Adaptive dynamic analysis-based optical coherence tomography angiography for blood vessel projection artifact suppression

Abstract

Optical coherence tomography angiography (OCTA) for blood vessel 3-D structure imaging suffers from blood vessel projection artifacts/tail artifacts when using a long decorrelation time (e.g., repeat B-scan acquisition in regular OCTA) or loss of micro vessel signal when using a short decorrelation time. In this work, we developed an adaptive first-order field autocorrelation function (g₁) analysis-based technique to suppress the projection artifacts under macro vessels while enhancing the dynamic signal of micro vessels. The proposed method is based on the differences of the decorrelation rate and the phase variations of g₁ between the vessel voxels and the artifacts regions. A short or long decorrelation time was applied to obtain the dynamic index of the projection artifacts region or the blood vessel region, respectively. Compared to the slab subtraction-based post-image processing-based techniques, the proposed approach addresses this problem on a physical basis and shows the ability to suppress the projection artifacts while enhancing the detection of the micro vessels.

1. Introduction

Optical coherence tomography angiography (OCTA) has become an important tool for studying the micro-vascular structural and dynamic changes [1,2]. OCTA obtains the blood flow information by comparing the signal changes at the same spatial location with the fact that the signals within blood vessels are changing over time due to the flow of red blood cells (RBCs), while the signal reflected from tissue are more stable. Thus, by detecting the changes of OCT amplitude/intensity, phase, or complex signals [3–8], OCTA is able to image the blood vessel with high spatial and temporal resolution. It has been widely used in ophthalmology [5,7,9], dermatology [10,11] and neurosciences [12–14].

Essentially, OCTA imaging of blood vessels is based on the detection of dynamic contrasts between the moving RBCs and the static tissues. As the main scatter particles in the blood vessels, RBCs have very high scattering anisotropy, leading to strong forward multiple scattering [15]. Photons carrying dynamic signals after hitting the moving RBCs may experience single-backward scattering, multiple-forward and backward scattering, and even transmit through the blood vessel and then be reflected back after hitting the tissues underneath the blood vessels. In the last two scenarios, the light paths of the multiple scattered photons are longer than that of the directly reflected photons (Fig. 2(A)). Hence the dynamic signal from photons hit with the moving RBCs may extend to regions underneath the bottom layer of the vessel, resulting in the long-standing challenge of the blood vessel projection artifacts, which is also called tail artifacts, in OCTA imaging. The blood vessel projection artifacts not only extend the blood vessel lumen in the axial direction but also bury the micro vessels’ signal in the projection artifacts background.

Fig. 2.

(A) Photon-RBC interaction (solid lines) and the equivalent light path (dashed lines). Revised from [23]. (B) Blood vessel projection artifacts and typical ${\mathrm{g}}_1(\tau )$ decorrelation from different regions (C1). V. stands for vessel, Trans. stands for transverse. (C2) Distinct phase changes were observed between the micro vessel with axial flow component (green) and artifacts (red) in the tail region.

Several hardware-based and post image processing-based approaches have been proposed to suppress the blood vessel projection artifacts in OCTA [16–21]. Leahy et al. applied a high NA objective to suppress the blood vessel projection artifacts by reducing multiple scattering signal for the out-of-focus region at the expense of a reduced depth of field [17]. Woo et al. calculated the mean intensity of all pixels in each individual A-lines and subtracted it from original image [16]. Zhang et al. used a subtraction method to minimize projection artifacts in OCTA for retinal vascular imaging [19]. Zhang et al. introduced an algorithm to remove the projection artifacts by resolving the ambiguity between in situ and projected flow signals [20]. Wang et al. introduced a reflectance-based projection resolved OCTA algorithm, which uses OCT reflectance to enhance the flow signal and suppress the projection artifacts in 3- dimensional OCTA [21]. Stefan et al. introduced a deep learning method for vessel visualization, enhancement and segmentation, which also suppressed the projection artifacts [18]. We previously proposed a ${\mathrm{g}}_1$ -OCTA method which used a short decorrelation to suppress the blood vessel projection artifacts [22]. The ${\mathrm{g}}_1$ -OCTA method had a limitation in balancing the goals of suppressing blood vessel projection artifacts and enhancing signals from slowly moving capillary networks. The blood vessel projection artifacts became stronger when using a long decorrelation time, while the micro vessel signals became weaker when using a short decorrelation time.

