J. Biophotonics 1–9 (2016) / DOI 10.1002/jbio.

201600194

FULL ARTICLE
Tail artifact removal in OCT angiography images
of rodent cortex
Utku Baran1, 2, Woo June Choi1, Yuandong Li1, and Ruikang K. Wang*, 1
1 Department of Bioengineering, University of Washington, Seattle, WA, USA
2 Department of Electrical Engineering, University of Washington, Seattle, WA, USA

Received 5 July 2016, revised 16 August 2016, accepted 21 August 2016

Published online 8 September 2016

Key words: optical coherence tomography, OCT angiography, artifact removal, stroke

Optical coherence tomography angiography (OCTA) is a

surging non-invasive, label-free, in vivo volumetric ima-
ging method, currently being translated to clinical
ophthalmology and becoming popular in neuroscience.
Despite its attractiveness, there is an inherent issue of
using OCT angiograms for quantitative cerebrovascular
studies: The dynamic scattering of moving erythrocytes
within pial vasculature creates tail-like artifacts that sha-
dow the capillary vessels in the deeper layers of cortex.
This false flow effect is relatively benign for qualitative
visualization purposes, but it might have a significant im-
pact on quantitative interpretation of angiographic re-
sults. In this work, we propose a simple image processing
method to remove these tail artifacts in depth-resolved
OCTA images using an adaptive enface mask generated
with OCT structural images. We demonstrate the effec-
tiveness of our method by comparing vessel densities and
vessel similarities of depth-resolved OCT angiograms in a
stroke study in a rodent model, in vivo. Thanks to the
ability of seeing through the tails of pial vessels, capillary
vessels beneath these vessels could be recovered to some
extend in the deeper layers of mouse cerebral cortex,
leading to a more accurate quantification.
Tail artifact removed enface OCT angiogram of deeper
layer in vivo mouse cortex.

1. Introduction ing techniques, such as laser speckle contrast imag-

ing [2], Doppler ultrasound [3], and functional mag-
Traditionally, histological analysis has been used to netic resonance imaging [4] also have been used to
study microvascular networks [1], which is totally in- study microvascular changes in many physiological
vasive and may not well represent the vascular mor- states. However, the limited spatial and/or temporal
phology in vivo. Popular non-invasive label-free imag- resolution prevent these techniques from imaging

* Corresponding author: e-mail: [email protected]

