ArticlePublisher preview available

Machine learning analysis of whole mouse brain vasculature

To read the full-text of this research, you can request a copy directly from the authors.

Abstract and Figures

Tissue clearing methods enable the imaging of biological specimens without sectioning. However, reliable and scalable analysis of large imaging datasets in three dimensions remains a challenge. Here we developed a deep learning-based framework to quantify and analyze brain vasculature, named Vessel Segmentation & Analysis Pipeline (VesSAP). Our pipeline uses a convolutional neural network (CNN) with a transfer learning approach for segmentation and achieves human-level accuracy. By using VesSAP, we analyzed the vascular features of whole C57BL/6J, CD1 and BALB/c mouse brains at the micrometer scale after registering them to the Allen mouse brain atlas. We report evidence of secondary intracranial collateral vascularization in CD1 mice and find reduced vascularization of the brainstem in comparison to the cerebrum. Thus, VesSAP enables unbiased and scalable quantifications of the angioarchitecture of cleared mouse brains and yields biological insights into the vascular function of the brain. VesSAP is a tissue clearing- and deep learning-based pipeline for comprehensively analyzing mouse vasculature, from large vessels to small capillaries.
This content is subject to copyright. Terms and conditions apply.
1Institute for Tissue Engineering and Regenerative Medicine (iTERM), Helmholtz Zentrum München, Neuherberg, Germany. 2Institute for Stroke and
Dementia Research (ISD), Ludwig-Maximilians-Universität (LMU), Munich, Germany. 3Graduate School of Neuroscience (GSN), Munich, Germany.
4Department of Computer Science, Technical University of Munich (TUM), Munich, Germany. 5Center for Translational Cancer Research of the TUM
(TranslaTUM), Munich, Germany. 6Munich School of Bioengineering, Technical University of Munich (TUM), Munich, Germany. 7Institute of Pharmacology
and Toxicology, University of Zurich (UZH), Zurich, Switzerland. 8Munich Cluster for Systems Neurology (SyNergy), Munich, Germany. 9German Center
for Neurodegenerative Diseases (DZNE), Munich, Germany. 10These authors contributed equally: Mihail Ivilinov Todorov, Johannes Christian Paetzold.
11These authors jointly supervised this work: Bjoern Menze, Ali Ertürk. e-mail:;
Changes in cerebrovascular structures are key indicators for a
large number of diseases affecting the brain. Primary angiop-
athies, vascular risk factors (for example, diabetes), traumatic
brain injury, vascular occlusion and stroke all affect the function of
the brain’s vascular network13. The hallmarks of Alzheimers dis-
ease, including tauopathy and amyloidopathy, can also lead to aber-
rant remodeling of blood vessels1,4, allowing capillary rarefaction to
be used as a marker for vascular damages5. Therefore, quantitative
analysis of the entire brain vasculature is pivotal to developing a
better understanding of brain function in physiological and patho-
logical states. However, quantifying micrometer-scale changes in
the cerebrovascular network of the brain has been difficult for two
main reasons.
First, labeling and imaging of the complete mouse brain vascu-
lature down to the smallest blood vessels has not yet been achieved.
Magnetic resonance imaging (MRI), micro-computed tomography
(micro-CT) and optical coherence tomography do not have suffi-
cient resolution to capture capillaries in bulk tissue68. Fluorescent
microscopy provides higher resolution, but can typically only
be applied to tissue sections up to 200 μm in thickness9. Recent
advances in tissue clearing could overcome this problem10, but so
far there has been no systematic description of all vessels of all sizes
in an entire brain in three dimensions (3D).
The second challenge relates to the automated analysis of large
3D imaging datasets with substantial variance in signal intensity
and signal-to-noise ratio (SNR) at different depths. Simple inten-
sity- and shape-based filtering approaches such as Frangi’s vessel-
ness filters and more advanced image processing methods with
local spatial adaptation cannot reliably differentiate vessels from
background in whole-brain scans11,12. Finally, imaging of the com-
plete vascular network of the brain at capillary resolution results
in datasets of terabyte size. Established image processing methods
do not scale well to terabyte-sized image volumes, as they do
not generalize well to large images, and require intensive manual
Here we present VesSAP (Vessel Segmentation & Analysis
Pipeline), a deep learning-based method for automated analysis of
the entire mouse brain vasculature, overcoming the above limita-
tions. VesSAP encompasses three major steps: (1) staining, clearing
and imaging of the mouse brain vasculature down to the capil-
lary level with two different dyes: wheat germ agglutinin (WGA)
and Evans blue (EB); (2) automatic segmentation and tracing of
the whole-brain vasculature data via CNNs; and (3) extraction of
vascular features for hundreds of brain regions after registration of
the data to the Allen brain atlas (Fig. 1). Our deep learning-based
approach for network extraction in cleared tissue is robust, despite
variations in signal intensities and structures, outperforms previ-
ous filter-based methods and reaches the quality of segmentation
achieved by human annotators. We applied VesSAP to the three
commonly used mouse strains C57BL/6J, CD1 and BALB/c.
