Open Access
Add to BibSonomy
Abstract
We recently proposed an optical coherence tomographic angiography (OCTA) algorithm, Gabor optical coherence tomographic angiography (GOCTA), which can extract microvascular signals from a spectral domain directly with lower computational complexity compared to other algorithms. In this manuscript, we combine a programmable swept source, an OCT complex signal detecting unit, and graphics process units (GPU) to achieve a real-time en-face GOCTA system for human skin microvascular imaging. The programmable swept source can balance the A-scan rate and the spectral tuning range; the polarization-modulation based complex signal detecting unit can double the imaging depth range, and the GPU can accelerate data processing. C++ and CUDA are used as the programming platform where five parallel threads are created for galvo-driving signal generation, data acquisition, data transfer, data processing, and image display, respectively. Two queues (for the raw data and en-face images, respectively) are used to improve the data exchange efficiency among different devices. In this study, the data acquisition time and data processing time for each 3D complex volume (256×304×608 pixels,) are 405.3 and 173.7 milliseconds respectively. To the best of our knowledge, this is the first time to show en-face microvascular images covering 3×3 mm2 at a refresh rate of 2.5 Hz.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
In human body, circulatory system supplies oxygen and nutrition for all tissues and the exchange process between blood and tissue cells occurs in capillary vessels. On the contrary, the microvascular distribution can be affected by tissue lesions, such as the higher vessel density around tumors. Therefore, the features of microvascular maps can be used as the evidence for disease diagnosis and treatment monitoring. Due to the advantages of high imaging speed, high resolution, non-invasiveness, and label-freeness, optical coherence tomographic angiography (OCTA) is one of the most widely used microvascular imaging techniques on human tissues (such as retina [1] and skin [2]) in clinical applications.
Since moving scatterers can modulate both amplitude and phase signals in optical coherence tomography (OCT), a variety of OCTA algorithms based on amplitude, phase, or complex analytical signals have been proposed for the past three decades. The amplitude-based algorithms include speckle variance OCT (SVOCT) [3,4], correlation mapping OCT (cmOCT) [5,6], split-spectrum amplitude-decorrelation angiography (SSADA) [7], differential standard deviation of log-scale intensity (DSDLI) [8], etc. In these algorithms, the dynamic information is contrasted by calculating the variance, standard deviation, or de-correlation coefficient between the multiple frames of amplitude images from the same position.
Phase based OCTA algorithms include phase variance OCT (PVOCT) [9], differential phase standard-deviation OCT (DPSD) [10], Doppler variance phase resolved (DVPR) [11] method, etc. Phase based technique usually can achieve higher sensitivity compared to amplitude-based algorithms, since phase is more sensitive to movements. However, phase-based methods also suffer severer bulk motion noise; and for the phase-unstable swept source lasers, phase jittering needs to be calibrated before vascular signals calculation [12,13]. Vascular information can also be extracted from complex signal, such as optical micro-angiography (OMAG) [14] and complex differential variance (CDV) [15] algorithms. Because phase signal is involved for calculation, bulk motion compensation and phase calibration for phase-unstable light sources are also needed.
Two key parameters of OCTA imaging system in clinical applications are the scanning speed and data processing speed. Some patients feel struggling to keep still during scanning, so strong motion artifacts can be caused for low scanning speed systems. Most of commercial systems acquire raw data first and then perform post-data-processing to calculate vascular images. Sometimes a patient needs to be scanned multiple times to acquire an acceptable dataset for the ophthalmologist’s review, resulting in a low screening efficiency. If high scanning speed and high data processing speed can be both performed on OCTA systems, many merits can be achieved for clinical applications, such as higher image quality and higher clinical efficiency.
Recently, multiple high-speed OCTA imaging systems have been presented and a comparison is shown in Table 1. Ref. [16–19] have high A-scan rate, but vascular images are calculated by using post-data-processing. Since graphics processing units (GPU) have more cores and parallel structure (which is more efficient for data processing compared to central processing units (CPU)), GPU has been used to accelerate data processing for vascular imaging in OCT, such as Ref. [20–22]. However, only cross-sectional vascular images are obtained in Ref. [20], and the refresh rates of real-time en-face vascular imaging at 0.06 Hz (with a 50 kHz laser) and 0.67 Hz (with a 200 kHz laser) are respectively achieved by Ref. [21] and Ref. [22]. Note that the refresh rates given here are the actually achieved performance by the systems instead of the possible theoretical performance.
