Cerebral artery segmentation based on magnetization-prepared … · We employed the Dice...
Transcript of Cerebral artery segmentation based on magnetization-prepared … · We employed the Dice...
1
Cerebral artery segmentation based on magnetization-prepared two rapid acquisition gradient
echo multi-contrast images in 7 Tesla magnetic resonance imaging
Uk-Su Choia, b
, Hirokazu Kawaguchic, 1
, and Ikuhiro Kidaa, b, *
a Center for Information and Neural Networks, National Institute of Information and Communications
Technology, Japan
b Graduate School of Frontier Biosciences, Osaka University, Japan
c Siemens Healthcare K.K., Japan
*Corresponding author
Ikuhiro Kida, Ph.D.
Center for Information and Neural Networks,
National Institute of Information and Communications Technology
Email: [email protected]
1 Present address: Division of Advanced Multidisciplinary Research, Tokyo Medical and Dental
University,
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
2
Abstract
Cerebral artery segmentation plays an important role in the direct visualization of the human brain to
obtain vascular system information. At ultra-high field magnetic resonance imaging, hyperintensity of
the cerebral arteries in T1 weighted (T1w) images could be segmented from brain tissues such as gray
and white matter. In this study, we propose an automated method to segment the cerebral arteries
using multi-contrast images including T1w images of a magnetization-prepared two rapid acquisition
gradient echoes (MP2RAGE) sequence at 7 T. The proposed method employed a seed-based region-
growing strategy with the following procedures. (1) Two seed regions were defined by Frangi filtering
applied to T1w images and by a simple calculation from multi-contrast images, (2) the two seed
regions were combined, (3) the combined seed regions were expanded using a region growing
algorithm to acquire the cerebral arteries. Time-of-flight (TOF) images were obtained as a reference
to evaluate the proposed method. We successfully performed vessel segmentations from T1w
MP2RAGE images, which mostly overlapped with the segmentations from the TOF images. As large
arteries can affect the normalization of anatomical images to the standard coordinate space in
functional and structural studies, we also investigated the effect of the cerebral arteries on spatial
transformation using vessel segmentation by the proposed method. As a result, the T1w image
removing the cerebral arteries showed better agreement with the standard atlas compared with the
T1w image containing the arteries. Thus, because the proposed method using MP2RAGE images can
obtain brain tissue anatomical information as well as cerebral artery information without need for
additional acquisitions such as of the TOF sequence, it is useful and time saving for medical diagnosis
and functional and structural studies.
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
3
1. Introduction
The visualization of the cerebral vasculature plays an important role in the diagnosis of
vascular diseases such as hemorrhage or aneurysms (Stapf et al., 2006; Matsushige et al., 2016) but
also to differentiate between anatomical areas in different brain regions (Neumann et al., 2016).
Uncertain segmentation of brain tissues and blood vessels can lead to misidentifying the localization
of the electroencephalographic signal source and lead to anatomical confounding in certain cortical
regions with large vessels in studies of brain function and of cortical characteristics, such as cortical
thickness(Fiederer et al., 2016; Viviani, 2016). Therefore, accurate segmentation of blood vessels is
essential for neuroimaging research and medical image analysis.
Time-of-flight (TOF) magnetic resonance angiography (MRA) has been widely used as a
non-invasive technique to contrast between blood and brain tissues (Lévy et al., 1994). The
unsaturated blood flow of cerebral arteries on the TOF image is brighter than the saturated
background of other brain tissues. As at ultra-high fields (UHFs), the longitudinal relaxation times
(T1) are extended both for blood and brain tissues, leading to enhanced contrast between them,
utilization of UHFs for vessel imaging is more efficient (Park et al., 2018). However, there is less
anatomical information such as gray matter (GM) and white matter (WM) in TOF images due to
background suppression.
Magnetization-prepared rapid gradient echo (MPRAGE) is a candidate for the visualization
of blood vessels indicating high signal intensity of arterial blood vessels while the background shows
intermediate signal intensities, resulting in high vessel-to-background contrast and is potentially
suitable for MRA display (Penumetcha et al., 2008, Wrede et al 2014). Although the T1 weighted
(T1w) image in the MPRAGE sequence provides good anatomical information and differentiation of
GM and WM, it is difficult to automatically segment the high signal arteries from the intermediate
signal tissues in this image. The enhanced contrast between cerebral arteries and brain tissues in the
T1w image at UHF can help automatically segment cerebral arteries from the brain (Maderwald et al.,
2008; Van de Moortele et al., 2009; Fiederer et al., 2016; Gulban et al., 2018).