Vascular staining, DISCO clearing and imaging. To reliably stain
the entire vasculature, we used WGA and EB dyes, which can be
visualized in different fluorescence channels. We injected EB dye
into live mice 12 h before WGA perfusion, allowing its long-term
circulation to mark vessels under physiological conditions16, while
we perfused mice with WGA during fixation. We then performed
Machine learning analysis of whole mouse brain
Mihail Ivilinov Todorov1,2,3,10, Johannes Christian Paetzold4,5,6,10, Oliver Schoppe4,5, Giles Tetteh4,
Suprosanna Shit4,5,6, Velizar Efremov4,7, Katalin Todorov-Völgyi2, Marco Düring2,8, Martin Dichgans2,8,9,
Marie Piraud4, Bjoern Menze 4,5,6,11 ✉ and Ali Ertürk 1,2,8,11 ✉
Tissue clearing methods enable the imaging of biological specimens without sectioning. However, reliable and scalable analysis
of large imaging datasets in three dimensions remains a challenge. Here we developed a deep learning-based framework to
quantify and analyze brain vasculature, named Vessel Segmentation & Analysis Pipeline (VesSAP). Our pipeline uses a con-
volutional neural network (CNN) with a transfer learning approach for segmentation and achieves human-level accuracy. By
using VesSAP, we analyzed the vascular features of whole C57BL/6J, CD1 and BALB/c mouse brains at the micrometer scale
after registering them to the Allen mouse brain atlas. We report evidence of secondary intracranial collateral vascularization
in CD1 mice and find reduced vascularization of the brainstem in comparison to the cerebrum. Thus, VesSAP enables unbiased
and scalable quantifications of the angioarchitecture of cleared mouse brains and yields biological insights into the vascular
function of the brain.
NATURE METHODS | VOL 17 | APRIL 2020 | 442–449 |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
... This shows that the segmentation approach is indeed transferable with respect to extracting the brain vasculature, but that further processing/improvement is required to remove nonspecific signals, such as those from the skin. v. Lastly, due to the increasing importance and prevalence of deep learning (DL) methods in biological sciences, particularly in cerebrovascular segmentation of preclinical models (Daetwyler et al., Biella, 2021;Todorov et al., 2020), there is a need to understand DL methods and their applicability to biomedical imaging data. We highlight several DL methods and show how these can be assessed against each other to understand their applicability and performance; specifically, the original U-Net (Ronneberger, Fischer, & Brox, 2015), SegNet (Badrinarayanan, Kendall, & Cipolla, 2017), and three modified versions of the original U-Net architecture (dU-Net). ...
... In this case, the vessel crosssectional intensities are single peaks with radial Gaussian-like distributions. This contrasts with biomedical image analysis, such as zebrafish transgenic reporter lines or mouse antibody stainings, where the cells that make up the vessel walls-called endothelial cells-are visualized (Kugler et al., 2018;Todorov et al., 2020). This leads to a cross-sectional double-peak (or ring-shaped signal) intensity distribution in lumenized vessels, whereas unlumenized (often very small) vessels display single peaks (Kugler et al., 2018;Todorov et al., 2020). ...
... This contrasts with biomedical image analysis, such as zebrafish transgenic reporter lines or mouse antibody stainings, where the cells that make up the vessel walls-called endothelial cells-are visualized (Kugler et al., 2018;Todorov et al., 2020). This leads to a cross-sectional double-peak (or ring-shaped signal) intensity distribution in lumenized vessels, whereas unlumenized (often very small) vessels display single peaks (Kugler et al., 2018;Todorov et al., 2020). Although this could be addressed experimentally by performing microangiography, this is (a) laborious and (b) only shows perfused vessels. ...