Table 1. Comparison of several high-speed OCTA imaging systems
All current OCTA algorithms calculate vascular signals in spatial domain after k-space resampling (for spectral domain OCT) and Fourier transform (FFT) which are computationally intensive, so it is very challenging to calculate en-face microvascular images (3D volume based) in real-time. We recently proposed an OCTA algorithm, Gabor optical coherence tomographic angiography (GOCTA) [23,24], which can extract vascular signals from spectral domain directly. In this way, the computational complexity of data processing can be greatly decreased since k-space resampling (in spectral domain OCT) and FFT are not needed. In Ref. [23,24], the datasets acquired by a commercial retinal imaging system are used to validate the theory of GOCTA. In this manuscript, we present a home-built, real-time, microvascular imaging system for human skin, in which a programmable swept source, an OCT complex signal detecting unit, and graphics processing units (GPU) are combined with the GOCTA algorithm to achieve the high-speed capability. The programmable swept source can balance the A-scan rate and spectral sweep range; the polarization-modulation based complex signal detecting unit can double the imaging depth range; and GPU can accelerate the data processing.
2. Method
2.1 System setup
Figure 1 illustrates the schematic of the en-face GOCTA imaging system, in which a 90:10 fiber coupler is used to split the light from light source into reference beam and sample beam. The laser is a programmable akinetic swept source (Insight, US) with a center wavelength of 1310 nm. The two collimators (L1 and L2) in sample arm and reference arm are both the F280APC-C, Thorlabs, US. In the sample arm, the objective lens (L3) has a focus length of 30 mm, giving a lateral resolution of 8.6 µm. A pair of galvos (6200H, Cambridge Technology, US) is used for 2D lateral scanning in the sample arm. The driving signals are generated by a programmable function generator (PCI-6731, National Instruments, US). The back reflected light from the reference arm and the backscattered light from the sample arm are combined by a cube beam splitter.
Fig. 1. Schematic of en-face GOCTA imaging system, L1-L9: achromatic lens; PC: polarization controller; P1-P2: polarizer; BS: beam splitter; QWP: quarter wave plate; PBS: polarizing beam splitter; BD: balanced detector; FG: function generator; DAC: data acquisition card.
Vakoc et al. proposed a polarization-modulation based OCT complex signal detecting technique [25] to simultaneously acquire the orthogonal components, in which a π/2 phase shift between the two orthogonal components is achieved by modulating the polarization of the two optical channels with fiber polarization controllers. In this work, a free-space system structure is applied so that a quarter wave plate can be used to more accurately introduce a π/2 phase shift. In the complex signal detecting unit, all the collimators (L4-L9) are F260APC-C (Thorlabs, US). Two polarizers (LPNIR050-MP2, Thorlabs, US; orientated at 45° (P1) and -45° (P2) respectively, referred to transverse-axis) are used to modulate the reference light and sample light. In one output channel of the cube beam splitter, a quarter wave plate (AQWP05M-1600, Thorlabs, US; orientated at 45° referred to transverse-axis) is used to delay the optical signal in the slow axis with a phase shift of π/2 compared to the signal in fast axis. Two polarizing beam splitters (CCM1-PBS254, Thorlabs, US) are used to combine the reference light and sample light in different polarizing directions, and the output interfered light is forwarded to the two channels of the balanced detectors. In this way, the orthogonal interference fringes can be acquired simultaneously by the two channels of data acquisition card (DAC, ATS9350, AlazarTech, Canada). Note that no interference occurs after the first cube beam splitter, since the polarizing directions of sample light and reference light are orthogonal.
From figure 4 in [24], we can see that the vascular image quality does not significantly degrade when reducing the spectral range to one forth. In this work, a programmable akinetic Insight swept-source is used to balance the spectral scanning range and A-scan rate. This laser has a full spectral range of 1266.4-1352.1 nm and the spectral scanning range can be reduced to improve the A-scan rate. In this work, the laser runs at an A-scan rate of 480 kHz with a spectral scanning range of 1266.4-1289.2 nm. The external clock is constant (400 MHz) and 448 points (valid points: 256) are acquired for each A-scan.
