Functional optical coherence tomography: principles and progress

Development of Doppler OCT in the laboratory has been aided by a relatively low need for additional instrumentation and few regulatory barriers, since only adaptions to signal processing methods are required to implement the approach. While some work has been done in humans and tissue phantoms, this approach has been primarily used in small animal imaging of zebrafish, tadpoles, mice, rats, birds, and other animals. For example, Yang et al used DOCT to image the cardiac dynamics of Xenopus laevis tadpole (figure 1) (Yang et al 2003a, 2003b). Larina et al used swept source (SS)-DOCT to image and analyze heartbeat and blood flow in live rat embryos with resolution at the single cell level (Larina et al 2009). Peterson et al used 4D OCT with Doppler, i.e. three-dimensional (3D) OCT images taken over time, to measure the shear stress in the developing hearts of quail embryos (Peterson et al 2012).

Zoom In Zoom Out Reset image size

The generation of DOCT images requires additional signal processing, but as noted, the system hardware may require little or no modification. The processing of Doppler signals is different for time domain OCT versus Fourier domain OCT and significant additional computer resources may be needed, particularly if the flow information is desired in real time.

In time domain OCT, the interference signal is already centered on a fixed Doppler frequency due to the moving mirror in the reference arm. Coherent demodulation with a lock-in amplifier set to this fixed frequency is used to detect the interference fringes generated by light scattered from the sample. The magnitude of this signal is proportional to the number of photons scattered from a given depth and provides the signal in the OCT image. Any motion in the sample can add or subtract from the signal content at this fixed frequency. To obtain the Doppler shifted signal, overlapped short-time Fourier transforms at a given reference arm position are used. The average velocity in the sample at a given depth is then given by $v=({{f}_{\text{s}}}{{\lambda}_{o}})/(4\pi n\text{cos}\theta )$ where ${{f}_{\text{s}}}$ is the frequency shift from the reference arm Doppler frequency, ${{\lambda}_{o}}$ is the center wavelength of the source, $n~$ is the local index of refraction of the sample and $\theta$ is the angle of the OCT beam relative to the motion of the particle within the sample. The range of speeds that can be measured is dictated by the bandwidth of the lock-in amplifier used for demodulation while the resolution is determined by the number of points in the short-time Fourier transform.

In Fourier domain OCT (FD-OCT), data are acquired from the whole sample depth simultaneously with a fixed pathlength in the reference arm so the approach for making Doppler measurements differs. Here two or more axial scans (A-scans) are acquired at a given lateral location and the phase difference between successive measurements is calculated. This phase variation is proportional to the speed of the target within the sample, $v$, with a relationship given by $v=({{\lambda}_{o}}\Delta \phi )/(4\pi nT\text{cos}\theta )~$ where $\Delta\phi ~$ is the phase shift between A-scans, $T$ is the time between A-scans, ${{\lambda}_{o}}$ is the center wavelength of the source, $n~$ is the index of refraction of the sample, and $\theta$ is the angle of the OCT beam relative to the motion of the particle within the sample. The signal phase is calculated from the Fourier transform of the frequency resolved intensity data. Since the intensity measurements are real valued, a Hilbert Transform is used to obtain the phase value. The Hilbert Transform consists of a Fourier Transform, where the negative frequency values are set to zero, followed by an inverse Fourier Transform. The final complex Fourier Transform will produce both the scattered intensity as a function of depth in the sample and the phase at each depth point. By comparing subsequent A-scans at the same point, the rate of change of the phase can be calculated.

Commercial systems with DOCT are available for research and clinical applications. Thorlabs (Newton, New Jersey) offers Doppler OCT as an option on both spectral domain (SD)-OCT systems operating at 930 nm and SS-OCT systems operating at 1300 nm. In September 2014, Optovue, Inc. (Fremont, California) introduced the Angiovue OCT angiography system for 3D visualization of the retinal vasculature. Of note, Angiovue is not yet approved for sale in the United States.

As shown by Izatt et al, the range of speeds that can be measured in Fourier domain OCT systems is determined by the method by which A-scans are acquired (Hendargo et al 2011). For both SD-OCT and SS-OCT the slowest speed that can be measured is set by the phase noise in the system, which is given by ${{v}_{\text{min}}}={{\lambda}_{o}}/(2{{\pi}^{2}}nT\text{cos}\theta 1/\sqrt{\text{SNR}})$ where SNR is the shot noise-limited SNR of the imaging system. The upper limit on detection speed is set by the integration time for a single pixel in the system, which varies for SD and SS system. For SD-OCT, this is given by ${{v}_{\text{max}}}={{\lambda}_{o}}/(4nDT\text{cos}\theta )$ where $D$ is the detector duty cycle, and for SS-OCT, this is given by ${{v}_{\text{max}}}=(M{{\lambda}_{o}})/(4nDT\text{cos}\theta )$, where $M$ is the number of acquisitions within a single frequency sweep of the laser. Thus, SS-OCT can measure a larger range of speeds in the sample for a given A-scan rate of the system.