Full-text available
With advancements in imaging techniques, data visualization allows new insights into fundamental biological processes of development and disease. However, although biomedical science is heavily reliant on imaging data, interpretation of datasets is still often based on subjective visual assessment rather than rigorous quantitation. This overview presents steps to validate image processing and segmentation using the zebrafish brain vasculature data acquired with light sheet fluorescence microscopy as a use case. Blood vessels are of particular interest to both medical and biomedical science. Specific image enhancement filters have been developed that enhance blood vessels in imaging data prior to segmentation. Using the Sato enhancement filter as an example, we discuss how filter application can be evaluated and optimized. Approaches from the medical field such as simulated, experimental, and augmented datasets can be used to gain the most out of the data at hand. Using such datasets, we provide an overview of how biologists and data analysts can assess the accuracy, sensitivity, and robustness of their segmentation approaches that allow extraction of objects from images. Importantly, even after optimization and testing of a segmentation workflow (e.g., from a particular reporter line to another or between immunostaining processes), its generalizability is often limited, and this can be tested using double-transgenic reporter lines. Lastly, due to the increasing importance of deep learning networks, a comparative approach can be adopted to study their applicability to biological datasets. In summary, we present a broad methodological overview ranging from image enhancement to segmentation with a mixed approach of experimental, simulated, and augmented datasets to assess and validate vascular segmentation using the zebrafish brain vasculature as an example. © 2022 The Authors. Current Protocols published by Wiley Periodicals LLC.
... Since the concept of tissue oxygenation is tightly linked to the vasculature, we also inspected changes in the vasculature of the spinal cord by means of vascular staining and light sheet fluorescence microscopy (LSFM). By applying a newly developed deep learning-based framework, VesSAP (Vessel Segmentation & Analysis Pipeline) [40], we reveal a reduction in the perfused vascular network in the spinal cord of EAE. ...
... We used the vessel segmentation and analysis pipeline (VesSAP) [40] to quantify the vasculature changes in the lumbar spinal cord by firstly manually defining a mask for the CNS tissue only separating it from the vertebra and surrounding vasculature. Next, we ran the segmentation, preprocessing and feature extraction to obtain the total vessel length (sum of vessel centerline voxels), bifurcation density (sum of segmentation skeleton bifurcations), and average radius of vessels (distance of all centerline voxels to the nearest segmentation mask). ...
... Next, we ran the segmentation, preprocessing and feature extraction to obtain the total vessel length (sum of vessel centerline voxels), bifurcation density (sum of segmentation skeleton bifurcations), and average radius of vessels (distance of all centerline voxels to the nearest segmentation mask). All measures were then corrected by a constant (described as the in vivo space in [40]) to account for shrinkage due to fixation and clearing. ...
Full-text available
Recent studies suggest that metabolic changes and oxygen deficiency in the central nervous system play an important role in the pathophysiology of multiple sclerosis (MS). In our present study, we investigated the changes in oxygenation and analyzed the vascular perfusion of the spinal cord in a rodent model of MS. We performed multispectral optoacoustic tomography of the lumbar spinal cord before and after an oxygen enhancement challenge in mice with experimental autoimmune encephalomyelitis (EAE), a model for MS. In addition, mice were transcardially perfused with lectin to label the vasculature and their spinal columns were optically cleared, followed by light sheet fluorescence microscopy. To analyze the angioarchitecture of the intact spine, we used VesSAP, a novel deep learning-based framework. In EAE mice, the spinal cord had lower oxygen saturation and hemoglobin concentration compared to healthy mice, indicating compromised perfusion of the spinal cord. Oxygen administration reversed hypoxia in the spinal cord of EAE mice, although the ventral region remained hypoxic. Additionally, despite the increased vascular density, we report a reduction in length and complexity of the perfused vascular network in EAE. Taken together, these findings highlight a new aspect of neuroinflammatory pathology, revealing a significant degree of hypoxia in EAE in vivo that is accompanied by changes in spinal vascular perfusion. The study also introduces optoacoustic imaging as a tractable technique with the potential to further decipher the role of hypoxia in EAE and to monitor it in MS patients.
... ultrasound, due to the lack of resolution necessary to visualize the vast network of microvessels extending beyond the major arterial branches of the circle of Willis. [9][10][11] Additionally, in vivo twophoton imaging and in vivo optical coherence tomography have limited imaging depth. 12,13 Although immunohistochemical and electron microscopy studies have also revealed crucial insights into the NVU major players, these modalities are often limited to localized brain regions, as they are not easily applied to whole-brain studies. ...
... 1(e)-1(g)]. 11,32,33 Although, LSFM is referred to as one major imaging category, there are several different types of light sheet microscope configurations. For example, types of LSFM can include selective plane illumination microscopy (SPIM) as well as inverted, multiview, Bessel beam, and stimulation emission depletion variations of SPIM. ...