In this work, we developed an approach to adaptively determine the decorrelation time ( $\mathrm{A}{\mathrm{g}}_1$ -OCTA) so that a short decorrelation time can be applied to the projection artifacts voxels while a longer decorrelation time can be applied to the vessel voxels. Results suggest that our method can greatly suppress the projection artifacts under macro vessels (large pial vein and large pial artery) and enhance the dynamic signal of micro vessels (arterials, venules, and capillaries). In the following sections, we will first illustrate the origination of the projection artifacts in the OCTA image and the features of the ${\mathrm{g}}_1$ decorrelation in the macro-, micro- vessels, and the projection artifacts regions; then describe proposed algorithm for $\mathrm{A}{\mathrm{g}}_1$ -OCTA data processing. In the results section, we will first show the ability of the proposed approach to maintain the micro vessel signal in the tail region, then compare the results with the long-, short- decorrelation time calculation and the proposed $\mathrm{A}{\mathrm{g}}_1$ -OCTA. At last, we will compare the projection artifacts suppression between the proposed $A{\mathrm{g}}_1$ -OCTA with the slab subtraction-based post image-processing method that performs the data processing based on the regular OCTA image that was obtained with repeated B-scan acquisition.

2. Materials and methods

2.1. Experimental set-up and scanning protocol

Our homemade OCT system (see Fig. 1 A) used a dual-SLD light source with a center wavelength of 1550 nm and a bandwidth of 210 nm. The spatial resolution in the brain of the system was 3.7 $\mu \mathrm{m}$ in axial and 3.5 $\mu \mathrm{m}$ in lateral using a 10 $\times$ objective. In this study, we employed an M-scan mode for data acquisition that 150 repeated A-lines at 60 kHz were acquired at one spatial location and then moved to the next transverse scan position. The full image was acquired with 400 $\times$ 400 transverse positions, thus the data is consisted of 150 × 400 × 400 A-lines. Each imaging session was around 6 minutes. The data was processed offline with MATLAB 2020b (MathWorks, Natick, MA).

Fig. 1.

(A) Experimental setup. (B) M-mode data acquisition with repeated A-lines.

2.2. Animal preparation

We acquired in vivo data from 5 female C57BL/6 mice prepared with a 3 mm diameter chronic cranial window. The surgeries were performed two weeks before the experiments. During the imaging session, the animals were anesthetized with 1-2% isoflurane and a region of interest (ROI) with 600 $\mu \mathrm{m}\times$ 600 $\mu \mathrm{m}$ (400 $\times$ 400 pixels in X and Y) area was acquired with our SD-OCT system. The animal experiment protocol was approved and supervised by the Animal Ethics Committee at the Southern University of Science and Technology.

2.3. $A{g}_1$ -OCTA for blood vessel projection artifacts suppression

We first obtained the autocorrelation function ${\mathrm{g}}_1(\tau )$ from the OCT field signal for each voxel:

$${\mathbf{g}}_1(\mathit{\tau })=\frac{⟨{\mathbf{R}}^{\ast }(\mathbf{t})\mathbf{R}{(\mathbf{t}+\mathit{\tau })}_{\mathbf{t}}⟩}{⟨{\mathbf{R}}^{\ast }(\mathbf{t})\mathbf{R}{(\mathbf{t})}_{\mathbf{t}}⟩}$$ (1)

where $\mathrm{R}(\mathrm{t})$ is the complex OCT signal at time t, ${\mathrm{R}}^{\ast }(\mathrm{t})$ is the complex conjugate, $\tau$ is the time lag, and $⟨⟩$ represents ensemble averaging. [19]