Another limitation in Doppler OCT is that time varying signals generated by bulk motion, as are typically seen in biological samples, can confuse the true Doppler signal. The most common technique to minimize the impact is to calculate Doppler shifts for the entire A-scan and look for common mode features. The common features are then assumed to be a result of bulk sample motion and subtracted to yield the Doppler signals of interest.

In all modes of Doppler imaging, accurate measurement of velocity requires knowledge of the angle $\theta$ between the OCT beam and the direction of the velocity in the sample. For blood vessels and capillary phantoms, it may be possible to measure this angle by acquiring a 3D OCT image and then identifying the location of the vessel or capillary within the volume image either manually or with automatic segmentation approaches. Huang et al showed that the location and angle of retinal vessels could be identified by using two concentric circular scans around the optic disk. In this approach, the relative position of a blood vessel in the two OCT cross sections is used to calculate the Doppler angle between the beam and blood vessel (Wang et al 2011).

Several variations on DOCT have been developed. Wang et al developed a technique referred to as optical angiography (OAG), which uses an SD-OCT system with a modulation in the reference arm pathlength (Wang et al 2007). In this implementation, the spectrometer operated at 10 000 lines per second, and the reference arm was modulated at the same frequency as the 2D (B-scan) rate (10 Hz) with a path length modulation magnitude of 50 microns. Motion in the sample, which produced a Doppler frequency greater than the Doppler frequency due to the reference arm modulation of 1.25 kHz, showed up in the complex conjugate part of the Fourier domain OCT image while the standard OCT image remained unchanged. Using this method, blood perfusion in a living adult mouse was imaged with higher signal to noise than phase resolved DOCT.

In another approach, Fingler et al used phase variance as a contrast mechanism to distinguish moving versus fixed scatterers in both an Intralipid phantom and live zebrafish (Fingler et al 2007). This system used SD-OCT with a center wavelength of 840 nm and a 25 kHz A-scan rate. This study showed that phase variance has a lower noise threshold compared to Doppler OCT and therefore can image motion without the angle dependence of Doppler, enabling detection of motion that is nearly perpendicular to the angle of incidence. In a similar approach, Mariampillai et al used a Fourier domain mode locked laser centered at 1310 nm and operating at 43 to 67 kHz, to demonstrate speckle variance OCT (Mariampillai et al 2008). This system provided high contrast between blood vessels and surrounding tissue, but did not provide any velocity information. However, there is significant utility in this approach as it could detect conditions such as vascular shutdown and transient vessel occlusion in near real time. Similar to the phase variance technique, the speckle variance was much more sensitive than Doppler detection for angles of incidence approaching 90 degrees.

Variance processing methods in FD-OCT have enabled depth-resolved visualization of capillary beds in the retina by acquiring A-scan data in the 100 kHz regime. However, acquisition time lasts several seconds, during which subject movement will alter visualization of the capillary beds. To address this limitation, Hendargo et al developed a technique to eliminate motion artifacts in speckle variance FD-OCT by creating a composite image of the retinal vasculature from multiple, sequentially-acquired volumes of data (figure 2) (Hendargo et al 2013).

Zoom In Zoom Out Reset image size

DOCT has shown potential for multiple clinical uses in ex vivo or in vivo human studies by observing modulation in blood flow in the retina, GI tract, and kidneys. In ophthalmology, OCT is used to render in vivo cross sectional views of the retina to diagnose pathologies such as macular holes, macular pucker, and vitreo-macular traction. DOCT can also potentially be applied to monitor progression of ocular pathologies that cause changes in blood flow dynamics, such as diabetic retinopathy and glaucoma. Yazdanfar et al first demonstrated in vivo color DOCT in the human retina using a time domain system (Yazdanfar et al 2003). This system was used to image Doppler shifts up to 3.3 kHz in artery-vein pairs with an axial resolution of approximately 20 μm and an acquisition time of less than 60 s. White et al demonstrated the first spectral domain OCT imaging of an in vivo human retina with an A-scan rate of 29 kHz (White et al 2003). The instrument was capable of detecting Doppler shifts between 24 Hz and 14.66 kHz, providing a dynamic range of over 600. By acquiring a sequence of 95 B-scans, the integrated flow of an artery-vein pair was measured over 3.28 s revealing a pulse rate of 73 beats per minute.