... 34 Most studies of the vasculature using LSFM have primarily utilized the standard Gaussian form of SPIM. 11,32,34 However, the strengths of Bessel beam SPIM were recently demonstrated for capturing the high-fidelity imaging of the brain vasculature while avoiding the streaking artifacts that result from Gaussian illumination. 35 Importantly, LSFM resolution can vary widely depending on the microscope and objective lens used, but overall this method can achieve submicron cellular resolution imaging for whole threedimensional (3D) volumes. ...
Significance: The cerebrovasculature has become increasingly recognized as a major player in overall brain health and many brain disorders. Although there have been several landmark studies to understand details of these crucially important structures in an anatomically defined area, brain-wide examination of the whole cerebrovasculature, including microvessels, has been challenging. However, emerging techniques, including tissue processing and three-dimensional (3D) microscopy imaging, enable neuroscientists to examine the total vasculature in the entire mouse brain. Aim: Here, we aim to highlight advances in these high-resolution 3D mapping methods including block-face imaging and light sheet fluorescent microscopy. Approach: We summarize latest mapping tools to understand detailed anatomical arrangement of the cerebrovascular network and the organizing principles of the neurovascular unit (NVU) as a whole. Results: We discuss biological insights gained from studies using these imaging methods and how these tools can be used to advance our understanding of the cerebrovascular network and related cell types in the entire brain. Conclusions: This review article will help to understand recent advance in high-resolution NVU mapping in mice and provide perspective on future studies.
... For example, Tahir et al. [19], Haft-Javaherian et al. [20] and Damseh et al. [21] used CNNs for segmentation of brain vasculature in two-or multi-photon images. Kirst et al. [7] and Todorov et al. [22] used CNN to segment mouse brain vessels imaged by light-sheet microscopy. Tetteh et al. [23] introduced DeepVesselNet, which uses 2.5D CNN instead of 3D CNN to segment vessels in 3D angiographic volumes. ...
... However, it does not work well for cavities in thick vessels. We also compare the proposed method with other similar methods, namely OTSU [34], VesSAP [22], 3D-Unet [33] and Vnet [35]. OTSU is a traditional segmentation method, and the other three are deep learning-based methods. ...
Full-text available
The popularity of fluorescent labelling and mesoscopic optical imaging techniques enable the acquisition of whole mammalian brain vasculature images at capillary resolution. Segmentation of the cerebrovascular network is essential for analyzing the cerebrovascular structure and revealing the pathogenesis of brain diseases. Existing deep learning methods use a single type of annotated labels with the same pixel weight to train the neural network and segment vessels. Due to the variation in the shape, density and brightness of vessels in whole-brain fluorescence images, it is difficult for a neural network trained with a single type of label to segment all vessels accurately. To address this problem, we proposed a deep learning cerebral vasculature segmentation framework based on multi-perspective labels. First, the pixels in the central region of thick vessels and the skeleton region of vessels were extracted separately using morphological operations based on the binary annotated labels to generate two different labels. Then, we designed a three-stage 3D convolutional neural network containing three sub-networks, namely thick-vessel enhancement network, vessel skeleton enhancement network and multi-channel fusion segmentation network. The first two sub-networks were trained by the two labels generated in the previous step, respectively, and pre-segmented the vessels. The third sub-network was responsible for fusing the pre-segmented results to precisely segment the vessels. We validated our method on two mouse cerebral vascular datasets generated by different fluorescence imaging modalities. The results showed that our method outperforms the state-of-the-art methods, and the proposed method can be applied to segment the vasculature on large-scale volumes.
... Many recent approaches have been developed for segmenting and analyzing large blood vessel datasets. [14][15][16][17][18][19] Such methodologies show excellent results even for large 3D volumes containing whole brain data. Still, there are two main advantages of Pyvane compared with the other methodologies. ...
... The first is that the default processors of Pyvane have been shown to provide good results for hundreds of images obtained from many different animals in previously published works. 5,13,[20][21][22] This contrasts with the recently developed approaches, which tested the algorithms on only three, 16 five, 14 nine, 19 and fifteen 15 microscopy images. Thus, it is expected that Pyvane can be easily adapted to new datasets. ...