© 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

2 U. Baran et al.: Tail artifact removal in OCT angiography images of rodent cortex

small blood vessels such as capillaries. On the other effectively suppressing the projection artifacts on
hand, two-photon microscopy [5] allows high-resolu- both enface and cross-sectional angiograms. Briefly,
tion imaging of neural and microvascular activities in this algorithm first detects the peaks in the A-lines
rodent models, but it is usually limited with speed, of OCTA, upon which to keep the decorrelation val-
field of view, penetration depth, and the necessity of ues at the successive peak positions (vessels), and
contrasting agents. then set the rest to zero. One drawback of this meth-
Optical coherence tomography (OCT), a well-es- od is that it heavily relies on an accurate peak detec-
tablished technique in biomedical imaging, enables tion. For ophthalmology applications, this may be a
volumetric morphological visualization of tissue mi- reasonable requirement where: 1) Most of the ves-
crostructures in vivo with a micrometer-scale image sels in retinal layers have similar diameters so that
resolution [6]. OCT angiography (OCTA), a popular the peaks are relatively easier to detect in an A-line,
extension of OCT, is able to image blood perfusion and 2) Retinal tissue is relatively transparent. How-
in the functional vessels down to capillary level, in ever, this algorithm is not amendable to tissues with
vivo [7, 8]. Recent studies have validated the accu- relatively high scattering properties, e.g. brain tissue.
racy of OCTA for in vivo blood flow imaging with In addition, it does not work well to recover small
multi-photon microscopy on a mouse cerebrovascu- vessels, such as capillaries that are located beneath
lar model [9] and with laser speckle contrast imaging larger superficial vessels [17]. Indeed, for the other
on a human skin microvascular model [10]. OCTA popular in vivo imaging subjects like human skin or
has been used to study the microvasculature of a mouse brain, the most of the capillaries are signifi-
variety of tissues in vivo, including healthy and dis- cantly smaller than the big superficial vessels, hence
eased human skin [11], human retina [12], and an accurate peak detection is challenging due to
mouse cerebral microvasculature [13]. their relatively small signal intensities compared to
Although OCTA is an attractive volumetric mi- tail artifacts.
crovascular imaging method, 3D visualization of mi- Here, we propose a simple post-processing meth-
crovasculature is typically difficult to interpret, which od to provide depth-resolved OCT angiography
in part drives most researchers to use enface maxi- images with tail artifacts minimized, allowing accu-
mum intensity projection (MIP) images of OCTA rate visualization and quantification of cortical vas-
for their studies. It is mostly because of the tail-like culature in brain tissue. We first enhance the effec-
artifacts that appear in the deeper layers coming tive signal-to-noise ratio (SNR) in OCTA cross-sec-
from the strong scattering property of erythrocytes tional images, then segment the
300 μm of OCT
within overlying vasculature. Such tails appear as and OCTA data into 10 depth-layers using an auto-
elongated vessels in the axial direction and cause mated segmentation method [18]. Finally, we utilize
overlap between superficial and deep layers [14]. the enface OCT structural images from the segment-
Since signal strength of the tail artifact is usually ed layers to create an adaptive mask that is used for
much stronger or comparable to the scattered signals suppressing the tail artifacts from enface OCTA
at the deep vessels, the flow signals of capillaries in images of the deeper cortex layers. This method is
the cortex subsequent to the pial vessels are over- applied to OCTA images of healthy and ischemic
lapped or hidden by the tail artifacts. This adverse mouse brain in vivo and its capability is demon-
effect is relatively benign for the visualization of strated with several quantitative analyses. This post-
overall vessel network from the OCT angiogram. processing method can also be adapted to data
However, it has a significant impact on quantitative acquired using common OCTA systems for various
interpretation of angiographic results. applications.
A few post-processing methods have been pro-
posed to tackle this problem in the field of ophthal-
mic imaging. The most of these methods rely on a
slab-subtraction algorithm. In these methods, the 2. System and methods
vascular pattern of superficial microvasculature is
subtracted from an enface MIP image of a deeper
layer [15, 16]. Unfortunately, this method replaces 2.1 Animal model
the tail artifact with a shadow cast (area with ab-
sence of blood vessels), leaving apparent gaps within All experimental animal procedures performed in
microvascular plexus that are difficult to reconstruct. this study are approved by the Institute of Animal
In another work, Zhang et al. developed a pro- care and Use Committee (IACUC) of the University
jection-resolved OCTA algorithm to improve the of Washington (Protocol number: 4262-01). Male 12-
shadow cast problem that occurs when using the pre- week-old C57/BL6 mice weighing between 23 g to
vious tail artifact removal methods [17]. This method 25 g were purchased from Charles River Labora-
can identify multiple vessels along an A-line of tories (Seattle, WA, USA). The mouse was deeply
OCTA image for ophthalmology applications after anesthetized using 1.5–2% isoflurane (0.2 L/min O2,