2.2 Scanning protocol
A bidirectional scanning protocol [26] is used in this work (illustrated in Fig. 2), where X- and Y-galvos are driven by a sinusoidal wave (with a frequency of 750 Hz) and a back-and-forth stepped triangle wave respectively. Data acquisition occurs within both forward and backward running of the two galvos, resulting in a full duty cycle. For each B-scan, 320 A-scans are acquired in total and 8 A-scans on both sides are cropped to remove the artefacts caused by the move of Y-galvo, resulting in an effective A-scan number of 304. Each 3D volume 608 B-sans (covering a 3×3 mm2 area) and each position is scanned twice. The odd and even frames are respectively used for vascular images calculation. In this way, two 3D volume datasets can be obtained within one cycle of Y-galvo. The obtained en-face vascular image is smoothed by a median filter with the kernel of 3×3 pixels and then nearest spline interpolation is performed to calibrate the image distortion caused by the non-linear scanning of X-galvo.
2.3 Software framework and data processing
Figure 3(a) illustrates the software framework of en-face GOCTA imaging system. Five parallel threads are created for galvo driving signal generation, data acquisition, data transfer (from host memory to GPU memory), GPU processing, and image display. The function generator outputs the driving signal in the continuous mode. The DAC works in the infinite data streaming mode and a queue is used to store the acquired data in host memory. Meanwhile, the data in the queue is being transferred to a 3D volume buffer in host memory and frame orientation and location are also corrected in this process. Each 4 B-scans are a batch and when the remainder of frame index over 4 is 2 or 3, the frame index in the destination array needs to plus or minus 1. And if the B-scan index is an odd frame, the frame is directly transferred. Otherwise, the frame needs to be flipped first. Once the buffer is full, the 3D volume will be transferred to GPU memory. GPU starts to process when a 3D dataset is transferred to GPU memory. The first step of GPU process is to remove the invalid pixels and flip the even frames. After a new valid dataset is obtained, the data transfer from host memory to device memory is enabled again, indicating that the old raw data can be replaced. When data processing is finished on GPU, the obtained en-face image is transferred back to host memory and stored in a second queue for display. Thread 4 then freezes until the next 3D dataset is ready for processing. Thread 5 flips the even en-face images upside down and then displays them. Thread 5 freezes when the en-face image queue is empty. In this way, the data acquisition, data processing, data reform and transfer, and image display can be running simultaneously, achieving high efficiency.
Fig. 3. The software framework of en-face GOCTA imaging system (a) and the flowchart of data processing on GPU (b).
Figure 3(b) is the flowchart of data processing on GPU. After removing the invalid pixels and reforming the dataset, the next step is to calculate surface profile. An evenly located sparse matrix (9×9 A-scans) of the 3D dataset is taken out first. By performing the Fourier transform on the sparse matrix and locating the maximum pixel value of each A-scan, the surface profile of the sparse matrix can be obtained. The estimated full surface profile for the 3D dataset can then be obtained by linearly interpolating the sparse surface profile. The Gabor-filter-kernels are generated based on the obtained surface profile and the designed imaging depth range [23].
For vascular imaging, bulk motions need to be compensated to achieve high quality vascular images. In this work, because the B-scan rate is high, the transverse shift between adjacent B-scans is negligible. Only the phase shifts are taken into accounts and they are compensated before calculating vascular signals. The bulk motion compensation method is briefly described here. After subtracting the DC component, the interference fringes can be expressed by
3. Results
3.1 Performance of the complex signal detecting unit
We experimentally tested the performance of the complex signal detecting unit with both laser’s full spectral range and the vascular imaging spectral range. The laser worked in a full spectral range scanning mode, in which the valid pixel number for each A-scan is 2816 pixel, and the A-scan rate is 141 kHz. The obtained fringes are also truncated into the vascular imaging spectral range (1266.4-1289.2 nm) for calculation and comparison. Figure 4(a) shows the spectral fringes of both channels. Figure 4(b) and (c) are the point spread functions (PSF) of full spectral range and sub-spectral range, respectively. Gaussian windows are applied to the fringes before FFT and sub-spectral fringes are zero-padded to the same length with the full-spectral-range fringes. We can see that the conjugate signal is suppressed at 36 dB and 31.5 dB respectively. The sub-spectral range PSF achieves a slightly weaker suppression of the conjugate image, the reason is that the PSF itself has a lower signal-to-noise-ratio. An infrared card and a healthy volunteer's finger were also scanned to further validate the performance, and the results are shown in Fig. 4(d) - (k). In comparison of Fig. 4(d) - (k), we can see that the performance of the complex signal detecting unit meets the need for tissue imaging.
