Assessment of microvasculature flow state with a high speed all-optic dual-modal system of optical coherence tomography and photoacoustic imaging.

We propose a high speed all-optic dual-modal system that combines spectral domain optical coherence tomography (SDOCT) and photoacoustic imaging (PAI) to evaluate microvasculature flow states. A homodyne interferometer was used to remotely detect the surface vibration caused by photoacoustic (PA) waves. The PA excitation, PA probing and SDOCT probing beams share the same X-Y galvanometer scanner to perform fast two-dimensional scanning. In addition, we introduced multi-excitation, dual-channel acquisition and sensitivity compensation to improve the imaging speed of the PAI sub-system. The total time for imaging a sample with 256 × 256 pixels is less than 1 minute. The performance of the proposed system was verified by in vivo imaging of the vascular system in a mouse pinna with normal and then blocked blood circulations. The experimental results indicate that the proposed system is capable of revealing different blood flow states (static and moving) and is useful for the study of diseases related to functional blood supply.

PAI detects ultrasonic signals produced by light absorption in a sample. Due to the high light absorption characteristic of hemoglobin, PAI is sensitive to all blood vessels regardless of flowing states inside. Both OCTA and PAI are powerful tools for angiography, and their combination can provide more useful information of blood flow for the vascular diseases, such as stroke.
In recent years, some multi-modal imaging systems combining PAI and OCT have been reported [16][17][18][19][20][21]. The integration of the two modalities is limited by the discrepancy of their detection components. The transducer used in PAI obstructs OCT scanning [22]. Optical interferometric methods, such as Mach-Zehnder interferometer [23][24][25] and Fabry-Perot interferometer [26][27][28], have been proposed as the alternatives to transducers. These techniques also have difficulties for the combination with OCT, because bulky and complicated optics are involved in general [29]. Fiber based Michelson interferometers have been reported for photoacoustic (PA) signal detection, including low-coherence and long-coherence interferometers [30,31]. Park et al. proposed an all-fiber heterodyne interferometer for non-contact PA tomography (PAT) [32]. Using this technique, Eom et al. established an all-fiber-optic PAT and OCT multi-modal system. In order to achieve heterodyne detection, acousto-optic modulation is required and the reported system is complex. The homodyne interferometer is relatively simple and easy to be implemented [31,33]. However, the sensitivity of the system varies due to ambient disturbance. Data acquisition is recommended to be performed when the system reaches its maximum sensitivity. This operation is time consuming and reduces the imaging speed of the multi-modal system.
In this paper, we present a high speed, all-optic dual-modal system that integrates OCT and PAI. To improve the imaging speed of PAI, a novel acquisition strategy was adopted, including multi-excitation, reference arm modulation, and 2-channel acquisition. It is unnecessary to lock the system at its maximum sensitivity. PA signals were compensated according to the instantaneous systematic sensitivity. To test the capability of the system for assessing blood flow state, the vasculatures of a mouse pinna was imaged in vivo with and without flowing blood.

OCT and PAI dual-modal system
The dual-modal setup used in this study integrated a spectral domain OCT (SDOCT) subsystem and a PAI subsystem with optical detection as illustrated in Fig. 1. In the SDOCT subsystem, we use a broadband infrared superluminescent diode (SLD, D-840-HP, Superlum) with a central wavelength of 840 nm and a full width at half maximum (FWHM) bandwidth of 80 nm as the illumination source to provide an axial resolution of ~6 μm in air (red beam in Fig. 1). The ex-fiber output power was rated at ~15 mW. Light from SLD went through an optical circulator and was split into a probing arm and reference arm through a 2 × 2 fiber coupler with 10:90 split ratio. The ninety percent power path went to the probing arm while the ten percent power path went to the reference mirror. The lights reflected back from both reference and probing arms met and interfered with each other at the fiber coupler, and the resulting interferogram was sent via the optical circulator to a custom-built high-speed spectrometer. The spectrometer consists of a transmission grating (1800 lines/mm), a camera lens with a focal length of 100 mm, and a line-scan camera (spL2048-140km, Basler). The theoretical depth range was calculated to be ~2.4 mm in air. The line-scan camera was running at 50,000 Hz to convert interferograms to digital signals. With each B-scan frame (i.e. X direction) containing 256 A-lines, the imaging speed of the OCT can reach 160 frames per second. The space between adjacent A-lines is ~5 μm and each B-scan spans is ~1.3 mm. The C-scan consists of 256 cross-sections with ~5 μm space interval covering ~1.3 mm. The acquired interferogram data were transmitted to a workstation through an image grabber (PCIe-1433, NI).