Then, we defined the dynamic contrast index ${\mathrm{I}}_{\mathrm{d}}$ as the maximum decorrelation during the defined decorrelation time of the ${\mathrm{g}}_1(\tau )$ after the first time lag:

$${{\mathbf{I}}_{\mathbf{d}}}_{(1:{\mathbf{n}}_{\mathit{\tau }})}=\mid {\mathbf{g}}_1(1)\mid -\mathbf{\text{min}}\mid {\mathbf{g}}_1{(\tau )}_{(1:{\mathbf{n}}_{\mathit{\tau }})}\mid$$ (2)

where || denotes magnitude, ${\mathrm{n}}_{\tau }$ is the autocorrelation time lag, and $\text{min}\mid {\mathrm{g}}_1{(\tau )}_{(1:{\mathrm{n}}_{\tau })}|$ denotes the minimum value of ${\mathrm{g}}_1(\tau )$ between the time lags of 1 to ${\mathrm{n}}_{\tau }$ . Usually, with a larger ${\mathrm{n}}_{\tau }$ the ${\mathrm{g}}_1{(\tau )}_{(1:{\mathrm{n}}_{\tau })}$ decays to a lower level thus resulting in a larger ${\mathrm{I}}_{\mathrm{d}}$ . The goal of $\mathrm{A}{\mathrm{g}}_1$ -OCTA is to adaptively select ${\mathrm{n}}_{\tau }$ for ${\mathrm{I}}_{\mathrm{d}}$ calculation, i.e. using a short ${\mathrm{n}}_{\tau }$ for the tail artifact regions and a long ${\mathrm{n}}_{\tau }$ for the blood vessel voxels. So, the key is to distinguish the vessel voxels from the projection artifacts. We noted that the ${\mathrm{g}}_1$ decays differently for voxels within the vessels and in the projection artifacts regions.

Firstly, we observed that for macro vessels (black dot in Fig. 2(B)) the blood flows at a high-speed which causes a fast decorrelation (black curve in Fig. 2(C1)); for micro vessels (blue and green curves in Fig. 2(C1)) the decorrelation is slower than that in the macro vessels but usually faster compared to the projection artifacts region (red curve in Fig. 2(C1)). However, when the decorrelation time is beyond a few milliseconds, the ${\mathrm{g}}_1$ of the projection artifacts region would decay to a similar level to that of the micro vessels, resulting in a non-negligible tail region beneath the macro vessels. This is the case of regular OCTA which employs B-scan repeats to detect the dynamic contrast and the time interval is usually longer than 5 ms.

Secondly, by comparing the imaginary signal of ${\mathrm{g}}_1$ between the artifacts and micro vessels with an axial flow angle, we further noted that the micro vessel’s signal has a clear phase change compared to the artifacts, as shown in Fig. 2(C2). This is due to the fact that the signal from axial flowing micro vessel is mainly from directional and transitional flow of RBCs, which results in a constant frequency shift or broadening. In contrast, the signal from the projection artifacts is a mix of various multiple scattered dynamic signals from the flowing red blood cells and the static signal reflected by the underneath tissue, which is featured with a random dynamic signal plus a static offset.

With these features, we propose a ${\mathrm{g}}_1$ analysis method to determine the blood vessel from the projection artifacts region, and then adaptively applied a shorter decorrelation time to suppress the projection artifacts and a longer decorrelation time to enhance the blood flow signal.

Figure 3 illustrates the adaptive dynamic approach. Firstly, based on the fact that the projection artifacts primarily appear under big pial blood vessels, we determined the macro vessel regions from the maximum intensity projection (MIP) image calculated with long ${\mathrm{n}}_{\tau }$ , as shown in Fig. 3(A). We used the binary map (Fig. 3(A1)) of blood vessels to obtain a vessel diameter encoded skeletonized vessel map (Fig. 3(A2)). Then the micro vessels were removed if the diameter is less than 25 $\mu \mathrm{m}$ , which was determined empirically. Finally, the transverse coordinates of the big vessels were obtained as shown in Fig. 3(A3), and the projection artifacts suppression and micro flow signal enhancement were performed in the axial projection regions of the big vessels.

Fig. 3.

$A{g}_1$ -OCTA data processing flowchart. (A) Step 1: identify the large blood vessels; (A1) -(A3): Stepwise results to obtain the large vessel binary map. (B) Step 2: determine macro vessel and the tail regions; Step 3: use the criteria of | ${\mathrm{g}}_1(\tau )$ | decorrelation rate to suppress artifacts in the tail region; (C)Step 4: identify the micro vessels from the tail region by the imaginary part of ${\mathrm{g}}_1(\tau )$ criteria; Step 5: apply long decorrelation time to get enhanced blood vessel signal and short decorrelation time to suppress the projection artifacts. Scale bar: 100um. (D1) Representative decorrelation of the imaginary part of ${\mathrm{g}}_1(\tau )$ for vessel voxels (green) and artifact voxel (red); (D2) Histogram distribution shows the maximum change of the imaginary part of ${\mathrm{g}}_1(\tau )$ of the artifacts voxels (red) and the vessel voxels (green). Note: different bin widths were used to draw the histograms.