Fig. 4. The performance of the complex signal detecting unit. (a) The interference fringes of the PSF detected by Channel A (black) and Channel B (red). (b) PSF of the full spectral range of complex fringes (a). (c) PSF of the sub-spectral range of complex fringes (marked by the dashed rectangle in (a)) and the sub-spectral range is the vascular imaging spectral range (1266.4-1289.2 nm). The obtained structural images of an infrared card with the full spectral range of one-channel fringes (d), full spectral range of complex fringes (e), sub-spectral range of one-channel fringes (f), and sub-spectral range of complex fringes (g). The obtained structural images of human finger with the full spectral range of one-channel fringes (h), full spectral range of complex fringes (i), sub-spectral range of one-channel fringes (j), and sub-spectral range of complex fringes (k). (d)-(k) share the same scale bar.
3.2 In vivo vascular imaging on human skin
With the advantage of real-time en-face imaging, we scanned an area of 7.5×7.5 mm2 on a healthy volunteer's finger and stitched the local vascular maps (in which each local area covers 3×3 mm2). The obtained large area vascular image is shown in Fig. 5. The Gabor filters used here are centered at 270 µm below the surface and have a depth range of 240 µm with tissue refractive index of 1.35 (full width at half maximum, FWHM). A video of real-time en-face vascular imaging was recorded on the local area marked a dashed red rectangle in Fig. 5(b), see Visualization 1. In the video, the vascular images from 3 different depth ranges are shown and the center depths of the Gabor filters are updated in real-time by a slide-bar on the software interface. The Gabor filters of the first 10 frames are centered at 70 µm below surface, the second 9 frames are centered at 170 µm, and the last 7 frames are centered at 270 µm. The depth ranges of the Gabor filters for the three depths are the same.
Fig. 5. (a) The photography of a healthy volunteer's left hand. (b) The mosaic vascular images of a local area (7.5×7.5 mm2) marked by a black rectangle in (a). Visualization 1 shows the real-time en-face vascular imaging on the local area marked by a dashed red rectangle in (b).
3.3 Data processing time analysis
We analyzed the data processing time of different data processing steps and the results are shown in Fig. 6. Here, a desktop (operating system: Windows 10; CPU: i5-7500; RAM: 32 GB; GPU: NVIDIA Geforce GTX 1060) is used for the data acquisition and data processing. CUDA version 10.1 and Visual Studio 2017 are used as the programming platform. The data transfer time for each 3D volume is 21 ms. In our software design, all the frames in each 3D volume dataset are processed simultaneously. The data processing times for different steps are shown in Fig. 6. Since the time for some steps (such as DC subtraction, Gabor filters generation, STD calculation) are too short to show in the graph for the comparison, these steps are combined with the adjacent steps to calculate the data processing time together. We can see that the bulk motion compensation and the skipped convolution are the two steps that cost the most of processing time (89.7 ms and 68.9 ms respectively). The total data processing time for each 3D volume is 173.7 ms while the data acquisition time is 405.3 ms, resulting in a refresh rate of 2.5 Hz for final en-face vascular image display.