In the GI tract, DOCT can be applied to observe modulated blood flow due to certain pathologic processes such as esophageal varices and neoplasms. Yang et al developed an endoscopic DOCT system and imaged both normal and pathologic tissues of 22 patients in a feasibility study (Yang et al 2005). Structural and microcirculatory patterns differed from site to site with observed differences between normal and diseased tissue. This endoscopic system uses a 2 mm probe that is compatible with the auxiliary channel of most endoscopes and achieved a velocity resolution of approximately 100 μm s⁻¹, which is 10 to 100 times better than can be measured with Doppler endoscopic ultrasound.

In solid organs, DOCT has been used to study microcirculation in kidneys. Blood flow was initially measured in vivo in rat kidneys, and the effects of mannitol and angiotensin II were observed (Wierwille et al 2011). The instrument provided lower and upper velocity detection limits of 0.01 and 3.9 mm s⁻¹, respectively. The same group then used DOCT to diagnose acute tubular necrosis in human donor kidneys. Twenty-eight patients were enrolled in the study, and each transplanted kidney was imaged prior to transplant (ex vivo), and then again following transplant and reperfusion (in vivo). Tubule size/shape and density/uniformity were grouped into categories of poor, moderate and good. The kidney with poorest post-transplant function did not show open tubules, and the patient suffered delayed graft function (figure 3) (Andrews and Chen 2014). DOCT has also been explored in other anatomical sites with groups having measured blood flow in skin and lips in healthy human subjects (Zhao et al 2000, 2001, Otis et al 2004).

Zoom In Zoom Out Reset image size

Although DOCT is not actively in clinical use, it offers significant potential benefits to physicians. It has been shown that DOCT can accurately measure blood flow in the retina, GI tract, and kidneys with micron-level spatial resolution and velocity resolution up to 10 μm s⁻¹. Computational methods to remove bulk-motion artifacts and precisely determine the angle between the illumination beam and direction of flow have also been developed. These studies and recent developments suggest that DOCT could be an important clinical tool to diagnose potentially pathologic tissues and monitor responses to treatment.

For PS-OCT, the system complexity necessarily increases since the approach requires collection of OCT signals at more than one polarization. The data from each polarization channel are typically processed independently until the final stage where the two images are analyzed to provide polarization contrast. There are numerous methods for collecting multiple polarizations. In an early example, de Boer et al used a time domain OCT system with a rotating quarter waveplate in the reference arm to collect OCT signals with varying polarizations as a function of time (de Boer et al 1997). Synchronizing the detection with the rotating waveplate enabled extraction of the OCT signal of the vertical and horizontal polarizations. This system was initially demonstrated through application to imaging fresh bovine tendon.

The next version of de Boer's system used a linear polarized light source set at 45° and a polarizing beam splitter (PBS) in the collection arm to simultaneously capture two orthogonal polarizations (de Boer et al 1999). Each polarization was collected by a separate detection channel with independently associated electronics. This was still a free space system, but enabled measurement of the depth-resolved Stokes parameters in rodent muscle and skin. Saxer et al then developed this approach further by building a high-speed fiber-based time domain OCT system with a polarization modulator in the source arm (Saxer et al 2000). This system operated at 1310 nm and was used to measure birefringence of in vivo human skin. Again the detection arm used a PBS with two detectors and independent electronics for each channel.

The next advance in time domain PS-OCT was 3 D PS-OCT, first demonstrated by Pircher et al in 2004 (Pircher et al 2004a, 2004b). The center wavelength was again at 1310 nm, but used a different interferometer topology. The Mach Zehnder configuration contained two PBS detection setups, with a total of four detectors, thus providing balanced detection for each orthogonal polarization; this is now the standard architecture in time domain PS-OCT. Human skin was measured in vivo with an acquisition time of 150 s per site.

The transition to Fourier domain OCT offers increased imaging speeds due to improved SNR (Choma et al 2003), but for PS-OCT, the approach requires more complex collection schemes in place of the standard spectrometer. Gotzinger et al developed a high speed PS-OCT system using two independent spectrometers each operating at 20 000 lines per second (figure 4) (Götzinger et al 2005). This implementation used a free space interferometer with a PBS in the collection arm and fiber optic coupling after the PBS to deliver light to the spectrometers. Yamanari et al moved to a fiber-based interferometer and located the PBS inside the spectrometer (Yamanari et al 2006). The PBS was placed after the grating and focusing lens and two CCD cameras were then used to capture the orthogonal polarizations at 27 000 lines per second. This system operated at 840 nm and was used to image chicken breast muscle, a finger pad and a caries lesion in a tooth.

