Segmentation of the Zebrafish Brain Vasculature from Light Sheet Fluorescence Microscopy Datasets

Elisabeth C. Kugler, Andrik Rampun, Timothy J.A. Chico, Paul A. Armitage

Preprint posted on 22 July 2020 https://www.biorxiv.org/content/10.1101/2020.07.21.213843v1

Article now published in Current Protocols at http://dx.doi.org/10.1002/cpz1.443

Microscopy and zebrafish: Understanding vasculature– part II.

Selected by Mariana De Niz

Categories: cell biology, developmental biology

Background

Zebrafish are useful models to study vascular biology, due to factors including their high genomic similarity to humans, rapid ex utero development, fecundity, larval transparency, and the availability of multiple reporter lines for multiple sub-cellular structures. Vascular diseases are a leading cause of death worldwide and many factors underlying pathology are currently not fully understood. Imaging modalities such as light sheet fluorescence microscopy (LSFM) allow the acquisition of vascular information in great anatomical detail throughout extended periods of time.

Current challenges in quantifying the zebrafish cranial vasculature include a lack of robust segmentation and quantification approaches. This is because most vascular research in zebrafish has been performed in the trunk vasculature; moreover, the cerebral vasculature is topologically challenging to study. LSFM is a relatively novel technique with great potential for the study of the vasculature in significant detail. However, approaches to validate image analysis methods are broadly lacking. Given that there is no gold standard for vascular segmentation, in their work, Kugler et al aim to overcome limitations in the knowledge about segmentation quality, by performing detailed validation of the segmentation methods using simulated data, manual measurements, and various in vivo biological datasets (1).

Figure 1. Pipeline for Segmentation of the Zebrafish Brain Vasculature from Light Sheet Fluorescence Microscopy Datasets

Key findings and developments

The work begins by exploring vessel enhancement using a shape-based filter with the assumption of local vascular tubularity. The Sato Enhancement filter (SE) already implemented into Fiji was evaluated for its applicability to images of the zebrafish vasculature acquired with LSFM. SE delivered meaningful results if data is presented as bright-on-dark. Applying SE to the simulated hollow tubes converted these to filled tubes when enhancement scale was approximately at the size of tubes. SE was able to convert double-to-single peak intensity distributions if applied at the scale of tubes, suggesting that lumenized and unlumenized vessels would both be enhanced similarly. However, there was a tendency for the simulated tube width to appear broader after enhancement, so any subsequent segmentation would need to be tuned to recover the correct vessel width.

To evaluate segmentation accuracy, vessel diameters obtained after applying the proposed enhancement and segmentation methodologies were compared to those obtained by the ‘gold standard’ manual measurement. When comparing full-width-half-maximum (FWHM) with the manual diameter measurements, no statistically significant differences were found. The authors conclude that the FWHM can be used to measure vascular diameter in zebrafish reporter lines, although outliers need to be considered. Upon comparing FWHM and manual measurements to general filtering and SE after thresholding, no significant differences were found. No segmentation method introduced significant artificial bias.

Segmentation robustness was assessed by processing data with varying signal properties. Two approaches were used for evaluation, namely data acquisition with a controlled decrease in image quality, and data augmentation by artificial noise addition. CNR was successfully decreased with this approach. Both were followed by segmentation after using general filtering methods (GF) and SE. The authors conclude that both segmentation approaches were robust over a broad range of contrast-to-noise ratio (CNR) levels

Segmentation sensitivity was assessed by using datasets with a predictable vascular volume difference via image acquisition before and after exanguination. No statistical difference was found in cerebral vascular volume after GF between control and exanguinated samples. SE was more sensitive, allowing extraction of true biological differences. In summary, evaluation of all segmentation approaches indicated that SE-based segmentation is more successful than the GF-based approach.

The authors then went on to quantify cerebral vascular volume throughout zebrafish development between days 3 and 5 post fertilization, using the transgenic reporter line Tg(kdrl:HRAS-mCherry)s916. The data suggested that SE-processed data were less variable and therefore lower sample numbers would be required to extract biologically relevant vascular volumes. The authors then evaluated whether this approach could be generalizable to other transgenic lines. They found that segmentation in reporter lines under the fli1a promoter displayed, in addition to vascular information, enhancement and segmentation of non-vascular signal. Quantification of the cerebral vascular volume showed a significantly higher vascular volume under the fli1a promoter in all examined reporter lines. Altogether, the authors conclude that vascular volume can be extracted in other transgenic lines, but accuracy and precision are reduced. They suggest that future work might include pre-processing to enhance vascular and decrease non-vascular signal, exclusion of non-connected components, exclusion of objects under a specified size-threshold, or exclusion based on peripheral position.