© 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.biophotonics-journal.org

J. Biophotonics (2016) 3

0.8 L/min air) during the experiments and eutha- tuted one B-frame (fast axis). In the slow axis (C-
nized at the end of the experiments. The body tem- scan), there were 400 different locations equally
perature of the animal was maintained at 36.8 C spaced, also covering a distance of
2 mm. At each
through homeothermic blanket system (507220-F, location, 24 repeated B-frames were acquired. With
Harvard Apparatus, MA, USA). The mouse was this scanning protocol, the data cube of one com-
subjected to two imaging sessions: Baseline and per- plete 3D scan was composed of 1024 by 400 by 9600
manent middle cerebral artery (MCA) occlusion. An (z–x–y) voxels, which took
54 s to acquire with an
open-skull cranial window was used to increase the imaging rate of 180 fps. The final 3D OCTA image
penetration depth of OCT signal, covering the distal was composed of 1024 by 400 by 400 (z–x–y) voxels.
branches of MCA. To obtain microvasculature down to capillary le-
Surgical procedure is briefly described as follows: vel at each location, an eigenvalue decomposition
First, a standard 4 × 4 mm cranial window [19] was (ED) based algorithm [22] was used to separate
created on the left parietal cortex 1 mm lateral from structural tissue from flowing erythrocytes from the
sagittal suture and 1 mm posterior from bregma by 8 B-frames to acquire 2 types of images: The first
drilling a circular grooved cranium and lifting the one using an ensemble of 8 consecutive B-frames,
central island. A round cover-glass was then placed and the second one using an ensemble of 8 B-frames
over the exposed brain surface and sealed onto the obtained by skipping 2 frames in between frames
bone with dental cement. Then, the cranial window out of 24 total frames. In this way, the time interval
was subjected to a baseline imaging by OCTA (see Δt for the 2nd image is three times that of the 1st
Section 2.3). After the baseline imaging, the mouse image. The increase of Δt in computing OCTA re-
was subjected to a permanent coagulation of the dis- sults in higher sensitivity [8], but with a cost of in-
tal MCA occlusion (dMCAO) [20]. Another set of creased noise.
OCTA images were taken at the same region under Figure 1 shows the steps taken to acquire OCTA
the cranial window right after dMCAO. with increased SNR by combining the relatively less
noisy original OCTA image with the higher sensitiv-
ity image. Briefly, after collecting 24 B-frames from
one location, we applied ED algorithm to each suc-
2.2 System setup cessive 8 B-frames of 24 total B-frames to get 3
OCTA images (Step 1). Then, we compared these 3
A spectral domain OCT (SD-OCT) system was used blood flow images and picked the pixels with the
for the experiments that comprises a superlumines- maximum intensity to create an OCTA image that
cent diode (Thorlabs Inc., Newton, NJ, USA) as the has a stronger signal (Step 2). In step 3, we trans-
light source [21]. This source has a central wave- formed the image obtained at Step 2 into a guidance
length of 1,340 nm with a bandwidth of 110 nm and image by replacing the intensity value at each pixel
provides a
7 μm axial resolution in the air. In the with the standard deviation around that pixel (using
sample arm, 10× scan lens (Thorlabs Inc., Newton, a moving 3 × 3 kernel). Accordingly, the standard
NJ, USA) was used to achieve
7 μm lateral resolu- deviations of the surrounding pixels are larger at ca-
tion with 0.12 mm depth of field. A home-built spec- pillaries locations. In Step 4, we generated another
trometer, which had a designed spectral resolution OCTA image obtained with the ensemble of 8 B-
of
0.141 nm, provided a detectable depth range of frames with 2 frames interval out of the 24 B-frames.

3 mm on each side of the zero delay line. The line The frame skipping increases a time spacing be-
rate of the linescan camera (1024 pixel detector-ar- tween the inter B-frames, beneficial to increase
ray, Goodrich Inc., Princeton, NJ, USA) employed OCTA signal strength at the capillaries by reducing
in this spectrometer was 92 kHz. The system had a a latency time between RBCs in single file flow at
measured dynamic range of 105 dB with the light the expense of raised additive noise [8]. The pixel
power of 3.5 mW incident onto the sample surface. intensities of capillaries in original image in Step 1
The operations for sample beam scanning, data ac- were replaced with those of the image in Step 4 of
quisition and storage are controlled by a custom which coordinates were guided by the pixel coordi-
software package written in Labview. nates in the image in Step 3. Accordingly, if the stan-
dard deviation value in Step 3 were bigger than an
empirically chosen threshold, those pixel coordinates
were assigned as a capillary location. Eventually, fi-
2.3 OCTA with increased SNR nal image (Step 5) had increased flow signal strength
at the capillaries relative to the original image (Step 1)
In order to acquire OCTA images of mouse cerebral while the background stayed less noisy than the
cortex, we used OCT-based microangiography image at Step 4. This way, we effectively increased
(OMAG) scanning protocol [8]. In this protocol, the SNR of OCTA image as shown in Step 5.
400 A-lines covering a distance of
2 mm consti-

www.biophotonics-journal.org © 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

4 U. Baran et al.: Tail artifact removal in OCT angiography images of rodent cortex

enhancement method [18] for this task. Briefly, we

detected the boundary of the brain surface of in an
A-line in OCT image. After this step, we selected

300 μm within the cortex and divided it into 10
depth-layers, each with
30 μm thickness. To ac-
quire enface projection images from these segmen-
ted layers, we applied sorted maximum intensity
projection (sMIP) algorithm [18] for each layer of
every A-line where the maximum intensity of flow
was selected and mapped to an enface plane. We fi-
nally applied a Gaussian filter (with 2 × 2 kernel) to
smooth the obtained enface projection maps.