4. Discussion and conclusion
Because clinical diagnostics using OCTA images are often made based on en-face microvascular images [2,27–30] and fast visualization is required in a number of scenarios (such as intraoperative imaging) in order to compensate for factors such as patient movement and workflow interruptions, we believe that real-time en-face microvascular imaging could offer improvements to diagnostic efficiency; however, such high-speed swept-source systems would incur significantly higher costs, making it difficult to be implemented in a clinical setting without further proof of diagnostic benefit. In this work, we present a system capable of high-speed en-face microvascular imaging, in which spectral scanning range is sacrificed to improve A-line scan rate, without an increase to the overall hardware costs. Due to GOCTA algorithm is performed to calculate en-face vascular images directly from fringes (without FFT), the cross-sectional vascular images cannot be obtained. However, the imaging depth range can be modified in real-time through an interactive slide bar on the software interface, so the en face vascular images within different depth ranges can be obtained right away (as shown in the Visualization 1). Note that tissue surface profile is used as the reference to calculate the vascular images within a specific depth range, tissue layer segmentation cannot be achieved. Since both the A-scan rate and spectral sweep range are adjustable in our system, if an abnormal vascular distribution is detected by en-face imaging mode and the cross-sectional images (or the accurate tissue layer sectioning) are needed for more detailed analysis, the standard scanning mode (with full spectral range) can be performed. The spectral bandwidth used in the presented en-face GOCTA imaging system is 22.8 nm, giving an axial resolution of only 33.2 µm, but the lateral resolution of 8.6 µm is high enough for capillary vessel detection. Note that the presented system can work with a high-speed swept source with a broader spectral range to overcome the axial resolution degradation, and the current GPU has the capability to accommodate larger datasets because the current data processing time is less than half of the data acquisition time.
The achieved data acquisition time and data processing time for each 3D volume are 405.3 ms and 173.7 ms respectively, so the refresh rate of en-face vascular images is limited by data acquisition. If a faster laser is used, the en-face vascular images refresh rate can be improved to 5.8 Hz at maximum with GTX 1060. If newer video cards (such as the NVIDIA RTX series) or dual video cards setting can be used for data processing, it is possible to reach video rate.
In the software, 3D volume-by-volume based data processing method is performed to exploit the capacity of the video card as much as possible. For current GPU-processing based works, such as Ref. [20–22], B-scan frame-by-frame based method is used to calculate vascular images, where multiple loops are needed to finish each 3D volume data processing. In this way, the total data processing time will be longer than 3D volume-by-volume based method, because CPU needs to be involved for each loop which limits the GPU’s performance. Moreover, only a small portion of video card’s capacity is used with the B-scan frame-by-frame based method, resulting in a waste of hardware capacity.
Compared to Ref. [21] and Ref. [22], the presented system achieved higher en face vascular imaging refresh rate. One reason is that a faster laser and a bidirectional scanning protocol are used. But it is important to note that, except for faster laser, the algorithm GOCTA itself also has speed advantage compared to SSADA and SVOCT. GOCTA calculates blood flow signals in spectral domain directly (without FFT) and the imaging depth range is controlled by skipped convolution which has much lower computing complexity compared to FFT (see Fig. 8 in Ref. [24]). If we only consider the data-processing speed regardless of data acquisition, the data-processing voxel rate can be calculated for quantitative comparison. The achieved data-processing voxel rate (voxel number per B-scan × repeated B-scan number × angiographic B-scan rate) by Ref. [20–22] are 87, 393, and 236 MVoxels/s respectively. The presented system achieves a data-process voxel rate of 545 MVoxels/s. For a fair comparison, the data processing speed of with bulk motion compensation in Ref [22] is used for the comparison. Note that this is just a rough comparison because the hardware speed used in these four studies are different.
Many techniques have been proposed to achieve full range OCT, such as phase modulator [31,32], beam displacement on galvo mirror [33,34], and 3×3 Mach-Zehnder interferometer [35] based methods. For the first two methods, phase modulation is performed along fast scanning direction while beam scanning. Hilbert transform (or FFT and inverse FFT) can then be performed to calculate complex analytical interference fringes for full range imaging, which however increases computational complexity. In the method of using a 3×3 Mach-Zehnder interferometer, only 27 dB of conjugate image suppression can be achieved. In this work, a polarization-modulation based method is used and the orthogonal components of interference fringes can be acquired simultaneously, which is faster than phase-modulation based methods and also achieves higher performance than 3×3 Mach-Zehnder interferometer-based method.
For in-vivo scanning, it is easy for the subject to move out of the effective imaging range in axial direction, therefore, a complex signal detecting unit is used in this work to double the imaging range. Another advantage of this complex detecting unit is that the two orthogonal components are acquired simultaneously. The obtained complex fringes can be then directly used for bulk motion compensation. Note that the presented system can also work with other full-range imaging techniques (such as Ref. [31–34]) or the standard one-channel signal detection, but in this way, FFT or Hilbert transform needs to be performed first to calculate complex signals for the bulk motion compensation.