Full-text available
Significance: A growing body of research supports the significant role of cerebrovascular abnormalities in neurological disorders. As these insights develop, standardized tools for unbiased and high-throughput quantification of cerebrovascular structure are needed. Aim: We provide a detailed protocol for performing immunofluorescent labeling of mouse brain vessels, using thin ( 25 μ m ) or thick (50 to 150 μ m ) tissue sections, followed respectively by two- or three-dimensional (2D or 3D) unbiased quantification of vessel density, branching, and tortuosity using digital image processing algorithms. Approach: Mouse brain sections were immunofluorescently labeled using a highly selective antibody raised against mouse Cluster of Differentiation-31 (CD31), and 2D or 3D microscopy images of the mouse brain vasculature were obtained using optical sectioning. An open-source toolbox, called Pyvane, was developed for analyzing the imaged vascular networks. The toolbox can be used to identify the vasculature, generate the medial axes of blood vessels, represent the vascular network as a graph, and calculate relevant measurements regarding vascular morphology. Results: Using Pyvane, vascular parameters such as endothelial network density, number of branching points, and tortuosity are quantified from 2D and 3D immunofluorescence micrographs. Conclusions: The steps described in this protocol are simple to follow and allow for reproducible and unbiased analysis of mouse brain vascular structure. Such a procedure can be applied to the broader field of vascular biology.
... Figure 7-figure supplement 3 provides an example of this potential using manual segmentation on a small patch of the data presented in Figure 7. Given that the manual segmentation of these vessels is relatively simple, albeit arduous, machine learning approaches (Hilbert et al., 2020;Tetteh et al., 2020) also seem promising, as they commonly perform well in visual tasks (LeCun et al., 2015;Rueckert et al., 2016;Zaharchuk et al., 2018) and have successfully been applied to large-scale vasculature segmentations of mouse tissue-cleared data (Todorov et al., 2020). In general, the main challenges for these algorithms include the small vessel size compared to the voxel size, the broad range of vessel diameters, and the high noise levels. ...
Full-text available
The pial arterial vasculature of the human brain is the only blood supply to the neocortex, but quantitative data on the morphology and topology of these mesoscopic arteries (diameter 50-300µm) remains scarce. Because it is commonly assumed that blood flow velocities in these vessels are prohibitively slow, non-invasive time-of-flight MRI angiography (TOF-MRA)-which is well-suited to high 3D imaging resolutions-has not been applied to imaging the pial arteries. Here, we provide a theoretical framework that outlines how TOF-MRA can visualize small pial arteries in vivo, by employing extremely small voxels at the size of individual vessels. We then provide evidence for this theory by imaging the pial arteries at 140-µm isotropic resolution using a 7T MRI scanner and prospective motion correction, and show that pial arteries one voxel-width in diameter can be detected. We conclude that imaging pial arteries is not limited by slow blood flow, but instead by achievable image resolution. This study represents the first targeted, comprehensive account of imaging pial arteries in vivo in the human brain. This ultra-high-resolution angiography will enable the characterization of pial vascular anatomy across the brain to investigate patterns of blood supply and relationships between vascular and functional architecture.
... The capillary network has recently been described thoroughly by multiple teams [Tsai et al. 2009, Xiong et al. 2017, Kirst et al. 2020, Todorov et al. 2020, Ji et al. 2021. Until recently, a challenge upon the description of the capillary network was the size of the capillaries and their density in the parenchyma: technical advances overcame them and allowed us to better describe the capillary bed. ...
BOLD-fMRI is the preferred technique to study brain activity in humans. Recently, functional ultrasound (fUS) was introduced as a novel technique to record blood volume changes while overcoming the poor spatiotemporal resolution of BOLD-fMRI. BOLD-fMRI and fUS both report signals related to changes in blood flow to determine activated brain areas, although through different physical mechanisms. Indeed, neuronal activation systematically triggers an increase in blood flow, through a cascade of cellular and molecular reactions named neurovascular coupling. Thus, the interpretation of vascular-based functional imaging techniques requires a thorough understanding of neurovascular coupling to recover the spatiotemporal dynamics of neuronal activation from blood flow changes. During my PhD, I linked microscopic measurements of neuronal and vascular responses to odor to the mesoscopic signal detected with functional ultrasound imaging in the mouse. My first project evaluates the correspondence between neuronal activation in the olfactory bulb measured with two-photon microscopy and the fUS output in the same co-registered 100x100x200 µm volume, through a variety of odorant stimuli. This study also allowed me to model the neurovascular coupling, between neuronal activation and the increase of red blood cell velocity in nearby capillaries. I showed that, at high stimulus strength, neurovascular coupling is no longer linear with the appearance of a delayed vascular component independent of neural activity. To achieve these results, I developed 'Iliski', an analysis software that is provided in a GitHub repository and is described in detail in a second article. The second part of my thesis focuses on a long-lasting controversy in the BOLD-fMRI field: do neuronally activated areas generate a local transient decrease in brain oxygenation that can be measured inside vessels before the increase in oxygenated blood flow? It has been hypothesized that this oxygen decrease would be spatially more specific than the oxygen increase, and allow a more precise interpretation of the BOLD-fMRI output. Previous studies reported this "initial dip" in oxygenation, but it was mainly measured in acute animal preparations, where neurovascular coupling is impaired by anesthesia and the invasiveness of the preparation. Here we report that the initial dip cannot be seen in chronic animals, whether anesthetized or awake. Overall, this corpus of work helps to better interpret functional imaging techniques based on blood flow changes by shedding light on the microscopic mechanisms underlying the mesoscopic signals.
