Fourier-domain optical coherence tomography: recent advances toward clinical utility

Abstract

With the advent of Fourier-domain techniques, optical coherence tomography (OCT) has advanced from high-resolution ‘point’ imaging over small fields-of-view to comprehensive microscopic imaging over three-dimensional volumes that are comparable to the dimensions of luminal internal organs. This advance has required the development of new lasers, improved spectrometers, minimally invasive catheters and endoscopes, and novel optical and signal processing strategies. In recent cardiovascular, ophthalmic, and gastrointestinal clinical studies, the capabilities of Fourier-domain OCT have enabled a new paradigm for diagnostic screening of large tissue areas, which addresses the shortcomings of existing technologies and focal biopsy.

Introduction

Optical coherence tomography (OCT) was originally developed for imaging the human retina as a means for diagnosing pathologic changes at an early phase [1,2]. The clinical adoption of OCT in ophthalmology is now well established and commercial systems are in routine use for research and clinical practice. In comparison, the translation of OCT for other clinical applications has lagged; although the first demonstration of catheter-based OCT for imaging internal organs was published in 1997 [3], the technology has not been adopted for routine practice. Recent advances, however, promise to change this status.

In the field of cardiology, there is a pressing need for improved characterization of coronary pathology in order to better understand factors associated with heart attack and to develop and guide the deployment of better therapeutic strategies. The resolution and image contrast of OCT are attractive for this application and suitable catheters have been developed that provide minimally invasive access to the main coronary arteries. The challenge, however, has been that blood is nearly opaque to light and strategies to either occlude flow or to displace blood through the injection of a transparent liquid such as saline can be applied for only a few seconds without risk of ischemia. Early clinical pilot studies with OCT demonstrated excellent image quality, but since the acquisition rate was limited to a few frames per second, only discrete locations within the coronary arteries could be visualized. The recent advance of new Fourier-domain strategies for OCT has overcome this limitation by increasing imaging speeds to more than 100 frames per second. This increase in the imaging speed allows long coronary artery segments to be imaged following a brief, non-occlusive injection of saline through the guiding catheter. As commercial Fourier-domain OCT systems become available, it is likely that the forecast for adoption of this technology in cardiology will dramatically change.

The benefits of Fourier-domain OCT may have a similar significance for endoscopic and laparoscopic applications in screening and surveillance for early neoplastic changes. In these applications, the rationale for high-speed acquisition is to enable wide-field imaging of large luminal surface areas in the exploration for early stage, focal disease. Although excisional biopsy can be safe and effective for focal diagnosis, the ability to survey large tissue volumes noninvasively could revolutionize diagnostic procedures. Current studies are, for example, investigating Fourier-domain OCT for diagnostic imaging of the entire distal esophagus with a resolution approaching that of histopathology. The following review will focus on the recent technical advances that have enabled Fourier-domain techniques and will highlight clinical applications for which this technology may have the most significant impact.

Time-domain OCT

Interferometric methods for length measurements have been used pervasively in the physical sciences for over a century. By measuring the cross-correlation between an electric field reflected from a target and a coherent replica of the original field, distances can be readily measured with a precision well below a single wavelength. Measurements can be performed using temporal delay or wavelength as the variable coordinate and many interferometer topologies have been exploited [4,5]. In the early 1990s, interferometry was investigated for length measurements in the human eye [6], exploiting the ability of this approach to determine the range to multiple reflection sites along a single optical axis. Not long after this, transverse scanning of the optical beam was implemented and the interferometric signal strength was converted to a color-scale or gray-scale to provide a cross-sectional image [1,2]. This approach has become recognized as time-domain optical coherence tomography (OCT), conventionally referring to an interferometric imaging system in which the reference delay is scanned.