Magnetization prepared two rapid acquisition gradient echoes (MP2RAGE) can obtain three
different images, a T1 map (T1) and a uniform T1w image with background noise (UNI) and without
background noise (UNIDEN) derived by two images at different inversion times, i.e., the first and
second inversion gradient echo images (INV1 and INV2, respectively) (Marques et al., 2010). The
MP2RAGE sequence has an advantage over the MPRAGE sequence for signal homogeneity in the
T1w image (UNI) because the inhomogeneity is cancelled out when the T1w image in the MP2RAGE
sequence is calculated from two inversion gradient echo images. As multi-contrast images in the
MP2RAGE sequence are useful to segment brain tissues such as GM, WM, and cerebrospinal fluid
(CSF) (Choi et al., 2019), the MP2RAGE sequence may also help segment the cerebral arteries.
In this study, we propose an automated method to segment the cerebral arteries using multi-
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
4
contrast images in the MP2RAGE sequence at 7 T. The proposed method is based on our brain tissue
segmentation (MP2rage based RApid SEgmentation; MP2RASE) (Choi et al., 2019) combined with a
seed-based approach and a region-growing algorithm. The proposed method using the MP2RAGE
sequence was evaluated using vessel segmentation with the TOF sequence as a reference. In addition,
we investigated the effect of cerebral arteries on the spatial transformation of T1w images to a
standard coordinate space, i.e., the Montreal Neurological Institute (MNI) space, which is ordinarily
used in the process of functional and structural MR imaging (MRI) studies.
2. Materials and Methods
2.1 Subjects
Four volunteers (four men, aged 34–49 years) without history of neurological disease or any other
medical condition participated in this study after providing written informed consent. All experiments
were approved by the Ethics and Safety Committees of National Institute of Information and
Communications Technology.
2.2 MRI acquisition
The experiments were performed on a 7-T investigational MRI scanner (MAGNETOM 7T; Siemens
Healthineers, Erlangen, Germany) with a 32-channel head coil (Nova Medical, Wilmington, MA).
The MP2RAGE was acquired using a research sequence from Siemens Healthineers and using the
following parameters: repetition time = 5000 ms, echo time = 3.43 ms, inversion times = 800 ms/2600
ms, flip angles = 4°/5°, matrix = 368 × 368 × 256, voxel size = 0.7 × 0.7 × 0.7 mm3, GRAPPA
acceleration factor = 3, and scan time = 9 min 27 s (Choi et al., 2019). To evaluate the proposed
method, we additionally acquired 3D TOF images with the following parameters: voxel resolution =
0.3 x 0.3 x 0.4 mm3, repetition time = 20 ms, echo time = 4.47 ms, flip angle = 18°, matrix = 704 ×
704 × 192, GRAPPA acceleration factor = 4, and scan time = 12 min 19 s (Wrede et al., 2014;
Matsushige et al., 2016). The acquisition region of the TOF sequence included large cerebral arteries
such as the anterior cerebral artery (ACA), middle cerebral artery (MCA), and internal carotid artery
(ICA).
2.3 Vessel segmentation using MP2RAGE images
The image processing pipeline in the proposed method is shown in Fig. 1. All MP2RAGE images
were stripped to remove non-brain structures using the BET toolbox in FSL 5.0.10 (FMRIB, Oxford,
UK) during the brain segmentation procedure. After removing non-brain structures, we normalized
the intensities of all images for the brain tissue segmentation using a feature-scaling method described
in Eq. 1 because MP2RAGE images have different intensity scales. Sraw indicated raw intensities, min
(Sraw) indicated the minimum intensity, and max (Sraw) indicated the maximum intensity of each
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
5
MP2RAGE image.