Our understanding of the cellular composition and architecture of cancer has primarily advanced using 2D models and thin slice samples. This has granted spatial information on fundamental cancer biology and treatment response. However, tissues contain a variety of interconnected cells with different functional states and shapes, and this complex organization is impossible to capture in a single plane. Furthermore, tumours have been shown to be highly heterogenous, requiring large-scale spatial analysis to reliably profile their cellular and structural composition. Volumetric imaging permits the visualization of intact biological samples, thereby revealing the spatio-phenotypic and dynamic traits of cancer. This review focuses on new insights into cancer biology uniquely brought to light by 3D imaging and concomitant progress in cancer modelling and quantitative analysis. 3D imaging has the potential to generate broad knowledge advance from major mechanisms of tumour progression to new strategies for cancer treatment and patient diagnosis. We discuss the expected future contributions of the newest imaging trends towards these goals and the challenges faced for reaching their full application in cancer research.
Full-text available
Astrocytes establish extensive networks via gap junctions that allow each astrocyte to connect indirectly to the vasculature. However, the proportion of astrocytes directly associated with blood vessels is unknown. Here, we quantify structural contacts of cortical astrocytes with the vasculature in vivo. We show that all cortical astrocytes are connected to at least one blood vessel. Moreover, astrocytes contact more vessels in deeper cortical layers where vessel density is known to be higher. Further examination of different brain regions reveals that only the hippocampus, which has the lowest vessel density of all investigated brain regions, harbors single astrocytes with no apparent vascular connection. In summary, we show that almost all gray matter astrocytes have direct contact to the vasculature. In addition to the glial network, a direct vascular access may represent a complementary pathway for metabolite uptake and distribution.
Full-text available
Optical coherence tomography provides volumetric reconstruction of brain structure with micrometer resolution. Gray matter and white matter can be highlighted using conventional and polarization-based contrasts; however, vasculature in ex-vivo fixed brain has not been investigated at large scale due to lack of intrinsic contrast. We present contrast enhancement to visualize the vasculature by perfusing titanium dioxide particles transcardially into the mouse vascular system. The brain, after dissection and fixation, is imaged by a serial optical coherence scanner. Accumulation of particles in blood vessels generates distinguishable optical signals. Among these, the cross-polarization images reveal the vasculature organization remarkably well. The conventional and polarization-based contrasts are still available for probing the gray matter and white matter structures. The segmentation and reconstruction of the vasculature are presented by using a deep learning algorithm. Axonal fiber pathways in the mouse brain are delineated by utilizing the retardance and optic axis orientation contrasts. This is a low-cost method that can be further developed to study neurovascular diseases and brain injury in animal models.
Full-text available
Brain vessel status is a promising biomarker for better prevention and treatment in cerebrovascular disease. However, classic rule-based vessel segmentation algorithms need to be hand-crafted and are insufficiently validated. A specialized deep learning method—the U-net—is a promising alternative. Using labeled data from 66 patients with cerebrovascular disease, the U-net framework was optimized and evaluated with three metrics: Dice coefficient, 95% Hausdorff distance (95HD) and average Hausdorff distance (AVD). The model performance was compared with the traditional segmentation method of graph-cuts. Training and reconstruction was performed using 2D patches. A full and a reduced architecture with less parameters were trained. We performed both quantitative and qualitative analyses. The U-net models yielded high performance for both the full and the reduced architecture: A Dice value of ~0.88, a 95HD of ~47 voxels and an AVD of ~0.4 voxels. The visual analysis revealed excellent performance in large vessels and sufficient performance in small vessels. Pathologies like cortical laminar necrosis and a rete mirabile led to limited segmentation performance in few patients. The U-net outperfomed the traditional graph-cuts method (Dice ~0.76, 95HD ~59, AVD ~1.97). Our work highly encourages the development of clinically applicable segmentation tools based on deep learning. Future works should focus on improved segmentation of small vessels and methodologies to deal with specific pathologies.