Zoom In Zoom Out Reset image size

In SS-OCT, the collection optics typically consists of either a single detector or two detectors in a balanced configuration so the PS implementation of SS-OCT is similar to that used in time domain. Interferometers for SS-OCT are typically fiber optic based so a fiber coupled polarizing beamsplitter may be used along with two fiber-coupled photodiodes. Polarization maintaining fibers must be used in these schemes to preserve the polarization dependence of the signal. Balanced detection may be implemented by extending this scheme to include a circulator and two fiber coupled polarizing beamsplitters. This scheme requires four detectors and the level of suppression of the opposite polarization will be limited by the extinction ratios of the two polarizing beamsplitters.

Numerous groups have developed polarization sensitive OCT systems based on swept source lasers. Some examples include Yamanari et al, who developed a fiber-based polarization-sensitive SS-OCT system using continuous source polarization modulation (Yamanari et al 2008). This system used a swept source from Santec Laser (Japan), centered at 1310 nm and operating at a sweep rate of 20 kHz. This system was demonstrated by imaging chicken breast muscle ex vivo and corneal angle from in vivo human anterior eye segment.

Oh et al developed a parallel acquisition system by generating two polarizations in the source arm and then modulating them at two different frequencies so that a single detector set can simultaneously capture information from both polarizations using frequency resolved detection (Oh et al 2008). Using a swept source laser with a center wavelength of 1305 nm and an A-scan rate of 50 kHz, this system was used to generate intensity and phase retardation images of chick muscle ex vivo, human finger in vivo, and human coronary artery ex vivo.

Baumann et al developed a high-speed polarization sensitive system with a passive polarization delay unit (Baumann et al 2012). This delay unit was placed in the sample arm and generated a 1.8 mm pathlength difference between the two polarizations. The detector scheme for this system was a standard four-detector setup with balanced detection for each polarization. A 1040 nm swept laser from Axsun Technologies (Billerica, MA) with a 100 kHz repetition rate was used with a semiconductor optical amplifier (SOA) from Inphenix Inc. (Livermore, California). The offset in polarizations resulted in two images in a B-scan with a depth offset between them of 1.8 millimeters. 3D retinal imaging was performed on volunteers with volumes of 500   ×   100 A-scans acquired in ~0.55 s.

The above section reviews only a few of the many PS-OCT systems developed in academic labs around the world. However, similar to Doppler OCT, there has been significant research into PS-OCT systems, but few commercial offerings. To date, the only commercially available PS-OCT is a module from Thorlabs that can be added to their SS-OCT systems. This requires a hardware change, but is available as an upgrade to any already deployed swept source systems.

Structural changes associated with certain disease processes cause changes in birefringence that can be measured with PS-OCT. PS-OCT has been applied to study birefringence in the eye, skin, esophagus, breast tissue, cartilage, and teeth. For example, PS-OCT was used to image hard exudates in eyes of 16 patients with diabetic macular edema with an axial resolution of less than 5 µm (Lammer et al 2014). Hard exudates in the eye indicate an impairment of the blood-retinal barrier, leading to retinal swelling and edema. PS-OCT enabled 3D imaging of hard exudates throughout retinal layers by measuring the uniformity of polarization across each image with a sliding evaluation window. Automated detection of hard exudates with PS-OCT showed reasonable correspondence with manual grading of color fundus photographs, which is the current gold standard for diagnosis.

PS-OCT has also been used to measure the birefringence of collagen, which transitions from a rod-like alpha helix to a random coil conformation after heating. In normal human skin, which contains well-ordered collagen, the birefringent nature of the tissue can be observed with PS-OCT. In contrast, the ordered collagen structure is disrupted in burned tissue resulting in decreased birefringence (figure 5). The mean difference in phase retardation between orthogonal polarization states was 0.401 and 0.249°/µm for normal and burned skin, respectively. This difference was statistically significant and over an order of magnitude larger than the measurement accuracy (0.003°/µm) (Pierce et al 2004).

Zoom In Zoom Out Reset image size

Other prospective applications of PS-OCT deserve mention. For example, PS-OCT has shown potential in assessing changes in cartilage organization. In one study, PS-OCT images were graded on a ternary scale by a blinded investigator and compared to histology from the same location. There were no significant differences in the paired scores derived from PS-OCT and histology (Drexler et al 2001). Another application has been found in the area of dentistry. Because dental enamel contains birefringent hydroxyapatite crystals, PS-OCT can be used to image changes in birefringence caused by dental caries with an initial in vitro demonstration having been completed (Baumgartner et al 2000). In summary, PS-OCT can measure birefringence in vivo in a variety of biological tissues. Recent demonstrations have achieved high spatial resolution and accuracy in determining the phase difference between orthogonal polarization states. This additional structural information can be leveraged to diagnose disease, evaluate injury, or monitor therapeutic response.