��������� � � ���� ��� ������
��� ������ ��� ������� (1)
First, we employed a Frangi filter (Frangi et al., 1998), which has been used for vessel segmentation
in TOF (Hsu et al., 2019) and MPRAGE (Fiederer et al., 2016; Oliveira et al., 2016; ), to construct a
binary vessel mask (Seed 1 in Fig. 1) from the normalized UNI (nUNI) image (Kroon, 2009). The
filtering was performed with the following parameters; α = 0.5, β = 0.5, γ = 5.0 (Antiga, 2007). The
filtered image was binarized with a threshold of a 10% robust range of non-zero voxels and masked
out by a WM binary mask, which was segmented using our previous method (Choi et al., 2019).
Second, a binary vessel mask (Seed 2 in Fig. 1) was constructed using three normalized contrast
images (nINV1, nINV2, and nT1) and a CSF binary mask, which was also segmented using our
previous method (Choi et al., 2019). The binary vessel masks, i.e., seed 1 and seed 2, were combined
for the region-growing procedure. Finally, the vessel mask was obtained by expanding the voxels of
the combined binary vessels masks by a region growing procedure using our in-house MATLAB code
including a recursive region growing algorithm (Daniel, 2011).
2.4 Vessel segmentation using TOF images
The intensities of the TOF images were normalized using the feature-scaling method described in Eq.
1. The vessels were segmented from the normalized TOF images using a Frangi filter with the
following parameters; α = 0.5, β = 0.5, γ = 5.0. The filtered images were binarized with a threshold of
a 25% robust range of non-zero voxels to construct the TOF binary vessel mask.
2.5 Comparison between vessel segmentations using MP2RAGE and TOF images
We employed the Dice coefficient to evaluate the similarity of the vessel segmentations of MP2RAGE
images with those of TOF images as a reference. Before evaluating the similarity, the two datasets
were coregistered because the dimensions between TOF and MP2RAGE images differ
(Supplementary Fig. 1). TOF images were initially registered to the UNIDEN of the MP2RAGE
sequence to acquire the transformation matrix using FLIRT in FSL. The matrix was applied to binary
vessel masks from TOF images with nearest-neighbor interpolation. Next, the pre-registered vessel
masks from the TOF images were registered to vessel masks from MP2RAGE for fine spatial
transformation using the non-linear registration method of Advanced Normalization Tools (ANTs)
(Avants et al., 2011) with nearest-neighbor interpolation. After registration, we defined two fields of
view (FOVs) for the evaluation with the Dice coefficient. One FOV was the acquisition region of the
TOF sequence, which was fully covered by the MP2RAGE image (Supplementary Fig. 1). Another
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
6
FOV was manually defined including large arteries such as the ACA, MCA, and ICA as local FOV
regions. The vessel segmentations within two FOVs were evaluated to calculate the Dice coefficient
with Eq. 2.
��� � �|��������� � �� ��������� � �� �|
|��������� � �� �|�|������� � �� �| 100 (2)
2.5. Spatial normalization to the standard space
Cerebral arteries in the UNIDEN T1w image were firstly removed based on vessel segmentation by
the proposed method using an in-house script. The T1w images with and without cerebral arteries
were spatially normalized to the MNI152 template in FSL using an in-house script including ANTs.
3. Results
3.1 Vessel segmentation using MP2RAGE images
Two binary vessel masks were constructed by independent processes: Frangi filtering of the nUNI
image and a simple calculation of MP2RAGE images. The masks partially overlapped, indicating that
the two seed regions represented independent parts of the cerebral arteries (Fig. 2). Most voxels in the
arteries were extracted as a vessel mask by Frangi filtering (Fig. 2a) but some voxels in relatively
larger arteries were not obtained. Conversely, the voxels, which were not detected by Frangi filtering,
were represented as a vessel mask using a simple calculation of MP2RAGE images (Fig. 2b).
Application of a region-growing algorithm for a combination of seed 1 and seed 2 extracted most
voxels in the arteries as a vessel mask (Fig. 2c), which shown in 3D reconstruction view of the
cerebral arteries (Fig. 3). Large arteries such as the MCA, ACA, and ICA were well segmented.