OCT angiography
The detailed description of OCTA has been described previously [34,35]. In SDOCT, the interferogram is captured by the line-scan camera. For simplicity, we ignore all components that do not contribute to the useful information about microstructure and flow. The interferogram can be expressed as: where k is the wavenumber; t is the time at which the interferogram is captured; E R is the amplitude of the light reflected from the reference mirror; S(k) is the spectral density of the light source; z is the depth; a(z, t) is the amplitude of the light backscattered from the sample; n is the refractive index; v is the axial velocity of blood cells. According to previous analysis [36,37], the sensitivity to the flow velocity detection is determined by the time interval between adjacent interferograms used for velocity calculation. The sensitivity increases with the time interval elongating. Here, high-pass filtering was applied along the slow scanning C-scan direction to isolate the scattering signal from moving blood cells from the static tissue. Since two B-scans were acquired in each cross-section along the C-scan direction with Δt B time interval, a differential operation was applied to the subsequent B-scan at each cross-section. This can be described by the following equation: where I j (t,k) denotes the flow signal at jth cross-section (total of 256 cross-sections) along the C-scan direction. As the differential operation is equivalent to high-pass filtering, it suppresses optical scattering signals from static elements. Then, by applying fast Fourier transforms (FFT) upon every wavenumber k of I j (t,k), we can obtain a depth-resolved flow image sensitive to blood flow for each cross section. With all the cross-sectional flow images, 3D volumetric perfusion map can be rendered using visualization software.

High speed PAI with multi-excitation and sensitivity compensation
In our PAI subsystem, the sample was illuminated by a focused short pulse laser. The absorption of laser energy by local absorbers inside a sample generated ultrasonic waves via thermal-elastic expansion. With plane-wave approximation, the ultrasonic waves leading surface vibration are expressed as [38]: where p(t) is the pressure of the ultrasonic wave, ε(t) is the surface displacement, and C is the acoustic impedance of the medium. Note that the scaling factor half represents the free boundary condition at the air-tissue interface. A fiber based homodyne interferometer was used to measure the surface displacement. In order to eliminate the DC component of the interference signal, balance detection was performed. Therefore, the intensity of the signal can be described as [39]: where the coefficient A is related to the intensities of the two interfering beams; λ is the wavelength of the detection laser; ( ) t φ is the optical phase difference between the probing and reference beams related to variations in ambient conditions. ( ) t φ varies slowly (less than a few hundred Hz where A is the amplitude of D(t). Hence, surface displacement caused by PA waves can be achieved as follows: The calculated ( ) t ε is used to reconstruct PAI images. The advantage is that the varying sensitivity is compensated. This provides the potential for improving the scan speed. In Eq. (8), when ( ) t φ is close to nπ , the sensitivity is approximately zero (PA 2,4,7,8 in Fig. 2). In such a case, sensitivity compensation is susceptible to system noise; therefore data acquired within those regions need to be exempted from image reconstruction. To avoid such influence, three pulses were generated by counter port of AO device (Fig. 1)  is close to zero (detailed description is shown in section 3.1).