2.5 Tail artifact removal in depth-resolved

OCTA images

We propose to utilize automatically segmented en-

face OCT structure images at deeper layers to re-
move the tail artifacts in depth-resolved enface
OCTA images of in vivo mouse cortex. The enface
OCT structural images at deeper layers contain a re-
gion with low intensity (shadow cast) due to the light
attenuation at the superficial blood vessels which
creates a contrast with the surrounding tissue. On
the other hand, same regions look brighter in enface
OCTA images due to the tail artifacts (caused by
forward scattering of light), as shown in Figure 2A.
Although, the capillaries underneath the surface ves-
sels are difficult to detect due to a strong forward
scattering event, at deeper layers, they generate
comparable scattering contrast to the shadow arti-
facts as shown in Figure 2B. Here, we created a
mask by normalizing the structural images with an
adaptive threshold, and multiplied the depth-re-
solved enface OCTA images with the resulting mask
Figure 1 The sketch of steps used to generate increased to remove the tail artifacts while keeping some of
SNR OCTA image. The final image in step 5 is formed by the signal in capillaries within the tail artifacts.
selectively combining background and big vessel pixels
The non-flat surface of the cortex makes the
from step 1, and capillaries from step 4. This selective com-
bination is done using step 3 as a guide to locate the capil-
OCT structure image intensity non-uniform in en-
laries. Accordingly, the standard deviations of the sur- face cross sections. This situation is more severe
rounding pixels (with 3 × 3 kernel) are larger at capillary when a large field of view is used. Hence, to create a
locations as shown in step 3. This information is utilized to structural image mask, first the enface OCT struc-
decide if the corresponding pixel is likely to be from a capil- ture images at various depths were made uniform
lary, when combining images in step 1 and step 4. Dashed using adaptive histogram equalization [23]. Histo-
circles are used to highlight the improvements in SNR in gram equalization is a contrast enhancement techni-
capillaries. que that increases the global contrast of images
when the information content is represented by close
contrast values. In this work, an extension of this
2.4 Automated segmentation and method, called adaptive histogram equalization, was
enhancement of OCT images used to enhance the foreground and suppress the
noise [23]. After this step, the enface OCT structure
In order to accurately map structure and microvas- images, IStr ðx; y; zÞ, were normalized as below:
culature at various depths within cortex, we automa-
tically segmented OCT and OCTA data. We utilized IStr-norm ðx; y; zÞ ¼ Normalize ðIStr ðx; y; zÞ;
ð1Þ
our recently developed automated segmentation and low ðzÞ; high ðzÞÞ

© 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.biophotonics-journal.org

J. Biophotonics (2016) 5

in the previous step as below:

IFlow-out ðx; y; zÞ ¼ ½IFlow ðx; y; zÞ  IFlow ðx; y; 2Þ  αðzÞ

 IStr-norm ðx; y; zÞ
ð2Þ
where IFlow-out ðx; y; zÞ represents the output pixel in-
tensity of an enface OCTA image in the layer z, and
IFlow ðx; y; zÞ refers to the original enface OCTA
pixel intensity at layer z. It is important to note that
the choice of αðzÞ is dependent on the layer depth; it
was chosen larger for the layers 4 and layer 5 to re-
move the strong tail artifacts and smaller for the
deeper layers where the OCT structure mask be-
comes more effective to remove the tail artifacts and
enhance signal contrast at capillaries. The exact val-
ue of αðzÞ depends on the system configuration, and
the tissue optical property as well as its physiology.