3.2 Comparison between vessel segmentations of MP2RAGE and TOF images
Figure 4 shows a comparison of vessel segmentations from MP2RAGE images by the proposed
method and from TOF images by Frangi filtering in 3D reconstruction view. Most arteries extracted
from the MP2RAGE images by the proposed method showed good agreement with the ones extracted
from the TOF images. Larger arteries such as the MCA, ACA, and ICA well overlapped, but
peripheral branches of arteries were mismatched between vessel segmentations of MP2RAGE and
TOF images. We evaluated the proposed method by comparison to TOF vessel segmentation using the
Dice coefficient for whole FOV regions, which overlapped between the two sequences, and the local
FOV regions including larger arteries (Fig. 5). The Dice coefficients with vessel segmentation from
TOF images were approximately 40% for the whole FOV regions and 55% for local regions when
vessels in the T1w MP2RAGE image were segmented by Frangi filtering (seed 1), which was the
same method as that used in TOF segmentation. The coefficients by simple calculation from the
MP2RAGE images (seed 2) for both regions were almost the same as those obtained with Frangi
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
7
filtering. After combining the two seeds, the similarities of vessel segmentation dramatically
improved from 40 to 57% for the whole FOV regions and from 55 to 73% for the local FOV regions.
The Dice coefficients showed a lower similarity for the whole FOV region because this approach may
not be able to properly handle the small branches of large cerebral arteries.
3.3 Influence of cerebral arteries on normalization to standard coordinate space
We examined the influence of the cerebral arteries on the spatial transformation of T1w images to a
standard coordinate space, i.e., the MNI space, which is ordinarily used in the process of functional
MRI and voxel-based morphometry (VBM). Figure 6 shows the spatial transformation in the T1w
image with and without cerebral arteries, segmented by the proposed method. By removing the MCA
and ACA from the T1w image, the GM intensities in the frontal orbital cortex and cingulate cortex
recovered when the T1w image was normalized to the standard coordinate space. Thus, the T1w
image after normalization excluding the cerebral arteries is very consistent with the MNI template in
FSL.
Discussion
We developed a novel method to automatically segment cerebral arteries with a region
growing algorithm for two combined seed masks from T1w MP2RAGE sequence images at 7 T. The
two seed masks were defined by Frangi filtering of the T1w image and by a simple calculation of
multi-contrast images in the MP2RAGE sequence. The seed mask of the region growing algorithm,
which has been widely used as a segmentation method for medical images (Malek et al., 2012), is
important to define core voxels of arteries for the robust region growth of cerebral arteries. Most of
the seed voxels were defined by Frangi filtering, which has been employed to segment blood vessels
in the retina (Oliveira et al., 2016) and the brain (Fiederer et al., 2016; Hsu et al., 2019). The filtering
can be useful to differentiate blood vessels from brain tissues in high contrast images, i.e., TOF
images but not with less suppressed tissue images such as T1w image (i.e., UNI) in the MP2RAGE
sequence. The advantage of multiple contrasts for segmentation in MP2RAGE images has been
described in previous studies (Wang et al., 2018; Choi et al., 2019). Therefore, as more seed voxels
might be needed for segmentation, we additionally defined the seed voxels using a simple calculation
of multi-contrast images from the MP2RAGE sequence, which provided better results.
As the major cerebral arteries are located in frontal parts of the brain such as the anterior
cingulate, orbitofrontal, and insular cortices, which are core brain regions in cognitive studies
(Bechara et al., 2000; Bush et al., 2000; Mutschler et al., 2009), the large cerebral arteries might
influence the standard space transformation of T1w images. Better agreement with the MNI atlas was
achieved when the T1w image was transformed to exclude the cerebral arteries using vessels mask
extracted by the proposed method compared with the agreement of T1w images that contained the
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
8
cerebral arteries (Fig. 6), which led to more precise transformation evaluation. Precise transformation
information could be important to assess the thickness of the cerebral cortex and volume using
methods such as VBM (Kotikalapudi et al., 2019) and to obtain accurate location of neuronal
activation such as olfactory functional MRI study (Gottfried et al., 2002). Especially, as recent UHF
studies have focused on fine brain structures and brain function at the submillimeter scale without
spatial smoothing, functional images with high spatial resolution can be affected because cerebral
arteries have the potential to induce errors of structural transformation in those cortical areas, which
can induce mis-localization of specific brain function.