In many respects, ophthalmic imaging is unique in comparison with other biological applications; the anterior portion of the eye and the vitreous are highly transparent with little optical scattering, the eye can be stabilized effectively, and the eye can be directly accessed with optical instrumentation. Recognizing the potential of OCT for imaging through minimally invasive catheters and endoscopes, however, research in the mid-1990s was directed to resolve several deficiencies of the prototype ophthalmic OCT systems. Improved resolution, approaching 1 µm, was achieved through the use of ultrafast mode locked lasers [7–9]. Imaging speed, required to overcome motion artifacts arising from respiration, cardiovascular function, and peristalsis, was increased 10-fold through the use of a phase-controlled rapid scanning delay line [10]. Since the dominant mechanism limiting the depth of penetration of the conventional near infrared (~800 nm) ophthalmic systems was optical scattering rather than absorption, the use of infrared light near 1300 nm improved imaging depths to a few mm in most biological tissues. [11,12] Although this is shallow in comparison with other clinical imaging modalities, such as ultrasound or diffuse-light techniques, it is sufficient for many biological and clinical applications, provided that minimally invasive probes can be utilized to deliver light to [13] and collect light from the tissue of interest. Appropriate probes were developed for intravascular [14], laparoscopic [15], and endoscopic delivery [13,16]. Combining these advances, small animal internal organ imaging was demonstrated for the first time in vivo in 1997 [3].

Shortly after the first demonstration of imaging in vivo with OCT, portable systems were developed and human clinical pilot studies were conducted to evaluate imaging in the upper and lower gastrointestinal tract [16–18], the common bile duct [19], the lung and airways [20], the cervix [21], the bladder [22,23], the larynx [24], and the coronary arteries [25–31]. These early studies showed that the resolution and contrast provided by OCT were significantly better than could be provided by conventional imaging techniques such as ultrasound and that a spectrum of pathologic states could be identified. The widespread adoption of OCT into clinical practice, however, did not follow. One significant barrier inhibiting adoption was the focal nature of OCT imaging; the catheter or endoscope probe was placed at discrete locations and cross-sectional images were obtained. The resulting diagnostic information could never therefore substantially exceed that provided by excisional biopsy.

In 2003, nearly simultaneous reports were published that demonstrated theoretically [32•,33•,34•] and experimentally [35•,36,37•] that shifting from the time-domain of conventional OCT to the Fourier-domain, in which the electric field cross-correlation is sampled as a function of wavenumber, provides several orders of magnitude improvement in detection sensitivity. The significance of this finding for clinical applications was enormous since it enabled dramatically faster imaging speeds and therefore allowed imaging over very large fields of view. [38••]

Fourier-domain OCT

Even at the inception of OCT, it was well known that interferometric ranging could be performed using wavenumber as the variable coordinate. In practice, this could be achieved either through the use of a wavelength-swept light source and a standard photodiode receiver or with a broadband light source and a spectrometer. In both cases, the acquired signal is the integrated spectrum of the light source, superimposed by fringes whose frequency encodes the pathlength imbalance of the interferometer. Through Fourier transformation, the sample reflectance as a function of depth is obtained. The terminology associated with these different configurations has not been standardized. Typically, the original coherence-domain OCT is now referred to as time-domain OCT. The configuration using a spectrometer (Figure 1a) has been referred to as spectral radar [39] and spectral-domain OCT; and the configuration using the wavelength-swept laser (Figure 1b) has been referred to as frequency-domain OCT, optical frequency domain imaging (OFDI), and swept-source OCT.

Figure 1.

Schematic layout of typical polarization-diverse Fourier-domain optical coherence tomography systems. (a) Spectral-domain systems rely on a broadband lightsource and a spectrometer as a detector. (b) In frequency-domain systems, the light source is a wavelength-swept laser and the receiver comprises single-element photodiodes. A frequency-shifter is typically used to resolve otherwise degenerate positive and negative depths relative to the reference arm pathlength. PBS: polarization beam splitter; BBS: broadband beam splitter.

The detection sensitivity in Fourier-domain OCT is enhanced because the receiver registers reflected light from all depth-points in the sample simultaneously over the duration of one complete axial profile. In time-domain OCT, the short temporal coherence of the light source is exploited to reject the reflected light from all but one depth point and to successively read reflection as a function of depth until a complete axial profile is collected. As a result, the enhancement of signal-to-noise ratio in Fourier-domain systems is given roughly by the number of axial points in an image. In either configuration, the theoretical enhancement of sensitivity can readily be made to be in the order of 100–1000. In practice, many factors can prevent realization of this improvement. In particular, backscattering arising in the optical fiber path of the interferometer sample arm gives rise to an elevated noise floor, but systems with enhancements of between 50 and 100 have been demonstrated.