Full-text available
The photothrombotic stroke model generates localized and reproducible ischemic infarcts that are useful for studying recovery mechanisms, but its failure to produce a substantial ischemic penumbra weakens its resemblance to human stroke. We examined whether a modification of this approach, confining photodamage to arteries on the cortical surface (artery-targeted photothrombosis), could better reproduce aspects of the penumbra. Following artery-targeted or traditional photothrombosis to the motor cortex of mice, post-ischemic cerebral blood flow was measured using multi-exposure speckle imaging at 6, 48, and 120 h post-occlusion. Artery-targeted photothrombosis produced a more graded penumbra at 48 and 120 h. The density of isolectin B4+ vessels in peri-infarct cortex was similarly increased after both types of infarcts compared to sham at 2 weeks. These results indicate that both models instigated post-ischemic vascular structural changes. Finally, we determined whether the strength of the traditional photothrombotic approach for modeling upper-extremity motor impairments extends to the artery-targeted approach. In adult mice that were proficient in a skilled reaching task, small motor-cortical infarcts impaired skilled-reaching performance for up to 10 days. These results support that artery-targeted photothrombosis widens the penumbra while maintaining the ability to create localized infarcts useful for modeling post-stroke impairments.
Full-text available
Analysis of entire transparent rodent bodies after clearing could provide holistic biological information in health and disease, but reliable imaging and quantification of fluorescent protein signals deep inside the tissues has remained a challenge. Here, we developed vDISCO, a pressure-driven, nanobody-based whole-body immunolabeling technology to enhance the signal of fluorescent proteins by up to two orders of magnitude. This allowed us to image and quantify subcellular details through bones, skin and highly autofluorescent tissues of intact transparent mice. For the first time, we visualized whole-body neuronal projections in adult mice. We assessed CNS trauma effects in the whole body and found degeneration of peripheral nerve terminals in the torso. Furthermore, vDISCO revealed short vascular connections between skull marrow and brain meninges, which were filled with immune cells upon stroke. Thus, our new approach enables unbiased comprehensive studies of the interactions between the nervous system and the rest of the body. © 2018, The Author(s), under exclusive licence to Springer Nature Limited.
Full-text available
U-Net is a generic deep-learning solution for frequently occurring quantification tasks such as cell detection and shape measurements in biomedical image data. We present an ImageJ plugin that enables non-machine-learning experts to analyze their data with U-Net on either a local computer or a remote server/cloud service. The plugin comes with pretrained models for single-cell segmentation and allows for U-Net to be adapted to new tasks on the basis of a few annotated samples. © 2018, The Author(s), under exclusive licence to Springer Nature America, Inc.
Full-text available
We present deep-learning-enabled super-resolution across different fluorescence microscopy modalities. This data-driven approach does not require numerical modeling of the imaging process or the estimation of a point-spread-function, and is based on training a generative adversarial network (GAN) to transform diffraction-limited input images into super-resolved ones. Using this framework, we improve the resolution of wide-field images acquired with low-numerical-aperture objectives, matching the resolution that is acquired using high-numerical-aperture objectives. We also demonstrate cross-modality super-resolution, transforming confocal microscopy images to match the resolution acquired with a stimulated emission depletion (STED) microscope. We further demonstrate that total internal reflection fluorescence (TIRF) microscopy images of subcellular structures within cells and tissues can be transformed to match the results obtained with a TIRF-based structured illumination microscope. The deep network rapidly outputs these super-resolved images, without any iterations or parameter search, and could serve to democratize super-resolution imaging. © 2018, The Author(s), under exclusive licence to Springer Nature America, Inc.
Full-text available
Fluorescence microscopy is a key driver of discoveries in the life sciences, with observable phenomena being limited by the optics of the microscope, the chemistry of the fluorophores, and the maximum photon exposure tolerated by the sample. These limits necessitate trade-offs between imaging speed, spatial resolution, light exposure, and imaging depth. In this work we show how content-aware image restoration based on deep learning extends the range of biological phenomena observable by microscopy. We demonstrate on eight concrete examples how microscopy images can be restored even if 60-fold fewer photons are used during acquisition, how near isotropic resolution can be achieved with up to tenfold under-sampling along the axial direction, and how tubular and granular structures smaller than the diffraction limit can be resolved at 20-times-higher frame rates compared to state-of-the-art methods. All developed image restoration methods are freely available as open source software in Python, FIJI, and KNIME. © 2018, The Author(s), under exclusive licence to Springer Nature America, Inc.