The proposed method has limitations to be addressed in further studies. The proposed
method could reliably detect major cerebral arteries such as the MCA, ACA, and ICA but failed to
segment small peripheral vessels possibly due to less contrast with the surrounding brain tissues
because of limited resolution, slow blood flow rate, and impaired image quality involving distortion
or inhomogeneity. B1 mapping of the SA2RAGE sequence (Eggenschwiler et al., 2012) could
possibly improve the inhomogeneity of the T1w image at UHFs. The enhanced contrast to noise ratio
between GM and WM could also help improve the segmentation by changing the acquisition MR
parameters of the MP2RAGE sequence such as the inversion times and flip angles (Tanner et al., 2012;
Marques and Gruetter, 2013). Although increased spatial resolution could be another possibility to
measure peripheral arteries, a low signal to noise ratio and longer scan time comprise trade-offs for
the MP2RAGE sequence.
We employed vessel segmentation of TOF images with Frangi filtering as reference to
compare with the proposed method based on the Dice coefficient. The quality of vessel segmentation
in both TOF and the proposed method depends on the parameter and threshold of Frangi filtering,
which also depends on the quality of the TOF (Phellan et al., 2017). In fact, parts of major arteries
were not segmented using TOF images with Frangi filtering. Further evaluation of the proposed
method will be needed by a quantitative verification with the different acquisition and analysis
parameters (Frangi filtering and threshold), which may lead to further improvement.
In conclusion, we proposed a novel method to automatically segment cerebral arteries using
MP2RAGE images at 7 T. As we previously developed a method to segment brain tissues using the
MP2RAGE sequence (MP2RASE) (Choi et al., 2019), together, a single MP2RAGE sequence
acquisition could provide automated brain tissue masks of GM, WM, and CSF as well as of blood
vessels. In addition, the proposed method could be used to derive accurate spatial transformation
information with reference to the standard coordinate space by removing the influences of cerebral
arteries from T1w images. Thus, because the proposed method of vessel segmentation using
MP2RAGE sequence can obtain brain tissue anatomical information as well as cerebral artery
information without any additional acquisitions such as TOF sequence, it is useful and time saving for
medical diagnosis and functional and structural studies.
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
9
Acknowledgements
We would like to thank Dr. Kober for the development of MP2RAGE sequence and technical supports.
This work was supported by the JSPS KAKENHI [grant number JP18K07701, JP18H04084, and
JP19H03537] from the Japanese Ministry of Education, Culture, Sports, Science and Technology (IK).
Declarations of interests
HK was employed by Siemens Healthcare K.K. Siemens Healthcare K.K. did not have any additional
role in the study design, data collection and analysis. UC and IK have no conflicts of interest.
References
Antiga, L., 2007. Generalizing vesselness with respect to dimensionality and shape. Insight J.,
Bergamo.
Avants, B.B., Tustison, N.J., Song, G., Cook, P.A., Klein, A., Gee, J.C., 2011. A reproducible
evaluation of ANTs similarity metric performance in brain image registration. NeuroImage 54,
2033–2044. https://doi.org/10.1016/j.neuroimage.2010.09.025.
Bechara, A., Damasio, H., Damasio, A.R., 2000. Emotion, Decision Making and the Orbitofrontal
Cortex. Cereb. Cortex 10, 295–307. https://doi.org/10.1093/cercor/10.3.295.
Bush, G., Luu, P., Posner, M.I., 2000. Cognitive and emotional influences in anterior cingulate cortex.
Trends Cogn. Sci. 4, 215–222. https://doi.org/10.1016/S1364-6613(00)01483-2.
Choi, U.-S., Kawaguchi, H., Matsuoka, Y., Kober, T., Kida, I., 2019. Brain tissue segmentation based
on MP2RAGE multi-contrast images in 7 T MRI. PLOS ONE 14, e0210803.
https://doi.org/10.1371/journal.pone.0210803.
Daniel, 2011. Region Growing (2D/3D grayscale).
https://mathworks.com/matlabcentral/fileexchange/32532-region-growing-2d-3d-grayscale.
Eggenschwiler, F., Kober, T., Magill, A.W., Gruetter, R., Marques, J.P., 2012. SA2RAGE: a new
sequence for fast B1+ -mapping. Magn. Reson. Med. 67, 1609–1619.
https://doi.org/10.1002/mrm.23145.
