullback scan was processed into horizontal slices and vector curves were extracted from each slice, matching the inner edge of the airway reflection. These slices were then stacked and displayed using a 3-D Matlab plot. The transition to the esophagus at the lower end of the scan is prominent. The 3-D reconstructions are a powerful diagnosis and research tool, and from images such as Fig. 9, it could be pos-
LEIGH et al.: ANATOMICAL OPTICAL COHERENCE TOMOGRAPHY FOR LONG-TERM, PORTABLE, QUANTITATIVE ENDOSCOPY
1445
K. Shepherd from the Department of Pulmonary Physiology, Sir Charles Gairdner Hospital, Nedlands, W.A., Australia; and S. Schwer and J. Ng from the School of Electrical, Electronic and Computer Engineering, University of Western Australia, Crawley, W.A., Australia. REFERENCES
Fig. 9. The 3-D image of the esophagus and hypopharynx formed from multiple horizontal slices of aOCT data. The black outline represents the outline of the uppermost single horizontal slice.
sible to identify the site(s) of upper airway collapse, the length of the collapsing segment, and the relationship of collapse to phase of respiratory cycle and sleep state. Such measurements will permit calculation of segmental compliance (volume versus pressure) and allow comparison of upper airway structure and function between normal and apneic subjects at a level of complexity and comprehensiveness that cannot be obtained from 2-D images. IV. CONCLUSION We have developed a system capable of imaging large hollow organs for extended periods of time, which gives an accurate quantitative assessment of size and shape. Such capabilities show great potential in upper airway clinical practice, as well as research into sleep apnea and other sleep and airway-related conditions. The preliminary results shown here indicate that aOCT is capable of resolving the changes in airway size and shape due to dynamic external factors, such as applied air pressure. We have also demonstrated quantitative imaging of an apnea event during sleep, with accompanying biometric data. The current standard for sleep apnea diagnosis is overnight polysomnography, and our instrument has been used in conjunction with a sleep clinic polysomnography suite. The integration of airway imaging with sleep state data has great potential for studying the mechanisms of airway collapse on an individual basis. Although the processing technique used currently is not suitable for routine 3-D display, such visualization is an analysis tool that fits well with the data produced by the aOCT system. These visualizations could potentially be useful in a wide range of clinical applications. ACKNOWLEDGMENT The authors would like to thank the following individuals for their support and contributions to this work: K. Maddison and
[1] P. P. Hsu, H. N. C. Han, Y. H. Chan, H. N. Tay, R. H. Brett, P. K. S. Lu, and R. L. Blair, “Quantitative computer-assisted digital-imaging upper airway analysis for obstructive sleep apnea,” Clin. Otolaryngol., vol. 29, pp. 522–529, 2004. [2] K. F. Mansour, J. A. Rowley, and M. S. Badr, “Measurement of pharyngeal cross-sectional area by finite element analysis,” J. Appl. Physiol., vol. 100, pp. 294–303, 2006. [3] D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, C. A. Puliafito, and J. G. Fujimoto, “Optical coherence tomography,” Science, vol. 254, pp. 1178–1181, 1991. [4] D. D. Sampson and T. R. Hillman, “Optical coherence tomography,” in Lasers and Current Optical Techniques in Biology, G. Palumbo and R. Pratesi, Eds. Cambridge, U.K.: Royal Society of Chemistry, 2004, ch. 17, pp. 481–571. [5] A. M. Sergeev, V. M. Gelikonov, G. V. Gelikonov, F. I. Feldchtein, R. V. Kuranov, N. D. Gladkova, N. M. Shakhova, L. B. Snopova, A. V. Shakov, I. A. Kuznetzova, A. N. Denisenko, V. V. Pochinko, Y. P. Chumakov, and O. S. Streltzova, “In vivo endoscopic OCT imaging of precancer and cancer states of human mucosa,” Opt. Exp., vol. 1, pp. 432–440, 1997. [6] G. J. Tearny, M. E. Brezinski, B. E. Bouma, S. A. Boppart, C. Pitris, J. F. Southern, and J. G. Fujimoto, “In vivo endoscopic optical biopsy with optical coherence tomography,” Science, vol. 276, pp. 2037–2039, 1997. [7] J. J. Armstrong, M. S. Leigh, I. D. Walton, A. V. Zvyagin, S. A. Alexandrov, S. Schwer, D. D. Sampson, D. R. Hillman, and P. R. Eastwood, “In vivo size and shape measurement of the human upper airway using endoscopic longrange optical coherence tomography,” Opt. Exp., vol. 11, pp. 1817–1826, 2003. [8] J. J. Armstrong, M. S. Leigh, D. D. Sampson, J. H. Walsh, D. R. Hillman, and P. R. Eastwood, “Quantitative upper airway imaging with anatomic optical coherence tomography,” Amer. J. Respir. Crit. Care Med., vol. 173, pp. 226–233, 2006. [9] N. V. Iftimia, B. E. Bouma, J. F. de Boer, B. H. Park, B. Cense, and G. J. Tearney, “Adaptive ranging for optical coherence tomography,” Opt. Exp., vol. 12, pp. 4025–4034, 2004. [10] G. J. Tearney, B. E. Bouma, and J. G. Fujimoto, “High-speed phase- and group-delay scanning with a grating-based phase control delay line,” Opt. Lett., vol. 22, pp. 1811–1813, 1997. [11] A. V. Zvyagin, E. D. J. Smith, and D. D. Sampson, “Delay and dispersion characteristics of a frequency-domain optical delay line for scanning interferometry,” J. Opt. Soc. Amer. A, vol. 20, pp. 333–334, 2003. [12] O. Skatvedt, “Continuous pressure measurements in the pharynx and esophagus during sleep in patients with obstructive sleep apnea syndrome,” Laryngoscope, vol. 102, pp. 1275–1280, 1992. [13] G. J. Klein, B. W. Reutter, M. H. Ho, J. H. Reed, and R. H. Huesman, “Real-time system for respiratory-cardiac gating in positron tomography,” IEEE Trans. Nuclear Sci., vol. 45, no. 4, pt. 2, pp. 2139–2143, Aug. 1998. [14] C. Xu and J. L. Prince, “Snakes, shapes, and gradient vector flow,” IEEE Trans. Image Process., vol. 7, no. 3, pp. 359–369, Mar. 1998.