The second step is to identify the macro vessel lumen area by using the axial profile analysis of the dynamic index ( ${\mathrm{I}}_{\mathrm{d}}$ ) in combination with the ${\mathrm{g}}_1(\tau )$ decorrelation. Similar to the approach used by Zhang [20], we first determined the blood vessel lumen by finding the axial profile peak and the width to peak from the upper boundary of the vessel, then examined the ${\mathrm{g}}_1(\tau )$ decorrelation which shall have similar decay rate.

The third step is to obtain the micro vessel voxels in the tail region from the artifacts background. The signal from the projection artifacts region is a mix of the dynamic signal due to multiple scattering of photons with the moving RBCs and the static signal that reflected from the tissue. Thus, the ${\mathrm{g}}_1(\tau )$ decorrelation in the projection artifacts region would slowly decay and not fully decorrelate to 0, while the signal within the vessel decays faster and would decay to a much lower level. With this feature, we first used a ${\mathrm{g}}_1(\tau )$ decorrelation threshold of 0.5 within the first 1 ms (Fig. 3(B)). Then, we applied the feature that for micro vessels the imaginary part of ${\mathrm{g}}_1(\tau )$ has a constant phase change that featured with a larger amplitude change (Fig. 3(D)) to determine the micro vessel voxels. Specifically, we applied an imaginary ${\mathrm{g}}_1$ criteria of the maximum absolute amplitude change within the first 1 ms greater than the threshold of 0.1 to recover the vessel voxels from the artifacts region (Fig. 3(C)).

Finally, to suppress the projection artifacts and enhance blood vessel signal, we applied a short decorrelation time for the dynamic contrast ${\mathrm{I}}_{\mathrm{d}}$ calculation in the artifact region, and a long decorrelation for the identified vessel voxels (Fig. 3(C)).

3. Results

We first tested the ability of the Ag1-OCTA technique to recover the micro vessel signals from the projection artifacts background, as shown in Fig. 4. Figure 4(A) shows a macro pial vessel and its tail region. We see that strong projection artifacts extend into the deep regions and the underneath micro vessel signals are merely identifiable. The ${\mathrm{g}}_1(\tau )$ decorrelation curves of the macro vessel (black dot along the red line in Fig. 4(A)), an artifact voxel (red dot along the red line in Fig. 4(A)), and a micro vessel (green dot along the red line in Fig. 4(A)) decay at different rates, as shown in Fig. 4(B1). We also see that the artifacts voxels decays slower in deeper layers, while the micro vessels have similar decorrelation rate in the upper and lower layers, as indicated by the solide and dashed curves in Fig. 4(B1). If only apply the decorrelation rate threshold ( ${\mathrm{g}}_{1\text{Thd}}$ ) to distinguish the artifacts and flow voxels, we would obtain a result with many broken-flow-micro vessels in the tail region as indicated by the green arrows in Fig. 4(B3). The phase change differences between the artifacts and the blood flows can further help us to recover the micro vessel voxels. As shown in Fig. 4(C1), the imaginary signal change of an artifacts voxel (red dot of the en face cross section image in Fig. 4(A)) and a micro vessel voxel (green dot of the en face cross section image in Fig. 4(A)) shows that the micro vessel has a dominant frequency change with a much larger phase variation. With this feature, we can recover the micro vessel voxels in the tail regions, as shown in Fig. 4(C3).

Fig. 4.

The phase information from ${\mathrm{g}}_1(\tau )$ can be used to recover the micro vessel signal after applied with the decay rate threshold criteria. (A) X-Z view of a macro pial vessel and its tail region. The en face cross sectional images show representative ${\mathrm{g}}_1(\tau )$ decorrelation of micro vessels and projection artifacts voxels at different depths. (B1) The magnitude of ${\mathrm{g}}_1(\tau )$ decorrelation of the black, red, and green voxels along the red line in (A). (B2&B3) Results were obtained by applying the decorrelation rate of ${\mathrm{g}}_{1\text{Thd}}$ only. (C1) The imaginary signal of ${\mathrm{g}}_1(\tau )$ of an artifact voxel and a micro vessel voxel marked by the red and green dots in the en face image in (A), respectively. (C2&C3) Micro vessel voxels in the tail region can be recovered with the ${\mathrm{g}}_1(\tau )$ imaginary information.