PAI sensitivity compensation
In order to evaluate the proposed high speed PA imaging method, we imaged a phantom with a fine tungsten filament (~100 μm) embedded in scattering gel. Focusing on the same position on the tungsten filament, PA signals were excited with three sequential laser pulses (~5 kHz frequency). Figure 3(a) shows three BD output waveforms acquired without a reference arm modulation. The excitations were completed in ~0.5 ms, and the interference phase ( ( ) t φ ) of two waveforms changed little during such a short period of time (red and pink lines in Fig.  3(a)). Therefore, three PA waves may all fall into the low sensitivity region, which will affect the accuracy of the sensitivity compensation. This problem can be solved by reference arm modulation. During the acquisition, the reference mirror was controlled to shift ~330 nm ( λ / 4 ), which introduced a π phase change. Therefore, at least one PA wave can be free from the low sensitivity regions (PA 1 and PA 3 in Fig. 3(b)). The high-pass filtered signal acquired by the other channel is shown in Fig. 3(c). Then, the three PA signals were compensated using the proposed methods, where ( ) 81. 6 ,9.4 , and 68.8 t φ =°°° respectively (Fig. 3(d)). We can see that the amplitudes of PA 1 and PA 3 are similar. The signal of PA 2 was over-compensated because it falls into the low sensitivity region and ( ) sin t φ approximates zero. To evaluate the effect of the reference arm modulation, the phantom was imaged without and with modulation, and the sensitivity compensated images were shown in Fig. 3(e) and 3(f), respectively. We can see that the intensity of tungsten filament is not uniform in the image (Fig. 3(e)). This means that under-compensation or over-compensation may occur in the low sensitivity region. The proposed method utilized reference arm modulation and high sensitivity region signal selection, and achieved a better imaging result (Fig. 3(f)).
Traditionally, PA signals need to be acquired at QPs for a homodyne interferometer based vibration measurement. The advantage of this operation is that the maximum sensitivity can be achieved for occurrence of Y) matrix PA occurrence fr overall imagin scanning spee position(~0.5 much closer dual-modal sy flow in short sensitivity tha amplitude of system is con average SNR detection (39 Fig Fig. 4(b) and n Fig. 4(b)). A ce OCTA imag ously, the micr blood flow blo 4(h). These tw combination ow status. Acc nna was flowin his is consistent d flow. After flo proposed syst eight-week-old in air). The mou ed mouse pinna ( Fig. 4(a) tem, we ima mouse (fem use pinna was a was then flatl )). Then, the bl ng results were reduce the infl ntroduce strong ferogram to sat ood circulation A. For each du ing, water drop nts were perfo and Use of La rsity. The incid ubsystem, the beam is ~3 mW ructure images is no obvious ed to extract bl with flowing b w blocking, th ted in Fig. 4(c) distribution of t (g)). The corre similar becaus n from OCTA . 4(c) and 4(d) , because bloo riment because he blood vesse l ) s aged the male) was depilated ly affixed lood flow shown in fluence of g specular turate and in mouse ual-modal plets were formed in aboratory dent light transient W.
captured structural lood flow blood are here is no ) and 4(g) the mouse esponding se PAI is and PAI ), we can od vessels e Fig. 4(c) els are not visible by the OCTA technique ( Fig. 4(g)), although vasculature was still present in the mouse pinna ( Fig. 4(h)). The dual-modal system may be applied to stroke research. Ischemic stroke is caused by a block of blood supply, for example, a thrombus that occludes the artery. For OCTA technique, two acquisitions (before and after occlusion) are required to determine the occluded parts of the blood vessels. The proposed system can use a single dual-modal scan to distinguish the flow state in blood vessels without having to compare it with previous results. OCTA is a motion sensitive technique; therefore it is unable to visualize some capillaries with very small flow velocity (blue circle in Fig. 4(c)). However, these vessels can be seen in PAI due to the presence of red blood cells and corresponding light absorption (blue circles in Fig. 4(d) and  4(h)). This result indicates that OCTA and PAI are complementary. Comparing the OCTA and PAI images, we can see that the diameters of the blood vessels in OCTA are larger than in PAI, especially in big vessels. This may be attribute to that OCTA can extract the entire flow area of the blood vessel from the cross-sectional scan images, while PAI only visualize red blood cells distributed in the center of the vessel for normal blood flow (axial flow) and sedimentate at the bottom when the flow stops. There are some low intensity speckles in our angiography images, and their positions correspond in OCTA and PAI (white arrows in Fig. 4(c),4(d),4(h)). The cause of speckles is unclear and should be uncovered in future research.