Increased detection sensitivity could be utilized to reduce the optical power incident upon the sample while still achieving image penetration to the multiple-scattering limit. Although this advantage may be useful in ophthalmology, most research groups have exploited the increased sensitivity of Fourier-domain systems to achieve higher imaging speed so that larger areas of the retina can be imaged without motion artifacts. In spectral-domain OCT, this requires a high-speed linear detector array, comprising more than 1000 individual elements. Silicon-base line-scan cameras were the first to be integrated into spectral-domain systems and provided readout rates of approximately 40 kHz. In comparison, InGaAs cameras, appropriate for use with 1.3 µm light sources were less mature, providing fewer pixels and slower readout rates. In addition to the line-scan camera, spectral-domain OCT system required the development of appropriate spectrometers [35•,40•] but leveraged from existing broadband light sources that were already developed for time-domain OCT.

The practical realization of high-speed frequency-domain OCT systems required the development of a new wavelength-swept laser with a narrow instantaneous linewidth, a broad tuning range, a linear sweep, and a high average power. Previous tunable lasers provided insufficient spectral range, sweep repetition rate, and power. The first demonstration of high-speed imaging relied on a novel laser configuration comprising a fiber optic ring, a semiconductor optical amplifier, and a wavelength filter constructed using a polygon scanner [41]. Subsequent improvements using the same general design yielded 115 kHz repetition rate scanning [42] and wavelength ranges extending over 145 nm centered at 1.3 µm [43]. At high-speed operation, the spectral sweep rate of the filter becomes large with respect to the resonator round-trip delay. Although the most straightforward solution to this limitation would be to miniaturize the cavity, another elegant solution is to synchronize the round trip time of the resonator with the filter [44]. In this approach, recently termed Fourier-domain mode locking [45], a long resonator is used so that on each round trip through the cavity, individual wavelengths return to the filter as the filter returns to the matching wavelength during the successive scan. Repetition rates as high as 370 kHz have been demonstrated in this way [46••]. Presently, the maximum imaging speed that can be achieved with frequency-domain OCT is limited by digital data transfer and storage.

One technical challenge to Fourier-domain approaches is the degeneracy between positive and negative depths in the sample; only the magnitude of depth is mapped to fringe frequency. A simple solution, applicable to frequency-domain systems, is to apply a frequency offset, through the use of an acousto-optic modulator, to shift the signal frequency corresponding to zero relative delay to an RF carrier [47]. This approach yields a doubling of the effective ranging depth. Alternatively, the intensity and quadrature of the Fourier-domain signal can be acquired to overcome the depth degeneracy [48–54].

With the availability of robust and portable Fourier-domain systems, clinical studies that exploit wide-field microscopic imaging have commenced. Using a frequency-domain system operating at 40 kHz A-line rate, endoscopic imaging of the entire distal esophagus in human subjects has been demonstrated (Figure 3). [55••] In contrast to previous endoscopic OCT studies in which diagnostic imaging was restricted to a similar field-of-view provided by excisional biopsy, the new generation systems are able to provide diagnostic information not accessible by biopsy. In ophthalmology, the high speed of Fourier-domain OCT has permitted comprehensive mapping of the microscopic structure of the retina. [56–58,59••] In the field of cardiology where intravascular OCT has been frustrated by the opacity of blood, Fourier-domain systems have yielded significant advances. Using only intermittent injection of transparent fluid to displace blood from coronary arteries for a few seconds, volumetric images have been obtained for entire coronaries (Figure 2). [38••,60••] This advance may provide dramatic improvements in understanding coronary atherosclerosis and response to intravascular interventions such as angioplasty and stenting.

Figure 3.

OFDI images of human Barrett’s esophagus, obtained in vivo. (A) Videoendoscope image shows islands of healthy squamous mucosa intermixed within regions of specialized intestinal metaplasia (SIM). (B) Circular transverse cross-sectional OFDI image with a layered appearance, and in some regions, an irregular surface and intraepithelial glands satisfying the OFDI diagnostic criterion of SIM. The 6 cm longitudinal segment was obtained in approximately 2 min at an acquisition rate of 9.8 frames per second (frame size: 2048 × 4096) and a pullback speed of 0.5 mm/s. (C) Expanded portion of (B) demonstrates surface irregularities (black arrowheads) and glands within the epithelium (red arrowheads). (D) Histopathologic image of the biopsy taken from the involved mucosa shows specialized and non-specialized columnar epithelium, consistent with the OFDI diagnosis of SIM. Scale bars and tick marks represent 1 mm.