Recently, the technique of sono-elastography has been combined with OCT to take advantage of the high resolution of OCT imaging. Optical coherence elastography (OCE) was first implemented by Schmitt in 1998 using a time domain OCT system and a piezoelectric actuator to generate linear displacements of 0 to 100 microns at rates up to a few Hz (Schmitt 1998). This system was used to image mechanical properties of gelatin phantoms, ex vivo pork meat, and in vivo human skin.

Later efforts by Adie et al resulted in an OCE system with a tunable piezoelectric stack and measured the tissue response as a function of the drive frequency over the range of 25 Hz to 1 kHz (figure 6) (Adie et al 2010). This system can measure the natural vibration modes or resonances up to the maximum drive rate. This approach was used to measure the complex mechanical spectra of rat mammary tumor margins, showing a difference in resonance frequencies between the normal tissue and the tumor.

Zoom In Zoom Out Reset image size

In 2010, Liang et al built a dynamic OCE system that acquired B-scans during an applied mechanical compression (Liang et al 2010). This approach used the phase difference between adjacent A-scans to measure scatterer velocity in the axial direction. Similar to Adie (Adie et al 2010), multiple driving frequencies were used, including 20, 45, 100, and 313 Hz. For the lowest three drive frequencies, an A-scan rate of 1 kHz was used, but for the 313 Hz drive frequency, the imaging A-scan rate was increased to 10 kHz. The system was tested on a three-layer tissue phantom and then on ex vivo rat tumor. The different excitation frequencies permitted differentiation of adipose and tumor tissue.

B. Kennedy et al presented the first 3D OCE images in 2011 (Kennedy et al 2011). This approach used an SD-OCT system with an A-scan rate of 5 kHz and mechanical excitation at 125 Hz to generate 3D-OCE imaging of normal and hydrated skin in vivo. The light source was a pumped titanium-sapphire laser with a center wavelength of 800 nm and a bandwidth of 100 nm providing an axial resolution of 2.8 µm. Images were acquired from the skin on the tip of the middle finger with a preload generated by pressing the finger against the actuator window and strapping it in place. The skin was then hydrated in warm water for 30 min and reimaged. The study observed a more elastic response of the stratum corneum for the case of hydrated skin.

To overcome the depth limitation of OCT to just the superficial layers of tissue, Kennedy et al integrated an OCT system with a needle biopsy probe to image deeper within human breast tissue (figure 7) (Kennedy et al 2013). The OCT system ran at a 5 kHz A-scan rate with a center wavelength of 836 nm. A motorized stage was used to drive the needle into the tissue at a known speed, with the needle tip providing the mechanical compression for OCE imaging. The needle probe did not offer any lateral scanning, so only provided mechanical information obtained from analyzing the forward looking A-scan.

Zoom In Zoom Out Reset image size

A more comprehensive review of OCE with a particular emphasis on the signal processing employed for various implementations of OCE can be found in Sun et al (2011). This review also provides details on the use of intravascular OCE for atherosclerotic plaque detection, which is further described below.

Clinical use of OCE relies on the principle that elastic properties of tissues can change due to disease. Edema, fibrosis, and calcification are all known to change the elastic modulus of the extracellular matrix. Tumors are often stiffer than normal surrounding tissue. Current applications of elastography include assessment of atherosclerotic plaques within arteries, examination of breast or brain tissue for malignancy, and evaluation of lung tissue for disease.

Intravascular ultrasound elastography has made significant strides and is able to extract arterial radial strains with a spatial resolution of 200 μm. However, vulnerable atherosclerotic plaques have structural elements on the order of 50–200 μm, below the resolvable limit of intravascular ultrasound. Intravascular OCE has inherently higher resolution and potential to provide high resolution characterization of strain in tissues that lie 1.0–1.5 mm below the vessel surface, which is the region most susceptible to plaque disruption (Chan et al 2004). Intravascular OCT is able to resolve thin cap fibroatheromas (TCFA), which are intracoronary plaques with a large lipid-rich necrotic core covered by a thin, inflamed fibrous cap, that are believed to be responsible for acute coronary syndrome (Davies 1992, Virmani et al 2003, Chan et al 2004, Jang et al 2005, Kubo et al 2007, Fujii et al 2008, Stone et al 2011). By detecting changes in the mechanical properties due to TCFA formation, intravascular OCE may be able to predict plaques at risk for rupture.