The authors then aimed to use deep learning approaches for segmentation of zebrafish vascular data from the single transgenic Tg(kdrl:HRAS-mCherry)s916. For this, they trained an original U-Net, SegNet, and three modified U-Nets (dU-Net1-3) on the training dataset using segmented masks obtained after SE as the ground truth. The resulting trained network was then applied to the evaluation dataset. U-Net architectures seemed to deliver better results than SegNet, which over-segmented the vasculature. The original U-Net and dU-Net1 delivered the best results. There was a tendency for the deep learning methods to systematically over-estimate the vascular volume. The deep learning approaches used in this study result in typical run times of between 7 and 23 minutes for segmenting a full stack. This was less than half the time required using SE-based segmentation, suggesting a time benefit for using deep learning methods.

What I like about this preprint

The authors systematically tested and compared segmentation methods for studying the zebrafish vasculature. I find this is a useful study and an important step forward for a current hindrance which is data analysis. I imagine advances like this will be a positive feedback loop for tools such as LSFM.

References

1. Kugler EC, et al Segmentation of the zebrafish brain vasculature from light sheet fluorescence microscopy datasets, bioRxiv, 2020.

Posted on: 3 September 2020

doi: https://doi.org/10.1242/prelights.24422

Read preprint

(No Ratings Yet)

Author's response

Elisabeth Kugler, Timothy Chico and Paul Armitage shared

Open questions

1. Did you test and/or found significant differences upon segmenting the vasculature in LSFM acquired samples as opposed to other forms of imaging (eg. confocal/ tissue sections)?

We have optimized our approach for LSFM data, but future work might expand our approach to confocal data. We think that it is crucial to spend time on understanding different data properties and optimizing analysis approaches depending on their application. Unfortunately, there is no “one size fits all”.

2. Did you see any difference in the segmentation and comparison outputs if you use different reporters/tags/dyes?

To examine exactly this, we here examined four different transgenic lines using the following three double-transgenics: (1) Tg(ﬂi1a:eGFP)y1, Tg(kdrl:HRAS-mCherry)s916, (2) Tg(fli1a:CAAX-eGFP), Tg(kdrl:HRAS-mCherry)s916, and (3)Tg(fli1a:LifeAct-mClover)sh467, Tg(kdrl:HRAS-mCherry)s916. We hypothesised that signal driven under the fli1a promotor would be more challenging to segment as we previously found these to have lower vascular specificity, lower CNRs, and higher image artefact levels.

Future work will focus on how we can remove non-vascular information (e.g. unspecific skin signal) from our data, either during pre- or post-processing, as well as how our approach performs on other data, such as dyes.

3. How reproducible are the comparisons you present here with what to expect in other animal models, for instance optically cleared rodents?

Our method achieves a very high reproducibility/accuracy, allowing us to to study subtle biological differences in the brain volume during development and experimental conditions.

To our knowledge, very few studies focus solely on validating suggested enhancement and segmentation approaches. As all image segmentation methods are specific to their biological question, it is difficult to directly compare different approaches, or compare other model organisms. It would be great to compare different analysis methods in different models thoroughly, which is something we would like to do in the future.

4. Beyond the factors you explored here, what other on factors that might require optimization depending on the experimental setup (eg. other reporter fish lines, or fluorophores, or promoters) do you suggest for users who are interested in using LSFM to study the vasculature?

LSFM is a unique technique, allowing unrivalled insights into vascular development and disease. There are many things one should consider when thinking about LSFM. Personally, I would say that the most important consideration is how people will actually store, handle, and analyse their data as LSFM datasets are rather large. This is of particular importance if wanting to perform quantitative and automized image analysis.

Have your say Cancel reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sign up to customise the site to your preferences and to receive alerts

Also in the cell biology category:

Clusters of lineage-specific genes are anchored by ZNF274 in repressive perinucleolar compartments

Martina Begnis, Julien Duc, Sandra Offner, et al.

Selected by Silvia Carvalho

Structural basis of respiratory complexes adaptation to cold temperatures

Young-Cheul Shin, Pedro Latorre-Muro, Amina Djurabekova, et al.

