Phase-sensitive swept source optical coherence tomography for imaging and quantifying of microbubbles in clear and scattering media
Ravi Kiran Manapuram, Venu Gopal Reddy Manne, and Kirill V. Larin
A novel phase resolved system based on swept source optical coherence tomography called as Phase Sensitive Swept Source Optical Coherence Tomography (PhS-SSOCT) to detect and quantify gas microbubbles in aqueous and tissue simulated media is developed. The system shows an axial resolution of 10 μm, phase sensitivity of 0.03 radians, imaging depth of up to 6 mm in air and a scanning speed of 20 kHz for a single A-line. The structural images of the bubbles show an accuracy of 10 μm where as the temporal phase response show an accuracy of 0.01 μm. Images of rapidly moving bubbles of sizes greater than 25μm are also presented which indicate that the PhS-SSOCT could be ultimately applied for the rapid assessment of microbubbles in biofluids, tissues and sclera of the eye.
Introduction
Microbubbles in blood and tissues are one of the root causes for diseases like Decompression Sickness (DCS), arterial or venous gas embroils (VGE) and barotrauma. The symptoms shown by these diseases are severe pains in joints, pulmonary problems, disorientation and mental dullness, vomiting and skin rash. Bubbles may act as emboli and block circulation, as well as cause mechanical compression and stretching of the blood vessels and nerves. Additionally, the blood-bubble interface acts as a foreign surface, activating the early phases of blood coagulation and release of vasoactive substances from cells lining the blood vessels. Formation and/or introduction of gas microbubbles in human blood and tissues remains a serious long-term sequel in patients undergoing cardiac valve replacement (with an annual risk of up to 4%); high-intensity focused US therapy; cesarean section and operative hysteroscopy; cardiopulmonary bypass and other open-heart surgeries; orthopedic surgery and various laser ablation and laparoscopic surgeries. Additionally, gas embolism happens in endoscopy, tissue biopsy, neurosurgery, liver transplantation, during central venous line insertion and removal and even during intravenous antibiotic delivery at home. The use of ultrasound bubble contrast media could also lead to gas emboli. Thus, formation and introduction of gas microbubbles in human blood and tissues is a significant everyday clinical problem affecting thousands of patients undergoing various surgical and therapeutic procedures. If detection of these bubbles at an early stage is possible, action can be taken to prevent neurological or other complications. Hence, noninvasive functional imaging, monitoring and quantification of microbubbles forming in blood and tissues have gained importance in searching for effective therapy and early diagnosis.
A high in-depth, high-speed and high-sensitive device is required for the imaging, monitoring and quantification of these microbubbles in epithelial tissues. Previously, several imaging techniques have been proposed and applied to the study of microbubbles in blood including Doppler Sonography, Magnetic Resonance Imaging (MRI), Nuclear Imaging and Computer Tomography. Doppler Sonography, the most popular technique since air-bubble interface produces strong ultrasonic reflection in the region of 1 MHz to 20MHz, is an ultrasound diagnostic imaging technique. Enhanced with Doppler Effect, it is capable of the assessment of moving bubbles by calculating the frequency shift of a particular sample volume. However, Doppler Sonography can detect only moving intravascular bubbles with a diameter of approximately 50 μm. Evidently, this resolution should be improved in order to achieve sensitive imaging and assessment of small microbubbles in blood and tissues.
Optics-based techniques have great intrinsic potential to achieve the goal of noninvasive imaging of microbubbles in tissues and bio fluids. Confocal-laser scanning microscopy (CLSM), two-photon fluorescence microscopy (2P-FM) and higher harmonic generation (HHG) microscopy are some examples of optical methods that have been applied in different fields of biological research. CLSM has significant axial and lateral resolutions, but it is limited to ~100 µm penetration depth due to high attenuation of visible/ultraviolet excitation light. 2P-FM has a higher penetration depth due to near-infrared (NIR) excitation of fluorophores, but general needs of exogenous fluorophores make this technique not truly “noninvasive.” HHG microscopy does not require application of exogenous fluorophores, but is bulky and very costly.