Discussion
To achieve non-invasive vasculature imaging using endogenous contrast, optical imaging could be an option. Scattering and absorption are two categories of light-tissue interaction, represented by OCTA and PAI, respectively. Based on the backscattered light from the sample, OCTA can extract intrinsic motion signal introduced by flowing blood cells from the surrounding static tissue. Thus, OCTA is able to visualize vessels with flowing blood, which represents blood flow perfusion (functional blood vessels). Most severe pathologies are associated with vascular abnormalities, such as stroke, trauma and cancer. Mapping blood flow perfusion is help to the pathogenesis of these diseases. With the progress of these diseases, another mechanism is activated and the damaged or blocked blood vessels begin to necrotize and are absorbed by surrounding tissue. However, because blood cells in these vessels don't move any more, OCTA cannot catch these vessels and monitor their progress. On the other hand, PAI can display all blood vessels with hemoglobin presence. The combination of OCTA and PAI is able to distinguish flow conditions in blood vessels, which makes it possible to monitor the entire process of disease development. The proposed dual-modal system integrates two complementary imaging techniques and can provide both anatomic and functional vasculatures. In our proposed system, the two detection units are all fiber based interferometers that bear the advantages of simplicity, compactness and flexibility. Although the PAI excitation part is still in free space, the high power laser was combined with the other two detection beams before the X-Y scanner. Therefore, the scanning part of the dual-modal system is compact and easy to operate. Two dichroic mirrors were used for combining three light beams, namely the OCT probing beam (~840 nm), the PA excitation laser beam (527 nm) and the PA signal detection beam (1310 nm). Dichroic mirrors provide high coupling efficiency for beams with different wavelength (>90%). However, the dichroic mirror introduces additional dispersion to the sampling arm of OCT subsystem and causes dispersion mismatch between the two interference arms, which degrades the axial resolution of the OCT. Using the dispersion compensation algorithm [42], the resolution reduction was partially compensated. The measured axial resolution is ~8 μm in air slightly lower than the theoretical value ~6 μm. Considering that the refractive index of the sample is ~1.35, this axial resolution is sufficient for imaging the capillary (7~8 μm).
The PAI subsystem uses fine optical focusing to provide lateral resolution, while the axial resolution is still derived from time-resolved ultrasonic detection, which is called optical resolution photoacoustic microscopy (OR-PAM) [43]. However, focusing the excitation laser beam restricts the imaging depth to the transport mean-free path (TMFP) [44]. The TMFP is the depth at which photon scattering directions become randomized and limits the penetration of ballistic optically focused microscopy systems (including OCT and PAM). In soft tissue, the TMFPs of the OCT and PAI are calculated as [45,46] Fig. 4(d) have low contrast and are difficult to distinguish, while the blood vessels in Fig. 4(c) are clear and have high contrast. Increasing the power of the excitation laser or optimizing the detection sensitivity can improve the penetration depth of PAI [47]. Because shutters are used to avoid incident light superimposing between OCT and PAI, their safety limits of American National Standards Institute (ANSI) should be considered separately. For PAI, the imaging area is 1.3 × 1.3 mm 2 ,and the power of the detection beam is 3 mW, so its average surface power is 0.18 W/cm 2 , which is much lower than the average power limit of 3 W/cm 2 set by the ANSI. The transient fluence of the excitation laser is 16 mJ/cm 2 , which is lower than the single pulse limit set by ANSI of 20 mJ/cm 2 . For multi-excitation situations, the maximum permissible exposure (MPE) for each single pulse within the group shall not exceed the single-pulse MPE limit multiplied by a correction factor of 0.25 n − , where n is the number of pulses [48]. With three laser pulse excitation at each point, the equivalent transient fluence limit is 15.2 mJ/cm 2 , which is close to that we used in this study. Therefore, the total incident power on the sample is slightly higher than ANSI limit in PAI subsystem. For OCT, its incident power is the same as that of the probing beam in PAI, which is lower than the ANSI limit.
Non-contact optical detection of ultrasound in biological tissues is of great interest because it expands the scope of PAI to biomedical applications where contact is impractical [49]. Heterodyne interferometer based method uses an acousto-optic modulator (AOM) to frequency-modulate the reference beam. The interference signal is demodulated using an in-phase and quadrature demodulator. The two demodulated signals are captured for image reconstruction [50]. Heterodyne interferometer is relatively less sensitive to ambient noise compared with the homodyne one and does not need active stabilization [32]. However, the detection sensitivity defined as the peak value divided by the noise root-mean-squire value of the intensity fluctuations was only 13.7 dB [32]. In their results, the noise level of PA signal is high. Additional noise may be introduced during modulation and demodulation. Attempts have also been made to photoacoustic detection using a low-coherence interferometer [30,51,52]. However, the sensitivity of the low-coherence interferometer decreases with the increase of optical path length difference (OPLD). To address this problem, Lu et al. present a PAI system using a homodyne interferometer with a long coherence length laser that allows for a constant high sensitivity in a large dynamic range of the OPLD [31]. The system locks at the maximum sensitivity during data acquisition. However, this reduces the imaging speed. In this paper, we improved the imaging speed of our previous system and integrated it with OCTA to achieve dual-modal imaging. Recently, a new technique named deep photoacoustic remote sensing (dPARS) has been proposed. Instead of measuring the sample surface displacement in heterodyne and homodyne methods, dPARS detects PA initial pressures by measuring the intensity reflectivity modulations of a low-coherence probe beam [53,54]. dPARS needs Z-scan to image structures at different depth. Besides this, it has many advantages, such as a higher SNR and deeper penetration depth. dPARS may offer new opportunities for clinical translation.

Conclusion
In summary, we presented a dual-modal optical system integrating non-contact PAI and OCT. PA signals were acquired using a fiber-based homodyne interferometer. The PA excitation, PA probing and SDOCT probing beams were combined using two dichroic mirrors. Three beams shared the same X-Y galvanometer scanner to perform fast two-dimensional scanning. The PAI acquisition speed was improved by the application of sensitivity compensation and reference arm modulation. PAI and OCTA were performed on the mouse pinna with and without flowing blood. The results indicate that the proposed system is capable of revealing blood flow states in vessels and is useful for the study of diseases associated with blood supply.