Selected by Pamela Ornelas

RIPK3 coordinates RHIM domain-dependent inflammatory transcription in neurons

Sigal B. Kofman, Lan H. Chu, Joshua M. Ames, et al.

Selected by Zoie Magri

Also in the developmental biology category:

Temporal constraints on enhancer usage shape the regulation of limb gene transcription

Raquel Rouco, Antonella Rauseo, Guillaume Sapin, et al.

Selected by María Mariner-Faulí

OGT prevents DNA demethylation and suppresses the expression of transposable elements in heterochromatin by restraining TET activity genome-wide

Hugo Sepulveda, Xiang Li, Xiaojing Yue, et al.

Selected by Mansi

Actin polymerization drives lumen formation in a human epiblast model

Dhiraj Indana, Andrei Zakharov, Youngbin Lim, et al.

Selected by Megane Rayer, Rivka Shapiro

preLists in the cell biology category:

BSCB-Biochemical Society 2024 Cell Migration meeting

This preList features preprints that were discussed and presented during the BSCB-Biochemical Society 2024 Cell Migration meeting in Birmingham, UK in April 2024. Kindly put together by Sara Morais da Silva, Reviews Editor at Journal of Cell Science.

Segmentation of the Zebrafish Brain Vasculature from Light Sheet Fluorescence Microscopy Datasets

Background

Key findings and developments

What I like about this preprint

References

Share this:

Have your say Cancel reply

Sign up to customise the site to your preferences and to receive alerts

Also in the cell biology category:

Clusters of lineage-specific genes are anchored by ZNF274 in repressive perinucleolar compartments

Structural basis of respiratory complexes adaptation to cold temperatures

RIPK3 coordinates RHIM domain-dependent inflammatory transcription in neurons

Also in the developmental biology category:

Temporal constraints on enhancer usage shape the regulation of limb gene transcription

OGT prevents DNA demethylation and suppresses the expression of transposable elements in heterochromatin by restraining TET activity genome-wide

Actin polymerization drives lumen formation in a human epiblast model

preLists in the cell biology category:

BSCB-Biochemical Society 2024 Cell Migration meeting

‘In preprints’ from Development 2022-2023

preLights peer support – preprints of interest

The Society for Developmental Biology 82nd Annual Meeting

CSHL 87th Symposium: Stem Cells

Journal of Cell Science meeting ‘Imaging Cell Dynamics’

9th International Symposium on the Biology of Vertebrate Sex Determination

Alumni picks – preLights 5th Birthday

CellBio 2022 – An ASCB/EMBO Meeting

Fibroblasts

EMBL Synthetic Morphogenesis: From Gene Circuits to Tissue Architecture (2021)

FENS 2020

Planar Cell Polarity – PCP

BioMalPar XVI: Biology and Pathology of the Malaria Parasite

Cell Polarity

TAGC 2020

3D Gastruloids

ECFG15 – Fungal biology

ASCB EMBO Annual Meeting 2019

EMBL Seeing is Believing – Imaging the Molecular Processes of Life

Autophagy

Lung Disease and Regeneration

Cellular metabolism

BSCB/BSDB Annual Meeting 2019

MitoList

Biophysical Society Annual Meeting 2019

ASCB/EMBO Annual Meeting 2018

Also in the developmental biology category:

GfE/ DSDB meeting 2024

‘In preprints’ from Development 2022-2023

preLights peer support – preprints of interest

The Society for Developmental Biology 82nd Annual Meeting

CSHL 87th Symposium: Stem Cells

Journal of Cell Science meeting ‘Imaging Cell Dynamics’

9th International Symposium on the Biology of Vertebrate Sex Determination

Alumni picks – preLights 5th Birthday

CellBio 2022 – An ASCB/EMBO Meeting

2nd Conference of the Visegrád Group Society for Developmental Biology

Fibroblasts

EMBL Synthetic Morphogenesis: From Gene Circuits to Tissue Architecture (2021)

EMBL Conference: From functional genomics to systems biology

Single Cell Biology 2020

Society for Developmental Biology 79th Annual Meeting

FENS 2020

Planar Cell Polarity – PCP

Cell Polarity

TAGC 2020

3D Gastruloids

ASCB EMBO Annual Meeting 2019

EDBC Alicante 2019

EMBL Seeing is Believing – Imaging the Molecular Processes of Life

SDB 78th Annual Meeting 2019

Lung Disease and Regeneration

Young Embryologist Network Conference 2019

Pattern formation during development

BSCB/BSDB Annual Meeting 2019

Zebrafish immunology