Full-text available
The distinct organization of the brain's vascular network ensures that it is adequately supplied with oxygen and nutrients. However, despite this fundamental role, a detailed reconstruction of the brain-wide vasculature at the capillary level remains elusive, due to insufficient image quality using the best available techniques. Here, we demonstrate a novel approach that improves vascular demarcation by combining CLARITY with a vascular staining approach that can fill the entire blood vessel lumen and imaging with light-sheet fluorescence microscopy. This method significantly improves image contrast, particularly in depth, thereby allowing reliable application of automatic segmentation algorithms, which play an increasingly important role in high-throughput imaging of the terabyte-sized datasets now routinely produced. Furthermore, our novel method is compatible with endogenous fluorescence, thus allowing simultaneous investigations of vasculature and genetically targeted neurons. We believe our new method will be valuable for future brain-wide investigations of the capillary network.
Full-text available
We present DeepVesselNet, an architecture tailored to the challenges to be addressed when extracting vessel networks and corresponding features in 3-D angiography using deep learning. We discuss the problems of low execution speed and high memory requirements associated with full 3-D convolutional networks, high class imbalance arising from low percentage (less than 3%) of vessel voxels, and unavailability of accurately annotated training data - and offer solutions that are the building blocks of DeepVesselNet. First, we formulate 2-D orthogonal cross-hair filters which make use of 3-D context information. Second, we introduce a class balancing cross-entropy score with false positive rate correction to handle the high class imbalance and high false positive rate problems associated with existing loss functions. Finally, we generate synthetic dataset using a computational angiogenesis model, capable of generating vascular networks under physiological constraints on local network structure and topology, and use these data for transfer learning. DeepVesselNet is optimized for segmenting vessels, predicting centerlines, and localizing bifurcations. We test the performance on a range of angiographic volumes including clinical Time-of-Flight MRA data of the human brain, as well as synchrotron radiation X-ray tomographic microscopy scans of the rat brain. Our experiments show that, by replacing 3-D filters with 2-D orthogonal cross-hair filters in our network, speed is improved by 23% while accuracy is maintained. Our class balancing metric is crucial for training the network and pre-training with synthetic data helps in early convergence of the training process.
Full-text available
Intracellular β-amyloid (Aβ) accumulation is an early event in Alzheimer's disease (AD) progression. Recently, it has been uncovered that presenilins (PSs), the key components of the amyloid precursor protein (APP) processing and the β-amyloid producing γ-secretase complex, are highly enriched in a special sub-compartment of the endoplasmic reticulum (ER) functionally connected to mitochondria, called mitochondria-associated ER membrane (MAM). A current hypothesis of pathogenesis of Alzheimer's diseases (AD) suggests that MAM is involved in the initial phase of AD. Since MAM supplies mitochondria with essential proteins, the increasing level of PSs and β-amyloid could lead to metabolic dysfunction because of the impairment of ER-mitochondrion crosstalk. To reveal the early molecular changes of this subcellular compartment in AD development MAM fraction was isolated from the cerebral cortex of 3 months old APP/PS1 mouse model of AD and age-matched C57BL/6 control mice, then mass spectrometry-based quantitative proteome analysis was performed. The enrichment and purity of MAM preparations were validated with EM, LC-MS/MS and protein enrichment analysis. Label-free LC-MS/MS was used to reveal the differences between the proteome of the transgenic and control mice. We obtained 77 increased and 49 decreased protein level changes in the range of - 6.365 to + 2.988, which have mitochondrial, ER or ribosomal localization according to Gene Ontology database. The highest degree of difference between the two groups was shown by the ATP-binding cassette G1 (Abcg1) which plays a crucial role in cholesterol metabolism and suppresses Aβ accumulation. Most of the other protein changes were associated with increased protein synthesis, endoplasmic-reticulum-associated protein degradation (ERAD), oxidative stress response, decreased mitochondrial protein transport and ATP production. The interaction network analysis revealed a strong relationship between the detected MAM protein changes and AD. Moreover, it explored several MAM proteins with hub position suggesting their importance in Aβ induced early MAM dysregulation. Our identified MAM protein changes precede the onset of dementia-like symptoms in the APP/PS1 model, suggesting their importance in the development of AD.