Matthew S. Leigh received the B.Tech. degree in optoelectronics from the University of Auckland, Auckland, New Zealand, in 2002. Currently, he is working towards the Ph.D. degree at the University of Western Australia, Crawley, W.A., Australia From 2002 to 2003, he was a Research Associate at the Optical+Biomedical Engineering Laboratory (OBEL), University of Western Australia. Currently, he is researching and developing endoscopic optical imaging techniques at the University of Western Australia.
1446
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 4, APRIL 2008
Julian J. Armstrong received the B.Sc. degree in physics and the B.E. and Ph.D. degrees in electronic engineering from the University of Western Australia, Crawley, W.A., Australia, in 1999 and 2007, respectively. His Ph.D. dissertation was entitled “Anatomical optical coherence tomography in the human upper airway,” which involved the development and clinical testing of an endoscopic optical coherence tomography system. Currently, he is a Research Associate at the Optical+Biomedical Engineering Laboratory (OBEL), University of Western Australia. His research interests include biomedical photonics and the development of new applications of optical coherence tomography.
Alexandre Paduch received the M.Sc. degree in photonics from the Ecole Nationale Supérieure de Physique de Strasbourg, Strasbourg, France, in 2000. He has been working on various projects in the field of photonics metrology. He worked as an R&D Engineer at Agilent Technologies, Böblingen, Germany, in the field of optical telecommunication components test (2000–2003). From 2003 to 2005, he was an Optical Test Engineer at Philips LCOS Microdisplay Systems. In 2005, he joined for one year the Optical+Biomedical Engineering Laboratory (OBEL), University of Western Australia (UWA), Crawley, W.A., Australia, as a Research Associate, where he developed an optical coherence tomography measurement systems profiling hollow organs in vivo. Since 2006, he has been an Optical R&D Engineer, in the field of optical surveying at Leica Geosystems, Heerbrugg, Switzerland.
Jennifer H. Walsh received the B.Sc. degree, the M.Sc. degree in science, and the Ph.D. degree in science from the University of Western Australia, Crawley, W.A., Australia, in 1995, 1999, and 2004, respectively. Her M.S. and Ph.D. degrees focussed on the effect of chronic and acute exercise on the thermoregulatory and blood vessel functions of cardiac patients. In 2004, she was a Sleep Scientist at the West Australian Sleep Disorders Research Institute (WASDRI), Nedlands, W.A., Australia, where she is currently a Postdoctoral Fellow, investigating the mechanisms, treatment, and diagnosis of upper airway collapse in obstructive sleep apnea. Her research focus is on the anatomic risk factors in obstructive sleep apnea.
David R. Hillman received the M.B. and B.S. degrees from the University of Western Australia, Crawley, W.A., Australia, in 1974, and the Anesthesiology Fellowship in 1981. Currently, he is Head of the Department of Pulmonary Physiology, Sir Charles Gairdner Hospital, Perth, W.A., Australia, Director of the West Australian Sleep Disorders Research Institute (WASDRI), Nedlands, W.A., Australia, and Clinical Professor at the Faculty of Medicine and Dentistry, University of Western Australia. His current research interests are centered on upper airway physiology during sleep and anesthesia.
Peter R. Eastwood received the B.S. degree in health science from Lock Haven University, Lock Haven, PA, in 1986, and the Ph.D. degree in respiratory physiology from the University of Western Australia, Crawley, W.A., Australia, in 1995. From 1995 to 1997, he undertook Postdoctoral research studies with Prof. J. Dempsey at the University of Wisconsin, Madison. Currently, he is a National Health and Medical Research Council (Australia) R. Douglas Wright Fellow at Sir Charles Gairdner Hospital, Nedlands, W.A., Australia, an Associate Professor at the University of Western Australia, and an Adjunct Professor at Curtin University of Technology, Perth, W.A., Australia. His research interests include investigations into the pathophysiology of upper airway dysfunction in individuals with sleep disordered breathing.
David D. Sampson received the B.Sc. degree with first class honors in chemical physics from the University of Western Australia, Perth, W.A., Australia and the Ph.D. degree in physics from The University of Kent, Canterbury, U.K., in 1992. He has spent periods in industry and as a tenure-track academic at The University of Kent and at The University of Melbourne, Melbourne, Australia, conducting research in optical communications and photonics. Since 1996, he has been at the University of Western Australia, where he founded and currently heads the Optical+Biomedical Engineering Laboratory (OBEL), School of Electrical, Electronic and Computer Engineering. He is also Associate Dean (Research) in the Faculty of Engineering, Computing and Mathematics. His research interests are in biomedical optics/biophotonics, more particularly in noninvasive diagnostics, holographic imaging, in vivo microscopy, optical coherence tomography and its medical applications, and tissue optics.