Fiederer, L.D.J., Vorwerk, J., Lucka, F., Dannhauer, M., Yang, S., Dümpelmann, M., Schulze-Bonhage,
A., Aertsen, A., Speck, O., Wolters, C.H., Ball, T., 2016. The role of blood vessels in high-
resolution volume conductor head modeling of EEG. NeuroImage 128, 193–208.
https://doi.org/10.1016/j.neuroimage.2015.12.041.
Frangi, A.F., Niessen, W.J., Vincken, K.L., Viergever, M.A., 1998. Multiscale vessel enhancement
filtering, in: Wells, W.M., Colchester, A., Delp, S. (Eds.), Medical Image Computing and
Computer-Assisted Intervention — MICCAI’98, Lecture Notes in Computer Science. Springer
Berlin Heidelberg, pp. 130–137. https://doi.org/10.1007/BFb0056195.
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
10
Gottfried, J.A., Deichmann, R., Winston, J.S., Dolan, R.J., 2002. Functional Heterogeneity in Human
Olfactory Cortex: An Event-Related Functional Magnetic Resonance Imaging Study. J. Neurosci.
22, 10819–10828. https://doi.org/10.1523/JNEUROSCI.22-24-10819.2002.
Gulban, O.F., Schneider, M., Marquardt, I., Haast, R.A.M., Martino, F.D., 2018. A scalable method to
improve gray matter segmentation at ultra high field MRI. PLOS ONE 13, e0198335.
https://doi.org/10.1371/journal.pone.0198335.
Hsu, C.-Y., Li, Y., Han, Y., Elijovich, L., Sabin, N.D., Abuelem, T., Torabi, R., Faught, A., Hua, C.-H.,
Klimo, P., Merchant, T.E., Lucas, J.T., 2019. Automatic image processing pipeline for tracking
longitudinal vessel changes in magnetic resonance angiography. J. Magn. Reson. Imaging 50,
1063–1074. https://doi.org/10.1002/jmri.26699.
Kotikalapudi, R., Martin, P., Erb, M., Scheffler, K., Marquetand, J., Bender, B., Focke, N.K., 2019.
MP2RAGE multispectral voxel-based morphometry infocal epilepsy. Hum. Brain Mapp. Epub
ahead. https://doi.org/10.1002/hbm.24756.
Kroon, D.-J., 2009. Hessian based frangi vesselness filter.
http://mathworks.com/matlabcentral/fileexchange/24409-hessian-basedfrangi-vesselness-filter.
Lévy, C., Laissy, J.P., Raveau, V., Amarenco, P., Servois, V., Bousser, M.G., Tubiana, J.M., 1994.
Carotid and vertebral artery dissections: three-dimensional time-of-flight MR angiography and
MR imaging versus conventional angiography. Radiology 190, 97–103.
https://doi.org/10.1148/radiology.190.1.8259436.
Maderwald, S., Ladd, S.C., Gizewski, E.R., Kraff, O., Theysohn, J.M., Wicklow, K., Moenninghoff,
C., Wanke, I., Ladd, M.E., Quick, H.H., 2008. To TOF or not to TOF: strategies for non-contrast-
enhanced intracranial MRA at 7 T. Magn. Reson. Mater. Phys. Biol. Med. 21, 159.
https://doi.org/10.1007/s10334-007-0096-9.
Malek, A.A., Rahman, W.E.Z.W.A., Yasiran, S.S., Jumaat, A.K., Jalil, U.M.A., 2012. Seed point
selection for seed-based region growing in segmenting microcalcifications, in: 2012 International
Conference on Statistics in Science, Business and Engineering (ICSSBE). Presented at the 2012
International Conference on Statistics in Science, Business and Engineering (ICSSBE), pp. 1–5.
https://doi.org/10.1109/ICSSBE.2012.6396580.
Marques, J.P., Gruetter, R., 2013. New Developments and Applications of the MP2RAGE Sequence -
Focusing the Contrast and High Spatial Resolution R1 Mapping. PLOS ONE 8, e69294.
https://doi.org/10.1371/journal.pone.0069294.
Marques, J.P., Kober, T., Krueger, G., van der Zwaag, W., Van de Moortele, P.-F., Gruetter, R., 2010.
MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at
high field. NeuroImage 49, 1271–1281. https://doi.org/10.1016/j.neuroimage.2009.10.002.
Matsushige, T., Kraemer, M., Schlamann, M., Berlit, P., Forsting, M., Ladd, M.E., Sure, U., Wrede,
K.H., 2016. Ventricular Microaneurysms in Moyamoya Angiopathy Visualized with 7T MR
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
11
Angiography. Am. J. Neuroradiol. 37, 1669–1672. https://doi.org/10.3174/ajnr.A4786.
Mutschler, I., Wieckhorst, B., Kowalevski, S., Derix, J., Wentlandt, J., Schulze-Bonhage, A., Ball, T.,
2009. Functional organization of the human anterior insular cortex. Neurosci. Lett. 457, 66–70.
https://doi.org/10.1016/j.neulet.2009.03.101.
Neumann, J.-O., Giese, H., Nagel, A.M., Biller, A., Unterberg, A., Meinzer, H.-P., 2016. MR
Angiography at 7T to Visualize Cerebrovascular Territories. J. Neuroimaging 26, 519–524.
https://doi.org/10.1111/jon.12348.
Okubo, G., Okada, Tomohisa, Yamamoto, A., Kanagaki, M., Fushimi, Y., Okada, Tsutomu, Murata, K.,
Togashi, K., 2016. MP2RAGE for deep gray matter measurement of the brain: a comparative
study with MPRAGE. J. Magn. Reson. Imaging 43, 55–62. https://doi.org/10.1002/jmri.24960.
Oliveira, W.S., Teixeira, J.V., Ren, T.I., Cavalcanti, G.D.C., Sijbers, J., 2016. Unsupervised Retinal
Vessel Segmentation Using Combined Filters. PLOS ONE 11, e0149943.
https://doi.org/10.1371/journal.pone.0149943.
Park, C.-A., Kang, C.-K., Kim, Y.-B., Cho, Z.-H., 2018. Advances in MR angiography with 7T MRI:
From microvascular imaging to functional angiography. NeuroImage, Neuroimaging with Ultra-
high Field MRI: Present and Future 168, 269–278.
https://doi.org/10.1016/j.neuroimage.2017.01.019.
Penumetcha, N., Jedynak, B., Hosakere, M., Ceyhan, E., Botteron, K.N., Ratnanather, J.T., 2008.
Segmentation of arteries in MPRAGE images of the ventral medial prefrontal cortex. Comput.
Med. Imaging Graph. 32, 36–43. https://doi.org/10.1016/j.compmedimag.2007.08.013.
Phellan, R., Peixinho, A., Falcão, A., Forkert, N.D., 2017. Vascular Segmentation in TOF MRA
Images of the Brain Using a Deep Convolutional Neural Network, in: Cardoso, M.J., Arbel, T.,
Lee, S.-L., Cheplygina, V., Balocco, S., Mateus, D., Zahnd, G., Maier-Hein, L., Demirci, S.,
Granger, E., Duong, L., Carbonneau, M.-A., Albarqouni, S., Carneiro, G. (Eds.), Intravascular
Imaging and Computer Assisted Stenting, and Large-Scale Annotation of Biomedical Data and
Expert Label Synthesis, Lecture Notes in Computer Science. Springer International Publishing,
pp. 39–46. https://doi.org/10.1007/978-3-319-67534-3_5.
Stapf, C., Mast, H., Sciacca, R., Choi, J., Khaw, A., Connolly, E., Pile-Spellman, J., Mohr, J., 2006.
Predictors of hemorrhage in patients with untreated brain arteriovenous malformation. Neurology
66, 1350–1355. https://doi.org/10.1212/01.wnl.0000210524.68507.87.
Tanner, M., Gambarota, G., Kober, T., Krueger, G., Erritzoe, D., Marques, J.P., Newbould, R., 2012.
Fluid and white matter suppression with the MP2RAGE sequence. J. Magn. Reson. Imaging 35,
1063–1070. https://doi.org/10.1002/jmri.23532.