Figure 2.

Volumetric OFDI imaging of the human LAD coronary artery, obtained in vivo. (A) Fly-through rendering view (proximal-distal) of the OFDI dataset, acquired during a single purge of Lactated Ringers solution (3 ml/s), an imaging catheter pullback rate of 2.0 cm/s and at an image acquisition rate of 100 frames per second. The fly-through depicts a yellow, elevated lipid-rich lesion with scattered macrophages (green). (B) OFDI cross-sectional image obtained at the location of the white arrowhead in (A) demonstrates OFDI evidence of a thin-capped fibroatheroma, a lipid pool (L), a thin cap (black arrow), and a dense band of macrophages at the cap-lipid pool interface. A thin flap of tissue (black arrowhead) can be seen over the cap. (C) Fly-through view (proximal to distal) shows a calcified lesion (black arrow) beneath a newly deployed stent. (D) OFDI cross-sectional image obtained at the location of the black arrow in (C) demonstrates a large calcific nodule (Ca) from 11–5 o’clock causing significant stent distortion, highlighted by the shorter inter-strut distances of the overlying stent compared to the opposing vessel wall. Color scale for (A) and (C): red—artery wall; green—macrophages; yellow—lipid pool; blue—stent. Tick marks, 1 mm. (*) Denotes guide wire artifact.

Conclusions

The recent development of high-speed imaging systems based on OCT principles will undoubtedly change the landscape of clinical implementation and adoption. While preserving the resolution and contrast of time-domain OCT, Fourier-domain systems enable comprehensive imaging over large fields of view and address the primary limitation of early OCT technology.

Although InGaAs linescan camera technology is presently advancing to rival the capabilities of the more mature silicon-based detectors, it is likely that the present division of applications between spectral-domain systems and frequency-domain systems will continue for some time. In ophthalmic applications, the preferred wavelength for imaging is near 800 nm owing to the relatively low absorption at this wavelength in the anterior eye and vitreous. Since the development of rapidly swept lasers at 800 nm is limited by the lack of appropriate semiconductor sources and since relatively inexpensive silicon-based cameras are already available that support imaging speeds approaching 100 frames per second, it is likely that spectral-domain OCT will remain dominant in ophthalmology. For catheter-based and endoscope-based imaging where low-loss fibers are used to deliver the light directly to the target organs, 1300 nm is preferable and results in greater imaging penetration within tissue. Since wavelength swept laser technology is now readily available for this spectral region and since the continued advancement of new high-speed cameras will require significant investment, frequency-domain OCT appears to be the most effective strategy. Additionally, frequency-domain systems are less prone to signal fading associated with optical phase variations in the optical fibers of catheters and endoscopes. Since appropriate technical strategies for Doppler and polarization-sensitive imaging have been developed for both the approaches, vasculature and birefringence imaging applications may not be a primary consideration in selecting between the two platforms.

Perhaps the most important area of continued technical development for the new generation OCT systems will be in signal and image processing. Current systems are capable of producing data rates approaching 1 GB/s. New strategies will be required for fast processing that includes interpolation and Fourier transformation. Dedicated digital signal processing such as field programmable gate array processing may yield effective strategies for computation and reduction of data volume before archiving. Additionally, new algorithms will be required for interpreting images for diagnosis or for preselecting portions of datasets for review by human experts.

Future work will also undoubtedly lead to the integration of techniques for expanding contrast and molecular specificity to Fourier-domain OCT. Many of the methods that were developed for time-domain OCT, including Doppler flow detection [61,62], birefringence characterization [63–65], and biochemical contrast [66,67•,68–72] can be directly translated into the new platforms. Indeed, methods for flow visualization [73–76] and polarization sensitivity [73,77–81] have already been demonstrated in Fourier-domain systems.

Although the path from the first OCT prototypes to the present capabilities has been long, the ground work to support adoption of the new Fourier-domain systems has heightened awareness in the biological and clinical communities and may be a factor in rapid commercialization. At present, there are more than a dozen companies with Fourier-domain systems either recently available or on the near-term horizon and prospects for wide spread application are excellent.