We then compared the results calculated with long-, short-, and the proposed adaptively determined- decorrelation time, as shown in Fig. 5. We see that at the top layer (Fig. 5(A1-C1)) the long and adaptive methods obtained similar results, while the short decorrelation-based method was unable to image many micro vessels and it has much weaker signal intensity at slow flowing macro vessels. In the middle layer (Fig. 5(A2-C2)), the long decorrelation-based calculation has strong tail signal, while both the short decorrelation-based and the proposed adaptive methods can greatly suppress the projection artifacts beneath the macro vessels (red arrows). However, in deeper layers (Fig. 5(A3-C3)) the short decorrelation-based method shows very weak signal in micro vessel regions, whereas the adaptive method shows strong signal which is comparable to that obtained with long decorrelation calculation (blue arrows), suggesting that the proposed method can not only suppress projection artifacts but also preserve blood flow signals of micro vessels.

Fig. 5.

Comparison of the long-, short-, and adaptively determined- decorrelation time obtained g1-OCTA results. En face MIPs obtained with long (A1-A3), short (B1-B3), and the proposed adaptively determined decorrelation time (C1-C3), respectively. (D1-D3) XZ-view stacks (MIP over 25 $\mu m$ in Y) marked by red shaded region in A1. (E1) Overlapped OCTA B-scan image of D1(gray background) and D3 (cyan color). Axial profiles obtained with the three methods at the dashed yellow line. Yellow arrows indicate preserved micro vessels under macro vessels. Red hollow arrows on the axial profile indicates suppressed projection artifacts region. (E2) Correlation coefficient of the en face maximum intensity projection (∼16 um thickness) at different depth with respect to pial macro vessel map in the upper layers.

Figure 5(D&E1) further compare the axial cross-sectional results obtained with these three methods. We see that the proposed adaptive method (Fig. 5(D3)) can greatly suppress the blood vessel projection artifacts under macro vessels, enhance signals in slowly flowing vessels (green arrows & red circles), and extract micro vessel signals buried in the projection artifacts background (yellow arrow, Fig. 5(E1)). The overlapped color-encoded B-scan image (Fig. 5(E1)) shows that a large amount of projection artifacts background is suppressed with the proposed adaptive method while micro vessels are preserved. The axial profiles in Fig. 5(E1) shows that the long decorrelation-based calculation has strong tail signal, while both the short decorrelation-based and the proposed adaptive methods can greatly suppress the projection artifacts beneath the macro vessels (red hollow arrows). The micro vessel signal obtained with the proposed method has similar signal intensity to the long decorrelation time-based method but with a suppressed projection artifacts background. Figure 5(E2) further compares the correlation coefficient of the en face MIP at different depth layers (stack thickness ∼16 $\mu$ m) related to the upper layers. We see that the correlation coefficients of all approaches slowly decrease to 0, indicating the reducing similarity between the upper layers and the deeper layers. Compared to the long $n_{\tau }$ ( $n_{\tau }$ =50) calculation, $A{g}_1$ -OCTA has a lower correlation coefficient in the middle depth layers, suggesting that $A{g}_1$ -OCTA can better suppress the projection artifacts. Compared to the short $n_{\tau }$ ( $n_{\tau }$ =5), $A{g}_1$ -OCTA has a larger value in the depth of 50-100 $\mu$ m, which suggests that some signals in the tail region were preserved with the $A{g}_1$ -OCTA. This may include two types of signals: some strong projection artifacts next to the bottom of pial macro vessel and recovered micro vessels in the axial projection region.

We further compared the proposed method with one of the representative post image processing-based methods, the slab-subtraction approach (SS-OCTA) [24]. For a fair comparison and a better visual assessment, we first normalized the results so that the mean signal intensity of the pial macro vessels is the same of ‘1’ for all three methods, and then adjusted the signal intensity in the deeper layers accordingly.