Similar to application to intravascular plaques, OCE can identify intravascular blood clots. Oldenburg et al used rehydrated, lyophilized platelets loaded with superparamagnetic iron oxides that adhere to sites of vascular endothelial damage, in order to identify the presence of clots. Using an approach termed magneto-motive OCT, where a magnetic field is used to mechanically displace magnetic nanoparticles, elastometry of simulated clots was conducted. The measured data were analyzed to determine the fundamental resonance frequency of the clot, which showed a qualitative positive correlation to its fibrinogen content (Oldenburg et al 2012).

OCE has been evaluated for other clinical applications, such as eye, skin, breast cancer and cystic fibrosis (Chhetri et al 2010, Liang and Boppart 2010, Srivastava et al 2011, Ford et al 2014). Ford et al demonstrated that corneal edema and collagen cross-linking affected the mechanical properties of human ex vivo corneas with OCE. Local lateral versus axial displacement ratios were 0.035, 0.021, and 0.014 µm µm⁻¹ for edematous, normal, and collagen cross-linked human corneas, respectively. The differences were statistically significant and demonstrated that OCE can quantify clinically-relevant mechanical properties in the cornea (Ford et al 2014). Liang and Boppart used OCE to measure Young's moduli of in vivo human skin at different locations on the arm. They demonstrated that Young's moduli are dependent on the site, stimulation direction and frequency. For example, with a driving frequency of 50 Hz, they measured Young's moduli of 101.180, 68.678, and 24.910 kPa from the volar forearm, dorsal forearm, and palm, respectively. Different conditions, such as hydrated or dehydrated skin, were also observed to affect the Young's modulus (Liang et al 2010).

Typically, malignant breast tissue has higher stiffness than the fatty tissues that compose the breast. This difference allows for discrimination between types of breast tissue by OCE, as demonstrated by Srivastava et al (2011). Tissue samples of normal breast, fibroadenoma (benign disease), and invasive ductal carcinoma (malignant disease) were subjected to axial compressive loading, and OCT was used to track the speckle movement within the tissue. Young's modulus was estimated by measuring the slope of each stress-strain curve, for each respective tissue type. Invasive ductal carcinoma showed the highest Young's modulus, approximately four times higher than normal breast tissue, which is qualitatively consistent with other reports in the literature. Although OCE has not found widespread clinical use, the preliminary results presented above highlight the potential to diagnose tissue based upon elastic properties derived from OCE measurements.

SOCT is still in the preliminary stages of development for clinical application, with the majority of studies to date restricted to tissue phantoms and animal models. Thus far, it has been demonstrated in pre-clinical studies as a promising technology for detection of blood oxygen saturation, precancerous lesions, intravascular plaques, and burn injury.

Several groups have used SOCT to measure blood oxygen saturation by leveraging the distinct absorption features of oxy- and deoxy- hemoglobin (HbO₂ and Hb, respectively). Faber et al measured the absorption spectra of HbO₂ and Hb in diluted blood and highly scattering whole blood (Faber et al 2003, 2005). The precision of the absorption coefficients extracted from the diluted blood samples was about 0.1 mm⁻¹ and there was a strong correlation (R² = 0.70–0.86) between the differential total attenuation coefficient measured at 780 and 820 nm and the oxygen saturation of the whole blood samples. Lu et al measured blood oxygenation by analyzing A-scans at different wavelengths (Lu et al 2008). They found a qualitative relationship between the measured absorption spectra and the oxygen pressures measured with a blood gas analyzer. Liu and Kang also measured blood oxygen saturation using a spectral normalization technique to correct for artifacts induced by the spatially- and spectrally- variant point spread function of the instrument and a low-pass filter to reduce speckle noise (Liu and Kang et al 2010). The blood used in these experiments was prepared to be either fully oxygenated or fully deoxygenated and the measured spectra showed good differentiation between these two states. Yi and Li demonstrated measurements of oxygen saturation in individual blood cells using sparse distributions of cells in a gel-based phantom to overcome the resolution limitations of the STFT (Yi and Li et al 2010). They found a strong correlation (R² = 0.80–0.87) between the oxygen saturation estimated from diffuse reflectance spectroscopy and SOCT. Robles et al quantified oxygen saturation in oxygenated and deoxygenated Hb absorbing phantoms and in vivo in a mouse model using a parallel frequency domain SOCT system (Robles et al 2010, 2011). The first study used visible light to measure oxy- and deoxy- hemoglobin concentrations in solutions and found a limit of detection of 1.2 g L⁻¹ with a pathlength of 1 mm. This is less than the average hemoglobin concentration in normal tissue, which is about 1.8 g L⁻¹, but approximately 3 times less than the concentration found in cancerous tissue. The detection limit was also significantly smaller than the average concentration of hemoglobin found in whole blood (150 g L⁻¹). Studies with an in vivo mouse model demonstrated en face images obtained with the molecular imaging true-color spectroscopic optical coherence tomography (METRiCS OCT) instrument (figure 9). This image was acquired from a dorsal skin-fold window chamber model after a retro-orbital injection of sodium fluorescein (NaFS). NaFS exhibits strong absorption at shorter wavelengths and its presence is therefore indicated by a redshift in hue. Oxygen saturation (SO₂) was quantified at specific tissue locations by fitting the measured spectra with the absorption spectra of HbO₂ and Hb.