3. Results

3.1 Depth-resolved OCTA within healthy

and ischemic mouse cerebral cortex

Cerebral microvasculature in mouse was imaged

through an open-skull cranial window using OCT.
The vessel densities in cortex changes with different
cortical layers, so it is important to reveal true vessel
density distribution for a better understanding of neu-
Figure 2 Overall procedure of acquiring tail artifact re-
moved depth-resolved enface OCTA images of rodent cor-
rovascular coupling during normal and diseased con-
tex. (A) After automated generation of depth-resolved en- ditions. The top panel in Figure 3 shows the depth-re-
face OCT structure and OMAG images, tail artifact algo- solved enface OCTA images of healthy mouse cortex
rithm removes the tailing artifacts at deeper layers while in vivo before applying the proposed method and
keeping some of the capillary plexus intact. (B) Magnified compares with the results from the proposed method
images of structure with shadow cast, microvasculature with at the right panel. At the depth of 90–120 μm under
tail artifact and the final tail artifact suppressed OCTA im- the surface, a bigger scaling factor α was necessary to
age that is obtained by multiplying the enface microvascu- remove the tail artifacts, but in return it also sup-
lar image with the mask created by normalizing the enface pressed the capillaries in these tails artifacts. Starting
structural image with adaptive threshold. Yellow dashed from
150 μm depth, capillaries could be visualized
lines point out the artifact locations. within the tail artifacts after the removal procedure.
Due to the contrast adjustment effect of adaptive en-
face OCT structure masks, the capillaries also appear
Here, low (z) and high (z) refer to the contrast brighter in the depth-resolved enface OCTA images.
thresholds for different layers, z, where z = {1, 2, …, Moreover, Figure 4 shows the original depth-re-
10}. These thresholds were adaptively chosen as the solved enface OCTA images of ischemic mouse cor-
mean and maximum intensity values of depth-re- tex in vivo and compares with the results from the
solved enface OCT structure images, IStr ðx; y; zÞ, re- proposed method. Procedure of dMCAO created a
spectively. large ischemic region which can be seen in the
To remove the tails of big surface vessels more images shown. Although the arteriolo-arteriolar ana-
effective in the deeper layers, we subtracted the con- stomosis recovered some of the missing blood in
trast-adjusted enface OCTA image of the second MCA through stealing blood from the collateral net-
layer (z = 2) from the deeper layers with a scaling work, the majority of the MCA vicinity still re-
factor, 0 < αðzÞ < 1. Finally, we multiplied the mained ischemic [21]. As shown in Figure 4, in the
depth-resolved enface OCTA images from layers (z) original OCTA images, the tail artifacts coming from
4 to 10 with the adaptive structure masks we created MCA appear in the deeper layers of ischemic corti-

www.biophotonics-journal.org © 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

6 U. Baran et al.: Tail artifact removal in OCT angiography images of rodent cortex

Figure 3 Comparison between original depth-resolved en-

face OCTA images of healthy mouse cortex in vivo with Figure 4 Comparison between original depth-resolved en-
the results from the proposed method. The open-skull cra- face OCTA images of ischemic mouse cortex in vivo with
nial window was used. The field of view in the images is the results from the proposed method. The open-skull cra-
1.5 mm × 1.5 mm. nial window was used. Ischemic region is formed after
dMCAO. Occluded MCA branch is pointed out. The field
cal area, which can lead to an inaccurate quantifica- of view in the images is 2 mm × 2 mm.
tion. Using the proposed method, capillary dropout
in that region can be easily distinguished.
densities (VADs) of angiograms at each layer (L1–
L10). The VAD is calculated as a ratio of a number
3.2 Comparison of vessel area density of pixels covered by the vessels to the total number
of pixels in the image [24]. Figure 5A and B show
and vascular dissimilarity in depth-resolved comparison of the calculated VADs with/without tail
enface OCTA images artifact removal for healthy and ischemic mice, re-
spectively. For healthy one, the VAD started to de-
To quantify the tail artifact removal effect on depth- crease from L3. The decrease in the VAD might be
resolved angiograms, we calculated the vessel area due to either cortical vessel geometry or attenuation

© 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.biophotonics-journal.org

J. Biophotonics (2016) 7

Figure 5 Comparison of cortical

vessel area density (VAD) as a
function of depth calculated from
OCT angiograms of (A) healthy
and (B) ischemic mice. The open-
skull cranial window was used.
Ischemic region was formed after
dMCAO.