The presented technique can be easily adapted for retinal imaging by applying the corresponding hardware. However, one issue for retinal imaging might be that more efforts are needed for motion artefact removing because stronger bulk motions may occur during retinal scanning compared to skin scanning. The bulk motion compensation algorithm used here only takes the axial movement into accounts. If transverse movements occur between the two B-scans of the same position, the strip bulk motion noise will show up in the en-face vascular images. In this case, the algorithm presented in Ref. [36] (en-face image-based method) can be performed to remove such noise. For large area scanning, another issue might be eye blink. In this scenario, each obtained local en-face image needs to be evaluated before being used for stitching. If eye blink occurs, the corresponding local area needs to be re-scanned. If the subject’s eye blink frequency is high, the local scanning area size can be decreased to reduce the scanning time of each local region for a higher scanning efficiency. Furthermore, cross-correlation based image-registration algorithm can be performed for automatically stitching the obtained local vascular images.
In summary, we present a real-time en-face Gabor optical coherence tomographic angiography (GOCTA) system for in vivo microvascular imaging on human skin, where a programmable swept source, a complex signal detecting unit, and GPU parallel processing are used. In the software, five parallel threads and two queues are used to improve the efficiency of data transfer among different devices. The achieved refresh rate for en-face microvascular images display is 2.5 Hz. This technique can be easily adapted for retinal vascular imaging. To the best of our knowledge, this study is the first time to present a volumetric data (608×304×256 pixels) acquisition and processing based en-face OCTA imaging system with a refresh rate of 2.5 Hz. We believe that this technique can be useful in clinical applications.
Funding
Natural Sciences and Engineering Research Council of Canada.
Disclosures
The authors declare that there are no conflicts of interest related to this work.
References
1. A. C. S. Tan, G. S. Tan, A. K. Denniston, P. A. Keane, M. Ang, D. Milea, U. Chakravarthy, and C. M. G. Cheung, “An overview of the clinical applications of optical coherence tomography angiography,” Eye 32(2), 262–286 (2018). [CrossRef]
2. K. H. Y. Cheng, A. Mariampillai, K. K. C. Lee, B. Vuong, T. W. H. Luk, J. Ramjist, A. Curtis, H. Jakubovic, P. Kertes, M. Letarte, M. E. Faughnan, V. X. D. Yang, and Brain Vascular Malformation Consortium HHT Investigator Group, “Histogram flow mapping with optical coherence tomography for in vivo skin angiography of hereditary hemorrhagic telangiectasia,” J. Biomed. Opt. 19(8), 086015 (2014). [CrossRef]
3. A. Mariampillai, M. K. K. Leung, M. Jarvi, B. A. Standish, K. Lee, B. C. Wilson, A. Vitkin, and V. X. D. Yang, “Optimized speckle variance OCT imaging of microvasculature,” Opt. Lett. 35(8), 1257–1259 (2010). [CrossRef]
4. C. Chen, K. H. Y. Cheng, R. Jakubovic, J. Jivraj, J. Ramjist, R. Deorajh, W. Gao, E. Barnes, L. Chin, and V. X. D. Yang, “High speed, wide velocity dynamic range Doppler optical coherence tomography (Part V): Optimal utilization of multi-beam scanning for Doppler and speckle variance microvascular imaging,” Opt. Express 25(7), 7761–7777 (2017). [CrossRef]
5. J. Enfield, E. Jonathan, and M. Leahy, “In vivo imaging of the microcirculation of the volar forearm using correlation mapping optical coherence tomography (cmOCT),” Biomed. Opt. Express 2(5), 1184–1193 (2011). [CrossRef]
6. C. Chen, W. Shi, and W. Gao, “Imaginary part-based correlation mapping optical coherence tomography for imaging of blood vessels in vivo,” J. Biomed. Opt. 20(11), 116009 (2015). [CrossRef]
7. Y. Jia, O. Tan, J. Tokayer, B. Potsaid, Y. Wang, J. J. Liu, M. F. Kraus, H. Subhash, J. G. Fujimoto, J. Hornegger, and D. Huang, “Split-spectrum amplitude-decorrelation angiography with optical coherence tomography,” Opt. Express 20(4), 4710–4725 (2012). [CrossRef]
8. W. Shi, W. Gao, C. Chen, and V. X. D. Yang, “Differential standard deviation of log-scale intensity based optical coherence tomography angiography,” J. Biophotonics 10(12), 1597–1606 (2017). [CrossRef]
9. J. Fingler, D. Schwartz, C. Yang, and S. E. Fraser, “Mobility and transverse flow visualization using phase variance contrast with spectral domain optical coherence tomography,” Opt. Express 15(20), 12636–12653 (2007). [CrossRef]
10. W. Shi, C. Chen, C. R. Pasarikovski, W. Gao, and V. X. D. Yang, “Differential phase standard-deviation-based optical coherence tomographic angiography for human retinal imaging in vivo,” Appl. Opt. 58(13), 3401–3409 (2019). [CrossRef]
11. Y. H. Zhao, Z. P. Chen, C. Saxer, Q. M. Shen, S. H. Xiang, J. F. de Boer, and J. S. Nelson, “Doppler standard deviation imaging for clinical monitoring of in vivo human skin blood flow,” Opt. Lett. 25(18), 1358–1360 (2000). [CrossRef]
12. W. Choi, B. Potsaid, V. Jayaraman, B. Baumann, I. Grulkowski, J. J. Liu, C. D. Lu, A. E. Cable, D. Huang, J. S. Duker, and J. G. Fujimoto, “Phase-sensitive swept-source optical coherence tomography imaging of the human retina with a vertical cavity surface-emitting laser light source,” Opt. Lett. 38(3), 338–340 (2013). [CrossRef]
13. B. Vakoc, S. Yun, J. de Boer, G. Tearney, and B. Bouma, “Phase-resolved optical frequency domain imaging,” Opt. Express 13(14), 5483–5493 (2005). [CrossRef]
14. R. K. Wang, S. L. Jacques, Z. Ma, S. Hurst, S. R. Hanson, and A. Gruber, “Three Dimensional Optical Angiography,” Opt. Express 15(7), 4083–4097 (2007). [CrossRef]
15. A. S. Nam, I. Chico-Calero, and B. J. Vakoc, “Complex differential variance algorithm for optical coherence tomography angiography,” Biomed. Opt. Express 5(11), 3822–3832 (2014). [CrossRef]
16. T. Tsai, H. Lee, O. O. Ahsen, K. Liang, M. G. Giacomelli, B. M. Potsaid, Y. K. Tao, V. Jayaraman, M. Figueiredo, Q. Huang, A. E. Cable, J. Fujimoto, and H. Mashimo, “Ultrahigh speed endoscopic optical coherence tomography for gastroenterology,” Biomed. Opt. Express 5(12), 4387–4404 (2014). [CrossRef]
17. Z. Zhi, W. Qin, J. Wang, W. Wei, and R. K. Wang, “4D optical coherence tomography-based micro-angiography achieved by 1.6-MHz FDML swept source,” Opt. Lett. 40(8), 1779–1782 (2015). [CrossRef]
18. R. Poddar, J. V. Migacz, D. M. Schwartz, J. S. Werner, and I. Gorczynska, “Challenges and advantages in wide-field optical coherence tomography angiography imaging of the human retinal and choroidal vasculature at 1.7-MHz A-scan rate,” J. Biomed. Opt. 22(10), 1–14 (2017). [CrossRef]
19. J. V. Migacz, I. Gorczynska, M. Azimipour, R. Jonnal, R. J. Zawadzki, and J. S. Werner, “Megahertz-rate optical coherence tomography angiography improves the contrast of the choriocapillaris and choroid in human retinal imaging,” Biomed. Opt. Express 10(1), 50–65 (2019). [CrossRef]
20. K. K. C. Lee, A. Mariampillai, J. X. Z. Yu, D. W. Cadotte, B. C. Wilson, B. A. Standish, and V. X. D. Yang, “Real-time speckle variance swept-source optical coherence tomography using a graphics processing unit,” Biomed. Opt. Express 3(7), 1557–1564 (2012). [CrossRef]
21. X. Wei, A. Camino, S. Pi, T. T. Hormel, W. Cepurna, D. Huang, J. C. Morrison, and Y. Jia, “Real-time cross-sectional and en face OCT angiography guiding high-quality scan acquisition,” Opt. Lett. 44(6), 1431–1434 (2019). [CrossRef]
22. J. Xu, K. Wong, Y. Jian, and M. V. Sarunic, “Real-time acquisition and display of flow contrast using speckle variance optical coherence tomography in a graphics processing unit,” J. Biomed. Opt. 19(2), 026001 (2014). [CrossRef]
23. C. Chen and V. X. D. Yang, “Gabor optical coherence tomographic angiography (GOCTA) (Part I): human retinal imaging in vivo,” Biomed. Opt. Express 8(12), 5724–5734 (2017). [CrossRef]
24. C. Chen, W. Shi, J. Ramjist, and V. X. D. Yang, “Gabor optical coherence tomographic angiography (GOCTA) (Part II): theoretical basis of sensitivity improvement and optimization for processing speed,” Biomed. Opt. Express 11(1), 227–239 (2020). [CrossRef]
25. B. J. Vakoc, S. H. Yun, G. J. Tearney, and B. E. Bouma, “Elimination of depth degeneracy in optical frequency-domain imaging through polarization-based optical demodulation,” Opt. Lett. 31(3), 362–364 (2006). [CrossRef]
26. M. Ju, M. Heisler, A. Athwal, M. V. Sarunic, and Y. Jian, “Effective bidirectional scanning pattern for optical coherence tomography angiography,” Biomed. Opt. Express 9(5), 2336–2350 (2018). [CrossRef]
27. S. A. Agemy, N. K. Scripsema, C. M. Shah, T. Chui, P. M. Garcia, J. G. Lee, R. C. Gentile, Y. S. Hsiao, Q. Zhou, T. Ko, and R. B. Rosen, “Retinal vascular perfusion density mapping using optical coherence tomography angiography in normals and diabetic retinopathy patients,” Retina 35(11), 2353–2363 (2015). [CrossRef]
28. Q. Zhang, C. S. Lee, J. Chao, C. L. Chen, T. Zhang, U. Sharma, A. Zhang, J. Liu, K. Rezaei, K. L. Peeple, R. Munsen, J. Kinyoun, M. Johnstone, R. N. Van Gelder, and R. K. Wang, “Wide-field optical coherence tomography based microanigography for retinal imaging,” Sci. Rep. 6(1), 22017 (2016). [CrossRef]
29. R. F. Spaide, J. M. Klancnik, and M. J. Cooney, “Retinal vascular layers imaged by fluorescein angiography and optical coherence tomography angiography,” JAMA Ophthalmol. 133(1), 45–50 (2015). [CrossRef]
30. N. Takase, M. Nozaki, A. Kato, H. Ozeki, M. Yoshida, and Y. Ogura, “Enlargement of foveal avascular zone in diabetic eyes evaluated by en face optical coherence tomography angiography,” Retina 35(11), 2377–2383 (2015). [CrossRef]
31. A. F. Fercher, R. Leitgeb, C. K. Hitzenberger, H. Sattmann, and M. Wojtkowski, “Complex spectral interferometry OCT,” Proc. SPIE 3564, 173–178 (1998). [CrossRef]
32. E. Götzinger, M. Pircher, R. A. Leitgeb, and C. K. Hitzenberger, “High speed full range complex spectral domain optical coherence tomography,” Opt. Express 13(2), 583–594 (2005). [CrossRef]
33. B. Baumann, M. Pircher, E. Götzinger, and C. K. Hitzenberger, “Full range complex spectral domain optical coherence tomography without additional phase shifters,” Opt. Express 15(20), 13375–13387 (2007). [CrossRef]
34. L. An and R. K. Wang, “Full range complex ultrahigh sensitive optical microangiography,” Opt. Lett. 36(6), 831–833 (2011). [CrossRef]
35. Y. Mao, S. Sherif, C. Flueraru, and S. Chang, “3×3 Mach-Zehnder interferometer with unbalanced differential detection for full-range swept-source optical coherence tomography,” Appl. Opt. 47(12), 2004–2010 (2008). [CrossRef]
36. P. Zang, G. Liu, M. Zhang, C. Dongye, J. Wang, A. D. Pechauer, T. S. Hwang, D. J. Wilson, D. Huang, D. Li, and Y. Jia, “Automated motion correction using parallel-strip registration for wide-field en face OCT angiogram,” Biomed. Opt. Express 7(7), 2823–2836 (2016). [CrossRef]