Van de Moortele, P.-F., Auerbach, E.J., Olman, C., Yacoub, E., Uğurbil, K., Moeller, S., 2009. T1
weighted brain images at 7 Tesla unbiased for Proton Density, T2� contrast and RF coil receive
B1 sensitivity with simultaneous vessel visualization. NeuroImage 46, 432–446.
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
12
https://doi.org/10.1016/j.neuroimage.2009.02.009.
Viviani, R., 2016. A Digital Atlas of Middle to Large Brain Vessels and Their Relation to Cortical and
Subcortical Structures. Front. Neuroanat. 10. https://doi.org/10.3389/fnana.2016.00012.
Wang, Yishi, Wang, Yajie, Zhang, Z., Xiong, Y., Zhang, Q., Yuan, C., Guo, H., 2018. Segmentation of
gray matter, white matter, and CSF with fluid and white matter suppression using MP2RAGE. J.
Magn. Reson. Imaging 0. https://doi.org/10.1002/jmri.26014.
Wrede, K.H., Dammann, P., Mönninghoff, C., Johst, S., Maderwald, S., Sandalcioglu, I.E., Müller, O.,
Özkan, N., Ladd, M.E., Forsting, M., Schlamann, M.U., Sure, U., Umutlu, L., 2014. Non-
Enhanced MR Imaging of Cerebral Aneurysms: 7 Tesla versus 1.5 Tesla. PLOS ONE 9, e84562.
https://doi.org/10.1371/journal.pone.0084562.
Abbreviations:
ACA: anterior cerebral artery
CSF: cerebrospinal fluid
FOV: field of view
GM: gray matter
ICA: internal carotid artery
INV1: first inversion gradient echo image
INV2: second inversion gradient echo image
MCA: middle cerebral artery
MNI: Montreal Neurological Institute
MRA: magnetic resonance angiography
MPRAGE: Magnetization-prepared rapid gradient-echo
MP2RAGE: magnetization-prepared two rapid acquisition gradient echoes
TOF: time-of-flight
UHF: ultra-high field
UNI: uniform T1w image with background noise
UNIDEN: uniform T1w image without background noises
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
13
VBM: voxel-based morphometry
WM: white matter
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
14
Figure caption
Figure 1. Image processing pipeline of the proposed method.
Dotted squared images were constructed from the brain tissue segmentation method (Choi et al.,
2019). CSF: cerebral spinal fluid, nINV1: normalized 1st inversion time image, nINV2: normalized 2nd
inversion time image, nT1: normalized T1 map, nUNI: normalized uniform T1w image, WM: white
matter.
Figure 2. Visual presentation of segmentation at major stages. (a) segmentation by Frangi filtering
applied to the T1w image, (b) segmentation by a simple calculation of multi-contrast images, (c)
segmentation in the final result of the proposed method. Segmentation masks were overlaid to the
T1w image, which indicated a sagittal view of regions around the anterior cerebral artery.
Figure 3. 3D visualization of vessel segmentation by the proposed method in all subjects.
Figure 4. Comparison of cerebral artery segmentation between the results from MP2RAGE and
TOF images in all subjects. Red color: overlapped segmentation between two methods, Green color:
segmentation only from MP2RAGE images, Blue color: segmentation only from TOF images.
Figure 5. Similarity between vessel segmentations at major stages from MP2RAGE and TOF
images. Seed 1: segmentation by Frangi filtering, Seed 2: segmentation by a simple calculation of
MP2RAGE images. The Dice values show vessel segmentation from the whole FOV region (whole)
and from large vessel FOV regions (local) in all subjects.
Figure 6. Effect of the cerebral arteries on normalization to the MNI coordinates.
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
15
(a) Normalized T1w image with cerebral arteries, (b) Normalized T1w image without cerebral arteries,
(c) MNI template. The 1st, 2nd, and 3rd columns from the left side indicate the axial view of the
orbitofrontal cortex and insular cortex and the sagittal view of the cingulate cortex, respectively. The
MCA in the 1st and 2nd columns and the ACA in the 3rd column are shown in (a) but not in (b).
Supplementary Figure 1. MP2RAGE and TOF images. (a) uniform T1w image without
background noise from the MP2RAGE sequence, (b) TOF image.
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted December 12, 2019. . https://doi.org/10.1101/2019.12.12.870840doi: bioRxiv preprint