of OCTA signal strength in depth. With the tail arti- tally correlated) and more than 0.3 at even much
fact removal, however, the VAD quickly dropped at deeper layer (L10). It indicates that the artificial
L4. This is because of the complete removal of tail flow signals from the surface vessels appearing at
artifacts of superficial vessels from L4 and L5 with- the deeper layers increase the vascular similarity in
out being able to distinguish capillaries underneath vessel geometry, resulting in substantial error for
at these tails (see Figures 3–4, 90–120 μm depth vessel visualization and quantification. Similar trend
images). Starting from L6, the capillaries in these was also found for the ischemic mouse cortex as
tails were partially preserved after the tail artifact re- shown in Figure 6b, but its correlation values were
moval (see Figures 3–4, 150–300 μm depth). In con- slightly higher than the healthy case (Figure 6A).
trast, without the tail artifact removal, the VADs re- This can be explained by the capillary dropout ob-
mained relatively high, due to the presence of strong served in MCA side after dMCAO which makes
tails in the enface OCTA images that deceive the images more similar in comparison (see Figure 4).
VAD calculation algorithm. This trend was similar
in the ischemic case as shown in Figure 5B.
Moreover, the local geometry of microvascula-
ture in the superficial cortical vessel network is quite 4. Discussion and conclusion
different from the vascular networks below the sur-
face that consists of many capillaries [1]. We com- Within the interconnected network of microvascula-
pared dissimilarity between the enface OCTA ture in brain, penetrating vessels that are directly at-
images by calculating correlations between the im- tached to the surface pial arterioles branch as a func-
age at the surface (L3) and four others below the tion of depth, resulting in a broad variation in the
surface (L4, L6, L8, and L10). The results were gra- density of microvasculature at different cortex layer
phically represented for healthy and ischemic mice [1]. The distribution of vessel density is strictly regu-
in Figure 6A and B, respectively. After the tail arti- lated at different cortical layers, corresponding to a
fact removal, all the sub-surface angiograms at L4, function of depths of neuronal cells. Moreover,
L6, L8, and L10 were highly uncorrelated with the pathological, physiological, and environmental states
surface angiogram at L3. On the other hand, with no can influence or promote changes in capillary den-
tailing artifact removal, the correlation value be- sity. For example, chronic hypoxia increases capil-
tween angiograms (L3 and L4) was close to 1 (to- lary density [25]. This adaptive increase in capillary

Figure 6 Comparison of correla-

tion values between enface angio-
gram at L3 and each enface angio-
grams at L4, L6, L8, and L10 for
(A) healthy and (B) ischemic
mice, respectively. The open-skull
cranial window was used. Ische-
mic region was formed after
dMCAO.

www.biophotonics-journal.org © 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