Optical interferometric techniques are extremely sensitive to local changes in scattering, absorption and refractive indexes of the tissues and cells. Since the refractive indexes of blood and air are quite different (1.4 and 1.0, respectively), an optical-based sensor will be capable of assessing formation of gas bubbles with ultra-high sensitivity and accuracy. This motivates the development of a phase sensitive, in-depth imaging device which would detect the microbubbles whose sizes are beyond the imaging capabilities of standard imaging techniques. This paper presents the development and application of phase-stabilized swept-source optical-coherence tomography (PhS-SSOCT) that could potentially be used for real-time, sensitive, accurate and noninvasive imaging, monitoring and quantification of microbubbles in the skin’s micro vessels and dermis area. High in-depth and lateral resolutions of OCT would allow direct monitoring and quantification of microbubbles in circulating blood and tissues.
Theory
OCT is a relatively new noninvasive optical diagnostic technique that provides depth-resolved images of tissues with resolutions of up to a few micrometers to depths of up to several millimeters. This technique was introduced in 1991 to perform tomography imaging of the human eye with 30µm resolution. Since then, OCT has been actively developed by several research groups for many clinical diagnostic applications (reviewed).
The basic principle of the OCT is to detect backscattered photons from a tissue of interest within a coherence length of the source using a two-beam interferometer. In Time-Domain configuration, each depth corresponds to a different time delays which are measured by moving the reference arm parallel to the tissue depth. Each depth information is obtained at different times making the information time encoded. Recently, novel OCT Fourier detection techniques have emerged which do not require mechanical in-depth scanning and can achieve very high detection sensitivities, enabling OCT imaging with a high increase in imaging speed over standard resolution OCT systems. These Fourier domain techniques measure the echo time delay of light by Fourier transforming the interference spectrum of the light signal as different echo time delays of light produce different frequencies of fringes in the interference spectrum. Fourier domain Optical Coherence Tomography (FD-OCT) offers significantly improved sensitivity and imaging speed compared to TDOCT. Fourier domain OCT detection can be performed in two ways: Spectral Domain OCT (SDOCT), using a broadband light source and a spectrometer with a multi-channel analyzer or Swept Source OCT (SSOCT), using a broadband narrow-pulse swept-laser source and an InGaAs detector.
In the SDOCT, high-performance CCDs, which are made of silicon-based detectors, are generally available only for wavelengths up to 1 µm. The penetration depth of a few samples is limited due to the high scattering of light with lower wavelengths. To enable imaging at higher wavelengths and to absolve the need of high performance CCDs and spectrometers, SSOCT is developed.
In SSOCT, the frequency chirped laser source is split into two arms: the reference and the sample arm. The back-scattered light from the sample which is the time-delayed copy of the reference at particular depths recombines with the reference to give fringes. The intensity of these interference fringes are detected with respect to time, RF modulated with the envelope of the signal and then passed through a band-limited filter. Each depth corresponding to a different delay in the back-scattered light gives a unique RF signal. All these frequencies are displayed at the same time.
For a Swept Source laser, k (t) = 2π/(t) is the wave number that does not always obey a linear relation k(t) = k0 + k1t. In theory if it obeys a linear relation, then the laser would have effectively mapped the k-space to time domain, i.e., i(t) = i(k). In reality, the frequency sweep of the chirped laser is non-linear. The k(t) contains higher order terms and causes non-linearity in the frequency sweep which leads to a non-uniform sampling interval. As the Fourier transform is applied to get the depth information, uniformly-spaced samples are required. By performing Non-Uniform Fourier Transformation (NUFT), depth information from the non-uniformly spaced samples can be obtained. Another approach is by using the interference fringes from a fabry-perot interferometer, uniformly spaced samples are obtained. The non-linearity in frequency sweep is analyzed by several groups and proposed different methods to overcome the problem. This paper employs a Mach-Zehnder interferometer based optical clock (MZI-OC) which generates equally-spaced frequency interferogram that range from 12.5GHz to 200GHz. All peaks, as well as the zero crossings in the recorded fringes of MZI-OC, are always equally spaced in optical frequency space and are used for synchronizing. The samples are collected at these zero crossings thus eliminating the non-linearity in the sampling interval. This also serves the purpose of converting into the uniform k-space. Without calibrating and remapping to uniform k-space, resolution degradation is observed along the z-space. In addition to that, this correction offered a significant improvement in the resolution at a particular depth as shown in Figure 1: the reflectivity profile from a weak reflector without calibrating (dotted) and with calibrating (solid) the signal.