Zoom In Zoom Out Reset image size

SOCT has also been used to study carcinogenesis in the hamster cheek pouch and rat colorectal cancer models (Graf et al 2009, Robles et al 2010). Both of these studies used DW spectroscopic analysis of normal and dysplastic tissues to reveal that the average diameter of cell nuclei was significantly greater in diseased tissue. The first study found an average nuclear diameter in the basal layer of the epithelium of 4.3 µm in healthy tissue and 9.5 µm in hyperplastic or dysplastic tissues. The second study found that the average nuclear diameter in colonic rat tissue at a depth of 35 µm from the tissue surface increased from 5.15 µm to 6.5 µm after 12 weeks of treatment with azoxymethane. Significantly, the latter study was able to detect systemic changes in the tissue, even when the number of pre-cancerous lesions was small.

SOCT may also benefit cardiovascular medicine, where a noninvasive method of detecting cholesterol plaques would avoid the complications associated with invasive procedures such as coronary catheterization. Fleming et al demonstrated more accurate detection of cholesterol by applying SOCT to porcine aorta injected with mayonnaise (Fleming et al 2013). Finally, SOCT has been used to evaluate burns in vivo in a mouse model. A number of data processing and analysis methods were compared for differentiating between healthy and burned tissue and overall classification accuracy was found to be as high as 91% (Maher et al 2014). In this study, images were color-coded based upon coefficients derived from power-law fitting the spectral data at each pixel or based upon the results of a logistic regression classification model (figure 10). These images demonstrate the diagnostic utility of the approach as they qualitatively differentiate the healthy tissue (light-beige coloring) from the burned tissue (red coloring).

Zoom In Zoom Out Reset image size

Some challenges remain to be addressed before SOCT can be successfully translated to the clinic. These include continuing to develop methods used to separate scattering and absorption coefficients (Robles and Wax 2010a) and improving the accuracy of hemoglobin concentration and oxygen saturation measurements through careful selection of relevant acquisition properties (e.g. illumination wavelength, illumination power, measurement time) and processing routines. However, taken together, the studies cited above suggest that SOCT is a promising technology for assessing the health and function of biological tissue with many directions for future research.

The first molecular imaging OCT approaches used spectroscopic OCT to detect the absorption of near infrared (NIR) dyes (Xu et al 2004). This preliminary study matched the dye absorption to the spectral band of the OCT light in the 780 nm region. By tracking the centroid of the returned light spectrum using SOCT methods described above, the presence of the dye was revealed. While this approach was simple and easily implemented, the simple absorption profile can be complicated by the presence of scattering, requiring a means for separating the two (Robles and Wax et al 2010).

A later approach by Yang et al (2004) developed a spectral triangulation method to detect indocyanine green (ICG). This approach relies on the fact that most scattering profiles present monotonic dependence on wavelength. Thus, the spectrum obtained with SOCT methods can be divided into just a small number of segments to detect the contrast agent and avoid the tradeoff in resolution seen for the STFTs needed for high spectral resolution. As noted above, Robles et al also used absorption profiles to detect the presence of sodium fluorescein (Robles et al 2011). This approach was able to distinguish the contrast agent from other spectral features by obtaining high resolution spectra using the DW approach (Robles et al 2009).

A different means for discriminating the absorption of a contrast agent can be realized by using a method that actively changes the absorption cross section by optical excitation. This approach termed pump-probe OCT (Rao et al 2003), relies on using an optical excitation to cause the contrast agent to transition to a triplet state. While this is usually undesirable for fluorescence imaging, since the contrast agent will temporarily cease to emit, the changes in absorption that are associated with this transition can be detected with OCT if carefully selected. In this demonstration, the triplet absorption cross-section of methylene blue in the 830 nm region overlapped with the spectral band of the OCT system. The limitation of this approach is that the triplet state, while long lived by fluorescence lifetime standards, was sufficiently short to only allow a single OCT scan per pump-probe cycle, resulting in a low efficiency. Later studies using this approach (Yang et al 2004) extended its utility by employing different contrast agents, such as bacteriorhodopsin and phytochrome A. The contrast agents used in this study used a slightly different mechanism where the absorption shifted depending on the protein state. With one absorption state offering a significantly longer lifetime, this approach did not require high intensity light to cause the transition to the triplet state. However, the limited choice of contrast agents and the long-lived state has limited further applications.

Molecular contrast can also be generated by leveraging nonlinear optical interactions, which generate coherent photons that are coherent with the excitation field. In OCT imaging, the processes of second harmonic generation (SHG) and coherent anti-Stokes Raman spectroscopy (CARS) have both been examined as molecular imaging schemes. SHG appears in media without inversion symmetry and in biomedical imaging, it is particularly useful for imaging collagen structures. Since the SHG light is coherent with the excitation, it can be detected with OCT by generating a reference field that is also at the second harmonic frequency. This approach has been demonstrated by multiple groups (Applegate et al 2004, Jiang et al 2004, Yazdanfar et al 2004) but became less favorable as multimodality imaging that combined OCT with multiphoton microscopy took hold (Chong et al 2013). Another nonlinear process, CARS has also been incorporated into OCT imaging (Bredfeldt et al 2005) but the complex scheme has prevented further development and widespread use.

The use of scattering based contrast agents within OCT also offers the possibility for molecular imaging. For example, plasmonic nanoparticles offer distinct optical signatures (Wax and Sokolov 2009) and can be used in place of fluorescent markers (Crow et al 2009). Gold nanoparticles have been explored for use in OCT, either through detection of their absorption signature (Li et al 2012) or by photothermal excitation (Skala et al 2008) or both (Nahas et al 2014). Recent extension to in vivo use in an animal model has shown the potential of gold nanoparticles as an imaging contrast agent but did not demonstrate molecular specificity (Tucker-Schwartz et al 2014).

Given the early stages of molecular imaging methods in OCT, studies have been limited thus far to animal models. Potential applications of these approaches include detection of neoplastic processes and arterial oxygenation.

Ground-state recovery pump-probe OCT (gsrPPOCT) is an extension of pump-probe OCT that provides molecular specificity with spatial resolution of 10 µm. Applegate and Izatt initially studied gsrPPOCT in the gills of euthanized zebrafish and demonstrated that the gsrPPOCT signal of hemoglobin correlated with OCT images of efferent filament arteries. Arteries were visualized as linear shadows by conventional OCT imaging, which was overlaid with the gsrPPOCT signal to confirm the presence of hemoglobin within these arteries. Limitations of this system included slow line rates of approximately 1 Hz and limited penetration depth due to the short wavelength probe (Applegate and Izatt 2006).

Applegate further developed the approach with a two-color Fourier domain PPOCT system (Jacob et al 2010) that demonstrated imaging depths beyond 725 μm and A-scan rates greater than 1 kHz. This system was used to image an ex vivo porcine eye with the cornea and aqueous humor removed (in order to avoid compensating for the refractive power of the cornea). The porcine iris is rich in eumelanin, which appears as a strong signal on the PPOCT image of the iris (figure 11) (Jacob et al 2010). This result suggested that PPOCT could be helpful in diagnosis of ocular melanoma, but further work with intact human eyes would need to be demonstrated to warrant clinical translation.

Zoom In Zoom Out Reset image size

Fluorescence-guided OCT has been employed in identification of malignancy in animal models. Jo et al incorporated OCT and fluorescence lifetime imaging microscopy (FLIM) into a single system and used it to image oral squamous cell carcinoma in an in vivo hamster model (Jo et al 2010). Compared to the control model, lesions in animals with oral cancer had a distinctive FLIM pattern: the central area had a fluorescence characteristic of NADH/FAD while the surrounding area was characteristic of collagen. This multi-modality approach offers good diagnostic potential without the need to directly detect the contrast agent with OCT. However, the depth gating capabilities of OCT are not extended to the molecular specificity with this approach.

Cardiovascular application of fluorescence guided OCT has been explored as well. Liang et al developed a multimodality fluorescence and optical coherence tomography probe based on a double-clad fiber (DCF) combiner. To evaluate this system, an ex vivo rabbit artery injected with highly saturated grease was stained for annexin V-conjugated Cy 5.5 and imaged. Annexin V is an antibody that targets apoptotic macrophages that accumulate in the core of plaques (Liang et al 2012). While the compelling combination of OCT and molecular imaging could offer significant utility, further work is needed for these modalities to be brought to the clinic.