8 U. Baran et al.: Tail artifact removal in OCT angiography images of rodent cortex

density during chronic hypoxia increases cerebral the enface OCTA images by calculating correlations
blood volume and restores tissue oxygen tension. between the image at the surface and four others be-
Hypertension also affects brain capillary density by low the surface. As expected, the vessel morphology
causing rarefaction (decrease in number) of capil- in the images acquired after the application of the
laries and impaired microvessel formation that can proposed method are highly dissimilar to the surface
increase vascular resistance [26]. An accurate quanti- vessel morphology, whereas the tail artifacts lead to
fication of microvasculature in deeper brain layers large correlation values between them (see Fig-
(capillary layer) is essential to understand cerebral ure 6).
vascular condition in heathy and diseased states.
In this work, we proposed to utilize enface
depth-resolved OCT structural images to remove the
tail artifacts in enface OCTA images acquired from 4.1 Limitations
the deeper layers of mouse cerebral cortex, in vivo.
The proposed algorithm is shown useful in suppres- OCTA offers a unique ability to quickly image a re-
sing tail artifacts present in enface OCTA images, latively large area in time-sensitive stroke experi-
leading to improved visualization of the deeper mi- ments. However, it comes with some limitations.
crovascular plexus under the big surface vessels, Firstly, the lateral resolution of our system is
7 μm
which are originally not visible due to these artifacts. with a depth of focus of 0.12 mm. The limited resolu-
Thus, the continuity of deeper vascular networks is tion makes it difficult to resolve capillaries that are
preserved for the benefit of qualitative and quantita- closely positioned to each other and can lead to in-
tive studies. accuracies in VAD calculations. On the other hand,
The proposed method assumes that the capil- the areas outside of the depth of focus appear
laries at deeper layers generate a comparable scat- blurred in the enface images and can also lead to in-
tering contrast to the shadow casts in enface OCT accurate VAD calculations. In this study, we tried
structure images. This assumption is not strictly true our best to keep all the crucial parameters, such as
when they are right underneath the big vessels the focus of the probe beam, positioning and orien-
(
90–150 μm under the surface) where the forward tation of the sample to be the same among all the
scattering event is still strong. Therefore, for these imaging sessions.
areas (layers 4 and 5 in our segmented volume) we The proposed algorithm is simple and yet effec-
subtracted surface vessels using a larger ratio, α, to tive to remove the tail artifacts in deeper layers,
eliminate the tail artifacts (see Eq. (2)). However, however it is not perfect. Especially the problem of
the information of capillary plexus beneath the big differentiating capillaries that are closer to the big
surface vessels up to
150 μm was also lost during surface vessels (up to
150 μm) remains as a signifi-
this operation (Figures 3–4). Another important cant challenge. Alternative hardware-based method,
assumption of this method is that the scattering dynamically focused optical coherence microscopy
contrast inside the shadows of enface OCT structure (OCM) angiography with a high numerical aperture,
images at the deeper layers (Figure 2B) come from has shown to be successful at reducing the tail arti-
the flowing erythrocytes in capillaries. This assump- facts in synthesized angiograms [27]. Moreover, the
tion is based on the fact that the high scattering re- multiplication of enface structural images with angio-
gions inside these shadows are aligned well with the graphy images might also suppress the signal in
surrounding microvasculature and are non-uniformly some of the capillaries that are not underneath a big
distributed. Moreover, these tail artifacts mostly ap- blood vessel as seen in Figure 2B. However, this side
pear under the big surface vessels and have a larger effect is relatively weak, thanks to the SNR increase
footprint than the capillaries in the deeper layers, in capillaries before the application of our method
which make their removal easily distinguished com- (see Section 2.3). Future work will also focus on
pared to the effect of this algorithm on capillary translating this method into a volumetric visualiza-
plexus (Figures 3–4). tion of tail artifact removed OCTA.
We utilized this method in a stroke study on In summary, the strongly forward scattering of
mouse for quantifying vessel density changes after photons propagating within the blood vessels causes
dMCAO, in vivo. Figure 5 demonstrates that the tail variations in OCT signals, leading to tail artifacts in
artifacts lead to an increase in the calculated VADs OCTA images beneath the functional blood vessels.
in the depth-resolved enface OCTA images of In spite of the availability of various algorithms, it is
healthy and ischemic mouse brain in vivo, which still difficult to exclude these artifacts in the OCTA
make these results difficult to interpret for compari- images while preserving the capillaries inside these
son studies. The ability of seeing through the tails of tails. Here, we have shown that using our simple
surface vessels enabled more accurate capillary den- method we can remove these artifacts while keeping
sity calculations in deeper layers of cerebral cortex. some of the small vessels underneath intact in enface
Moreover, we quantified the dissimilarity between OCTA images for rodent brain imaging applications.