There are several advantages of SSOCT over SDOCT and TDOCT. In addition to the fact that Fourier domain-detection does not need a complex setup for a mechanical scanning of the reference arm, there are sensitivity advantages over TD detection. The SSOCT utilizes a photo detector in a dual-balanced detection mode, in which the common noise is subtracted out of the phase fringes which are then added to enhance the signal strength and sensitivity. In OCT, sensitivity drops off as depth is increased. In SSOCT, however, due to the narrow instantaneous line width of the tuning laser source, interference is observed even at deeper depths as the coherence function remains nearly constant. This effect is more distinctly observed in SDOCT as the laser source is not pulsed. The constant coherence function at deeper depths allows the SSOCT to have higher imaging depths. However, SSOCT suffers from a major drawback: the phase is highly unstable from successive A-line scans in system utilizing mode-unlocked laser source. For a typical SSOCT (without any phase stabilization techniques employed), phase variations from successive A-scans for a homogenous media comes to up to π radians (usually caused by the bulky movements of the polygon mirror inside the laser).
In this paper, we present results on development of a phase resolved sensing system based on SSOCT and evaluated its performance by quantifying microbubbles of different diameters in clear and scattering tissue-simulating media.
Experiment
A. Set up:
The schematic of the system as shown in the Figure 2 consists of four main units: the source, the interferometer (Mach-Zehnder in this particular setup), the calibration system and the data acquisition electronics. The laser source output is split into two arms: one arm containing the interferometer and the other arm containing the recalibration and triggering unit. A 90-10 fiber coupler (Thorlabs) is used to send 90% of the light to the interferometer so that the sample arm gets the maximum possible power. The remaining 10% is further split by a 99-1 fiber coupler with 99% going to a Fiber Bragg Grating (FBG) and 1% to the MZI-OC. The 99% light goes to the FBG through a three-arm circulator and the reflected pulse is passed to the detector by the third arm of the circulator. This detector outputs a voltage signal which is converted to an electrically tunable TTL pulse by a pulse generator (Stanford Research Systems, Inc.,). The TTL signal is tuned to a required duty cycle and used to trigger the analog to digital converter (ADC). The other 1% is fed to the MZI-OC whose signal is detected in balanced detection mode, and these electric signals from the detector are acquired by one of the channels of the ADC. In the interferometer arm, the 90% light is split into 1% and 99%, each going to the reference arm and sample arm respectively via circulators. The light coming from the reference arm is passed through an adjustable pin hole to allow attenuation if required. The reflected light from the reference arm and sample arm are coupled into a 50-50 fiber coupler where they recombine and form the interference fringes. These fringes are then detected by a balanced photo detector (BPD), which subtracts the two signals to remove the common mode noise. As the fringes would be out of phase, this BPD is effectively adding the fringes but subtracting the common mode noise. After removing the common mode noise, the fringe encoded voltage is then amplified by a Transimpedance amplifier (TIA) and then RF modulated and acquired by a PC through the other channel of the ADC. Both the information from MZI and the interferometer is acquired simultaneously by the ADC with the receiving of the trigger from the FBG.
A.1 Data Acquisition and signal processing
The fringes are acquired by a 14- bit high-speed digitizer (model, company). The digitizer is operated at 50 MS/s and acquires 2500 sample points per A-scan. The sampling rate is chosen by considering the fact that the sampling interval should be smaller than the instantaneous line-width; otherwise, a large sensitivity drop off along the depth scan would be observed. Out of these 2500 points, the first 200 and last 252 points are deleted so as to select the data corresponding to the laser wavelength scans. Both the raw signal and the MZI-OC signal will now have 2048 points. After recalibration using MZI-OC, the number of points is decreased due to the fact that the number of peaks and zeros are always less than the total number of points. As a matter of fact, if the nearest neighborhood algorithm is used to find the peaks, which gives 1 point out of 3 points, only one-third of the sample signal would be utilized. Thus, the number of peaks and zeros registered are around 600 indicating the presence of 600 raw data points using spline interpolation, three points are inserted between consecutive raw data points thus having 2400 points. Again, the signal is windowed to have 2048 points. It must be noted that the points that are deleted from the raw signal should not contain any information from the signal. If, in the process of getting 2048 from 2400, any information containing the fringes is lost, then instead of getting 2048 points in the first deletion, more points are retained. The selection of points is made in such a way that the calibrated signal always has 2048 points without losing any fringe information. A complex FFT of this signal gives the depth profile and, due to the symmetry of FFT, each A-line corresponding depth is constructed using 1024 points. It is worth noting that inserting a lower number of points, e.g. two, might result in losing some of the higher frequency components which in turn might reduce the imaging depth of the system. Since the back-scattered light at a depth greater than the coherence length of the laser source cannot form fringes, the maximum imaging depth would not exceed the coherence length of the laser source.
Signal processing includes several steps, including reference subtraction, recalibrating and resampling to uniform k-space and image construction. The background signal is recorded at the beginning of every acquisition by blocking the sample arm. This signal contains the 1% residual light reflected from the reference arm and from the residual signal in the detectors due to imperfect symmetry of the BPD. Subtracting the reference from the interference signal helps to increase the contrast and remove the artifacts. It removes the low-frequency components introduced by the reference or ambient light. Each depth profile is obtained from the Fourier transform of the interference fringe signal. The logarithm of the absolute value of this complex FFT is mapped to the gray scale to get an image of a single A-scan. By generating a transversal set of similar 1-D depth profiles with the scanning galvo-mounted mirror, a 2-D image is constructed. Thus, the 2-D image contains both axial and transverse information. By scanning the Y-mirror, enface imaging is achieved, thus allowing for 3-D imaging.
A.2 PhS-SSOCT:
Jitter present on the output source spectrum causes the acquired fringe data to slowly drift over time. These drifts introduce varying delays between the trigger signal and the subsequent digital fringe data. As the phase is extracted from the complex FFT of the fringe signal, the mismatch in fringe signal and the trigger would result in phase jumps of up to a maximum value of π as shown in Figure 3(a). These jumps could be removed by dynamically triggering the ADC using an electrically tunable TTL signal generated from a narrow band (0.1 nm) Fiber Bragg Grating (FBG) as shown in Figure 2. A reflected optical pulse is generated whenever the source sweeps the FBG reflection wavelength. This pulse is converted into a TTL signal using a signal generator (Stanford Research Systems). By triggering the ADC with the above TTL signal, the jitter due to electronics is reduced by introducing perfect synchronization between the source and data acquisition, which in turn reduces the phase variations: standard deviation of 0.016 radians for 512 A-line scans has been achieved as shown in Figure 3(b).
Regardless of the above removal of π jumps, the MZI-OC introduces lot of phase noise to the system. This is due to the fact that zeros and peaks keep changing with every laser scan. Thus, the phase measured from the sampled fringe signal inhibits high phase variations. To avoid this, the MZI-OC is recorded only once and hence the zeros and peaks will not change with every scan and the recalibrated signal will be stable in the phase. With the stabilized phase, the refractive index change as small as 6.3 x 10-6 (in a 1 mm) can be detected using the PhS-SSOCT.
Methods
Performance of the developed system was evaluated in water and scattering media containing gas microbubbles of different diameters. Water was injected into a 500 µm flow-through cuvette, and bubbles were generated by introducing different pressures using a peristaltic pump (Fisher Scientific). 1.54% polystyrene spheres (PS) were used to simulate the scattering media with a scattering coefficient of 100 cm -1 for the 1324 nm wavelength. The beam is scanned across the cuvette as shown in the Figure 4(a). The amplitude of the interference signals in the time-delay domain was recorded from the cuvette. Four characteristic interferometric peaks were observed corresponding to the interferences between the four surfaces of the cuvette as described in Figure 4(b).
In these experiments, the optical delay is calculated as a function of dynamic refractive index (modified by presence/absence of microbubbles) from the interferometric peaks that are produced by the reflection from the inner walls of the cuvette in the time-delay domain. The phase is extracted from the complex Fourier transform of the interference fringes and monitored at the interferometric peak that corresponds to the self interference between the inner glass surfaces. Phase-sensitive measurements of water are taken before and after injection of the microbubbles.
Bubbles are generated by creating a pressure difference and are circulated using the pump. Bubbles of diameters greater than 10 μm, were first imaged to test the consistency of the PhS-SSOCT imaging. The images were taken in the inter-interference mode, in which the reference arm can be placed in such a way that all surfaces of the cuvvette can be shown in real distances as shown in Figure 4(a). Phase stability was high in self-interference mode compared to the inter-interference mode. Thus, in all the results presented in this paper, the images were taken in inter-interference mode, and the phase response was measured in self-interference mode.
Results and Discussion
a. Clear media:
Figure 5(a) shows the image of a 500 µm cuvette filled with water containing a bubble of diameter 224 μm (actual diameter). A very high display threshold is selected so as to suppress the self-interference image. The 1-D profile of the corresponding self-interference image and the temporal phase response are shown in Figures 5(b) and (c) respectively. It should be noted that the imaging and the phase response were taken one after the other in very quick succession, as the imaging was done in inter-interference mode and the phase response was studied in self-interference mode. Although there could be a very slight manual error, there was an attempt to maintain correlation between the image and the phase response as much as possible.
The reference arm is placed at a distance so that the whole cuvette can be imaged without any negative images that arise due to the symmetry of FFT. Therefore, for Figure 3(a), all four bright lines correspond to all four surfaces of the cuvette. In the clear aqueous media, the maximal phase variations were as low as 0.03 radians after a 5-point averaging, which implies that any microbubble that introduces a random phase greater than 0.03 radians can be easily detected.
Generally, for an optical path difference of one wavelength the phase shift would be 2π in a homogenous media. Equation 2 describes a relationship between the changes in the refractive index and the phase of PhS-SSOCT. For an air bubble in water, the change in the refractive index would be 0.33 which translates to a minimum bubble size of 2 µm for the phase to be unwrapped by one 2π jump. Since the system’s resolution is 10µm, bubbles with diameters greater than 10µm can be clearly seen in the structural image, as shown in Figure 5(a). When there is no bubble, the optical path length between the inner surfaces of the cuvette is 665 µm (refractive index of water is 1.33, so 500 x 1.33 = 665 µm), which is observed as a peak at 665 µm in corresponding 1-D depth profile. As the beam interacts with the bubble, the optical path length keeps decreasing until it reaches the center of the bubble and increases again to the original value. This change in the optical path length is reflected as a shift in the peak in the corresponding 1-D profile as shown in the Figure 5(b). The larger the bubble, the greater is the decrease in the optical path length and hence the greater is the shift. The number of 2π jumps by which the phase to be unwrapped is then calculated from the peak shifted in the 1-D profile. This unwrapped phase is then added to the PhS-SSOCT phase response to get the true phase response. For the clear media, each depth pixel corresponds to a path difference of 5.88 µm, which is equal to 2.94 “2π” jumps. Thus, for the bubble shown in Figure 5(a), the number of 2π jumps would be 69/2*2.94 = 113 as the peak shifts by 69 depth pixels [Figure 5(b)]. By plugging this true phase in Equation 2, the size of the bubble obtained is 219 µm. The actual diameter of the bubble measured is 224 µm. The true phase and the PhS-SSOCT phase (phase before adding required 2π jumps) is plotted in the same graph with two different scales on the right and left of the Y-axis.
Similarly, several bubbles with different diameters (52 µm, 94 µm, 160 µm etc.,) were taken and quantified. The obtained error ranged from 0.19 µm to 10 µm which can be attributed to the 2π ambiguity. The change in the path differences less than 10 µm is not reflected in the 1-D profile (due to limited imaging resolution of 10 µm). In this case, the number of 2π jumps by which the phase should be unwrapped cannot be determined and can take any integer value between 1 and 5. This means that microbubbles with diameters with multiples of 2 µm (up to 10 µm) would show the same phase response.
For instance, if the number of 2π jumps calculated from the phase shift is 60, the actual number of 2π jumps can be any number between 60 and 65. Likewise, if the measured diameter of a bubble is 120 µm, then the actual bubble diameter could be 120, 122, 124, 126 or 128 µm. However, this ambiguity can be resolved by implementing a fast real-time continuous unwrapping algorithm which will be developed in our future studies. This algorithm would acquire multiple phase recordings between consecutive changes in the path difference of 2 µm and continuously unwrap the phase information. It would also allow the exact quantification of microbubbles that are beyond the imaging capabilities of the system.
Figures 6(a) show an example of small bubbles which cannot be resolved from SSOCT structural imaging. These bubbles can, however, be detected by the PhS-SSOCT. The sizes of the bubbles were estimated to be 1.9 µm (but could be 3.9 µm, 5.9 µm, 7.9 µm or 9.9 µm due to the ambiguity discussed above) for Figure 6(a). Therefore, PhS-SSOCT is an effective device for ultra-sensitive quantification of microbubbles with diameters significantly less than imaging capabilities of the employed system.
b. Scattering media:
The detection, imaging and quantification of microbubbles in scattering media were also performed on a similar note. Figure 7(a) depicts the image of the cuvette when filled with scattering media. A high dynamic range is chosen to identify if any self-interference lines are observed. It can be noted that a very faint line is visible at a 705 μm path length. The phase is monitored at that peak. The phase variations obtained were 0.04 radians, indicating the minimum size of the bubble that could be detected is 0.01 μm. Figure 7(b) shows the image of a bubble with an actual diameter of 142.3 μm, and the diameter obtained from PhS-SSOCT measurements was 143 μm from Figure 7 (d). Note that the phase response in Figure 7(d) is shown for few A-lines so that the phase response could be clearly displayed.
The bubbles that are beyond the imaging capabilities of SSOCT are shown in Figure 8. The portion circled in blue in the image is the place where the phase indicates that there could be three bubbles of diameters all around 0.8 μm (but that could be 2.8 μm, 4.8 μm, 6.8 μm or 8.8 μm). As the sizes of the bubbles are smaller than the focused beam spot (~ 25 μm), they act as a hindrance to the beam passage and create a shadow on the other interface as shown in Figure 8(a).
The PhS-SSOCT is tested for fast moving large bubbles. Figure 9 depicts the M-mode image of air bubbles and air gaps that encountered the beam. The sizes of the bubbles can be predicted from the depth scan (Y-axis) and the speed of the bubbles from the transverse scan (X-axis). The large air gaps have a bright line at the end of the cuvette which was less deep than when there was no air gap, indicating the presence of some medium that has a lower refractive index than the scattering media.
Conclusions
In this paper, we demonstrated that the PhS-SSOCT is capable of imaging, detecting and quantifying of large and small microbubbles in clear and scattering media. The results suggest that small micro bubbles with diameter beyond imaging capabilities of the system can be detected and quantified using the PhS-SSOCT. Potentially, micro bubbles of diameters as small as 0.01 µm (that introduce a phase shift of 0.03 radians) can be detected and quantified with this method. Our future studies will focus on the development of the effective phase unwrapping algorithm that will quantify the micro-bubbles in both clear and blood simulated media as well as in tissues in vivo with no 2π ambiguities.
Acknowledgments
The study is supported by a grant from Office of Naval Research. Authors would like to thank Narendran Sudheendran and Esteban F. Carbajal for extensive discussion and analysis of the results presented in this paper.
List of Figures
Figure 1: Improvement in resolution after calibration
Figure 2. The schematic outline of Phase-Stabilized SSOCT system (ADC: Analog to Digital Converter, BPD: Balanced Photo Detector, C: Collimators, FBG: Fiber Bragg Grating, G: Galvanometer Scanner, L: Lens, RM: Reflective Mirrors, SLS: Swept Laser Source, TTL: TTL Signal Generator).
Figure 3: The temporal phase response of 500 μm cuvette at 500 μm depth (a) before stabilization (b) after stabilization (Note the scale)
Figure 2-22 (a) Orientation of the cuvette with respect to the beam. (b) Corresponding 1-D depth profile (A- glass thickness, B- optical path thickness, C- A+B).
Figure 5 (a) Image of a cuvette of 500 µm thick containing a bubble of 224 µm (fast moving) (b) corresponding 1-D profile showing the shift in peak (c) Temporal Phase response at 665 µm peak
Figure 6: Image of a cuvvette of 500 µm thick with (a) very small bubbles and corresponding temporal Phase response at 665 µm peak (b) of (a)
Figure 7: Image of the cuvette with scattering media (a) without any bubble (c) with a bubble and their corresponding temporal phase response in (b) and (c).
Figure 8: Cuvette with very small bubbles and their corresponding temporal phase response
Figure 9: Image of 500 µm cuvette containing fast moving bubbles