Compared to the regular OCTA (Fig. 6(A)), it shows that the SS-OCTA (Fig. 6(B)) removes the blood vessel projection artifacts and the proposed approach (Fig. 6(C)) does a good job in suppressing the projection artifacts. In the deeper layers (Middle MIP: 133-363 $\mu \mathrm{m}$ ), we noted that the SS-OCTA artefactually produces many disconnected micro vessel segments in the macro vessel tail regions. In contrast, $\mathrm{A}{\mathrm{g}}_1$ -OCTA can not only suppress the projection artifacts but also preserved the micro vessel signal in the tail regions, as indicated by the green arrows. The bottom row of Fig. 6 compares the coronal cross-sectional MIP of the three methods. Clearly, the regular OCTA has the strongest projection artifacts, and the SS-OCTA removes almost all signals beneath the macro vessels. In contrast, the proposed A ${\mathrm{g}}_1$ -OCTA can preserve and enhance the micro vessel signals while suppressing the projection artifacts.

Fig. 6.

Comparison with the regular OCTA (A), the slab-subtraction based method (B), and the proposed $A{g}_1$ -OCTA (C) at two depth layers. (D1-D3): corresponding XZ stacks (MIP over 66 $\mu m$ in Y) marked by blue shaded region in figure A. The green circles show the ability of suppressing projection artifacts. Scalebar: 100 $\mu m$ . (E1): blood vessel density at different depth; mean $\pm$ std, n = 3; (E2) Statistic comparison of the correlation coefficient of the en face maximum intensity projection (∼16 um thickness) at different depth with respect to the macro vessel map in upper layers, n = 3, mean $\pm$ std.

We further calculated the vessel density (Fig. 6(E1)) for each MIP stack with a thickness of ∼16 $\mu m$ , and the correlation coefficient [20] (Fig. 6(E2)) for each stack with respect to the macro vessel distribution in the upper layers. From Fig. 6(E1) we see that in the layers beneath the macro pial vessels (depth 100 $\mu m$ - 250 $\mu m$ ), both the SS-OCTA and $A{g}_1$ -OCTA have a decreasing vessel density, suggesting the projection artifacts suppression ability. The SS-OCTA has the lowest vessel density in the deeper layers is due to the fact that it removes all signals beneath the macro pial vessels, including the underneath micro vessel pixels. Figure 6(E2) compares the correlation coefficient related to the upper layers. We see that $A{g}_1$ -OCTA has a lower correlation coefficient in the middle layers compared to the regular OCTA and slowly decreases to 0, indicating that the similarity between the upper layers and the deeper layers is reducing, suggesting the ability of $A{g}_1$ -OCTA to suppress the projection artifacts. The negative value of SS-OCTA is due to the fact that it removes most signals in the macro vessel tail region, leaving a shadow in the deep layers.

Worth noting that, we do notice that for some slowly flowing transverse vessel segments, the A ${\mathrm{g}}_1$ -OCTA is not able to obtain the flow signal compared to the regular OCTA, as indicated by the red arrows in Fig. 6. This is due to the fact that $\mathrm{A}{\mathrm{g}}_1$ -OCTA calculates a much shorter period of decorrelation time (∼1 ms) compared to the regular OCTA (>5 ms). In this case, it may not detect the RBC signal due to the discontinuity, single-passage file, and even the RBC stalling effect [25] in the capillaries. In addition, the regular OCTA is usually obtained with 10 times of averaging while the A ${\mathrm{g}}_1$ -OCTA doesn’t do averaging. Increasing the decorrelation time for ${\mathrm{g}}_1$ calculation can greatly improve the micro vessel signal and its continuity, but at the expense of a longer acquisition time.

4. Discussion

To summarize, in this work we show that the proposed $\mathrm{A}{\mathrm{g}}_1$ -OCTA has the ability to suppress the blood vessel projection artifacts and enhance the detectability of the micro vessels in the tail regions. This method first identified the macro vessels and the underneath tail regions, then both the magnitude and phase differences of the ${\mathrm{g}}_1(\tau )$ decorrelation were analyzed in the tail region to differentiate the vessel voxels from the artifact voxels. Finally, a short decorrelation time was applied to suppress the projection artifacts signal, while a longer decorrelation time was used to enhance the dynamic contrast of the blood flow signal.

For the regular OCTA, the time interval between the repeated B-scans at the same location is usually greater than 5 ms, which is too long that the artifacts signal would decay to a level comparable to the micro vessels leading to the projection artifacts issue in the axial direction. To address this problem, we previously proposed a ${\mathrm{g}}_1$ -OCTA method [22] which employs repeated A-line acquisition to obtain high temporal resolution data and then by applying a short decorrelation time to suppress the projection artifacts signal. However, this method would also suppress the blood flow signals, leading to deteriorated detection of blood flows, especially for the slowly flowing micro vessels. In this work, we further investigated both the magnitude and phase changes of the ${\mathrm{g}}_1(\tau )$ decorrelation and proposed the A ${\mathrm{g}}_1$ -OCTA method to differentiate the flow signal from projection artifacts. The results suggest that the A ${\mathrm{g}}_1$ -OCTA can not only suppress the blood vessel projection artifacts but also enhance the flow signal and further recover the micro vessel voxels from the projection artifacts background. Compared to the slab subtraction-based post image processing method to remove the macro vessel’s projection artifacts which results in broken flows in the tail region, the A ${\mathrm{g}}_1$ -OCTA addresses this problem from the physical basis and maintains a good connectivity of the flow of the micro vessel network.

Noting that we found it’s challenging to detect the slowly flowing transverse vessel voxels. This is due to the fact that the OCT signals from micro vessels are usually composed with the slow flowing signal from moving RBCs and static signal from vessel walls, resulting in a quite similar ${\mathrm{g}}_1(\tau )$ decorrelation to the projection artifacts regions. Further, for transverse flowing vessels, the ${\mathrm{g}}_1(\tau )$ doesn’t have a dominant phase change of the imaginary ${\mathrm{g}}_1(\tau )$ like the vessels with an axial velocity component do. Thus, the voxels of slowly flowing transverse micro vessels may be treated as projection artifacts region and the signal would be suppressed with A ${\mathrm{g}}_1$ -OCTA. However, from our observation, most micro vessels close to the macro vessels have an axial velocity component, enabling the differentiation from the artifact region. Further, in deeper layers, the projection artifacts are weaker and usually decay slower compared to flow signal, giving $\mathrm{A}{\mathrm{g}}_1$ -OCTA the ability to different vessel voxel from artifacts region and enhance the detection of micro vessels in deep layers. To further enhance the detection of those slowly flowing micro vessels, the idea proposed in the VISTA method [26]can be applied to combine the regular OCTA (longer interval/observation time) for the micro vessel detection in regions outside of the macro vessel projection area, and with the A ${\mathrm{g}}_1$ -OCTA method (shorter interval/observation time) to suppress the projection artifacts of macro vessels while enhancing the detection of micro vessels in the tail region.

It is worth to mention that the A ${\mathrm{g}}_1$ -OCTA suffers from a major limitation of a long data acquisition time to obtain an angiography image compared to the regular OCTA. In principle, a regular OCTA image can be obtained with a twice-Bscan repeat acquisition, which may just take a few seconds, while it may take 2-4 mins to get an A ${\mathrm{g}}_1$ -OCTA image. Although as we demonstrated in our previous publication that the observation time ${\mathrm{n}}_{\mathrm{t}}$ can be reduced to 50 [22], it will be still challenging to apply A ${\mathrm{g}}_1$ -OCTA for clinic studies, particularly the ophthalmology exam. However, as demonstrated in this work the A ${\mathrm{g}}_1$ -OCTA will be a powerful tool for the micro circulation study in animal models, where the relatively longer acquisition time would not be an issue. We expect the A ${\mathrm{g}}_1$ -OCTA will play an import role in the cerebral micro circulation-related mechanism and disease studies.

Funding

National Natural Science Foundation of China10.13039/501100001809 (62275116, 62220106006); the Shenzhen Science and Technology Innovation Program10.13039/501100017610 (20210316161406001); the Guangdong Science and Technology Department10.13039/501100007162 (2022A1515011984); and the Guangdong Provincial Key Laboratory of Advanced Biomaterials (2022B1212010003); Special Funds for the Cultivation of Guangdong College Students’ Scientific and Technological Innovation (pdjh2022c0086).

Disclosures

The authors declare no conflicts of interest.

Data Availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.