© 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.biophotonics-journal.org

J. Biophotonics (2016) 9

Due to its post-processing implementation, it can be Kindermann, G. Pellacani, and G. B. E. Jemec. Micro-
easily used in the standard OCT systems. In this vasc. Res. 107 (2016).
study, we focused on a stroke model in mice to show [11] U. Baran, W. J. Choi, and R. K. Wang. Skin Res Tech-
the usefulness of this method but it is reasonable to nol 22(2), 238–246 (2015).
believe that this post-processing method can be [12] Q. Q. Zhang, C. S. Lee, J. Chao, C.-L. Chen, T.
adapted for various other applications. Zhang, U. Sharma, A. Zhang, J. Liu, K. Rezaei, K. L.
Pepple, and R. Munsen. Scientific reports 6 (2016).
Acknowledgements The work was supported in part by [13] U. Baran and R. K. Wang. Neurophotonics 3(1),
National Institutes of Health grants (R01HL093140, and 010902 (2016).
R01EB009682). [14] V. J. Srinivasan and H. Radhakrishnan. NeuroImage
102, 393–406 (2014).
[15] L. Liu, S. S. Gao, S. T. Bailey, D. Huang, D. Li, and
Y. Jia. Biomed. Opt. Express 6(9), 3564–3576 (2015).
References [16] A. Zhang, Q. Zhang, and R. K. Wang. Biomed. Opt.
Express 6(10), 4130–4143 (2015).
[1] B. Pablo, P. S. Tsai, J. P. Kaufhold, P. M. Knutsen, H. [17] M. Zhang, T. S. Hwang, J. P. Campbell, S. T. Bailey,
Suhl, and D. Kleinfeld. Nat. Neurosci 16(7), 889–897 D. J. Wilson, D. Huang, and Y. Jia. Biomed. Opt. Ex-
(2013). press 7(3), 816–828 (2016).
[2] D. A. Boas and A. K. Dunn. J. Biomed. Opt. 15(1), [18] U. Baran, W. Zhu, W. J. Choi, M. Omori, W. Zhang,
011109 (2010). N. J. Alkayed, and R. K. Wang. J Neurosci. Methods
[3] C. K. Willie, F. L. Colino, D. M. Bailey, Y. C. Tzeng, 270(1), 132–137 (2016).
G. Binsted, L. W. Jones, M. J. Haykowsky, J. Bella- [19] Y. Li, U. Baran, and R. K. Wang. PLoS ONE 9(11),
part, S. Ogoh, K. J. Smith, J. D. Smirl, T. A. Day, S. J. e113658 (2014).
Lucas, L. K. Eller, and P. N. Ainslie. J. Neurosci. [20] G. Llovera, S. Roth, N. Plesnila, R. Veltkamp, and A.
Methods 196(2), 221–237 (2011). Liesz. JoVE 89, e51729 (2014).
[4] D. P. Cardenas, E. R. Muir, S. Huang, A. Boley, D. [21] U. Baran, Y. Li, and R. K. Wang. Neurophotonics 2
Lodge, and T. Q. Duong. NeuroImage 119, 382–389 (2), 025006 (2015).
(2015). [22] S. Yousefi, Z. Zhi, and R. K. Wang. IEEE T Biomed
[5] C. Schaffer, B. Friedman, N. Nishimura, L. Schroeder, Eng. 58(8), 2316–2323 (2011).
P. Tsai, F. Ebner, P. Lyden, and D. Kleinfeld. PLoS [23] Y. Siavash, J. Qin, Z. Zhi, and R. K. Wang. Quant.
Biol. 4(2), e22 (2006). Imaging Med. Surg. 3(1), 5–17 (2013).
[6] P. H. Tomlins and R. K. Wang. J. Phys. D: Appl. Phys. [24] R. Reif, J. Qin, L. An, Z. Zhi, S. Dziennis, and R. K.
38(15), 2519–2535 (2005). Wang. J. Biomed. Opt. 2012, 9 (2012).
[7] R. K. Wang, S. L. Jacques, Z. Ma, S. Hurst, S. R. [25] M. J. Cipolla. The Cerebral Circulation. San Rafael
Hanson, and A. Gruber. Opt. Express 15(7), 4083– (CA): Morgan & Claypool Life Sciences; Chapter 2,
4097 (2007). Anatomy and Ultrastructure. (2009).
[8] L. An, J. Qin, and R. K. Wang. Opt. Express 18(8), [26] I. A. Sokolova, E. B. Manukhina, S. M. Blinkov, V. B.
8220–8228 (2010). Koshelev, V. G. Pinelis, and I. M. Rodionov. Micro-
[9] H. Wang, U. Baran, Y. Li, W. Qin, W. Wang, H. vasc. Res. 30(1), 1–9 (1985).
Zeng, and R. K. Wang. J. Biomed. Opt. 19(10), [27] L. Conor, H. Radhakrishnan, M. Bernucci, and V. J.
106011 (2014). Srinivasan. J. Biomed. Opt. 21(2), 020502 (2016).
[10] L. Themstrup, J. Welzel, S. Ciardo, R. Kaestle, M.
Ulrich, J. Holmes, R. Whitehead, E. C. Sattler, N.

www.biophotonics-journal.org © 2016 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim