F1000ResearchF1000Research2046-1402F1000 Research LimitedLondon, UK10.12688/f1000research.26977.1Research ArticleArticlesTAPAS: Towards Automated Processing and Analysis of multi-dimensional bioimage data
[version 1; peer review: 2 approved with reservations]
GillesJean-FrançoisSoftwareValidation1BoudierThomasConceptualizationMethodologySoftwareSupervisionWriting – Original Draft PreparationWriting – Review & Editinga12
Sorbonne Universite, Paris, France
Academia Sinica, Taiwan, Taipei, Taiwanthomas.boudier@sorbonne-universite.fr
Modern microscopy is based on reproducible quantitative analysis, image data should be batch-processed by a standardized system that can be shared and easily reused by others. Furthermore such system should require none or minimal programming from the users.
We developed TAPAS (Towards an Automated Processing and Analysis System). The goal is to design an easy system for describing and exchanging processing workflows. The protocols are simple text files comprising a linear list of commands used to process and analyse the images. An extensive set of 60 modules is already available, mostly based on the tools proposed in the 3D ImageJ Suite.
We propose a wizard, called TAPAS menu, to help the user design her protocol by listing the available modules and the parameters associated. Most modules will have default parameters values for most common tasks. Once the user has designed her protocol, she can apply the protocol to a set of images, that can be either stored locally or on a OMERO database.
An extensive documentation including the list of modules, various tutorials and link to the source code is available at
https://imagej.net/TAPAS.
Image processingimage analysisautomationOMEROImageJFijiThe author(s) declared that no grants were involved in supporting this work.Introduction
Modern microscopy, through new systems like light-sheet or high-throughput microscopes, is generating a vast amount of complex data that needs to be analysed. These data can be large in size or in number. Furthermore, for the purpose of reproducible quantitative analysis, these data should be batch-processed by a standardized system, that can be easily shared and reused. Some batch systems already exist such as CellProfiler
1, ICY
2 protocols or ImageJ macros
3. However, these systems may require some programming knowledge or time to set up for inexperienced users or are not yet fully multi-dimensional.
In the last 10–20 years, and more recently with deep-learning methods, a lot has been accomplished in the field of image processing, especially for image segmentation. However, there is no real standardization for image analysis protocols. Arganda-Carreras and Andrey
4 designed a first version of a systematic image analysis pipeline. Furthermore, due to the recent advances in fast volumetric microscopy, more and more data are produced, but it lacks a systematic way of organizing raw data and subsequent analysed data and results. With the spread of database systems such as OMERO
5, more and more imaging facilities and labs are storing their data in a more organized fashion.
Methods
We developed TAPAS (Towards an Automated Processing and Analysis System) as a system for describing and exchanging processing workflows. The protocols are simple text files comprising a linear sequence of commands. An extensive set of 60 commands is already available, mostly based on the 3D ImageJ Suite
6,
7. The design of the protocol allows simplified tracking of processed data and quantitative results, by using keywords to design the image data, such as ?
image?.
TAPAS is focusing on data organization rather than complex segmentation or analysis algorithms. TAPAS focused originally on data stored on an OMERO database, by allowing to retrieve, perform classical segmentation procedures and analysis, and push back the results, both images and tables, to the database. Data on OMERO are, by design, organized by user, then
projects and
datasets. In TAPAS the current analysed image is simply referred by the keyword
?image?, and the corresponding project and dataset the data belongs to by
?project? and
?dataset? respectively. Subsequent processed data are then referred as
?image?-processing, for instance
?image?-nucleus for the result of nucleus segmentation. Similarly, additional datasets can be created such
?dataset?-labels to store the results of segmentation. The results tables can be stored using the name of the image as reference such as
?image?-nucleus-volume.csv. Results tables will be linked to the original raw image using OMERO
attachments.
Implementation
The system is implemented in Java, with a core library, including OMERO and BioFormats input/output utilities, and a plugins library including a comprehensive set of modules. Each module is generic as it will process a generic
Image class, and each class will get an
Image as input and will return an
Image as output. Parameters are managed as simple
String files, allowing flexible management of parameters, even for inexperienced Java programmers. The current system uses the ImageJ
ImagePlus class as implementation for the
Image class, however any other class can be used, allowing the use of any Java library.
The processing pipeline is constructed as an ordered list of processing classes, with their corresponding parameters. Since the classes are generic, a processor is also specified as how to process the image data, by default an
ImagePlus processor is built. An experimental version of processor with a set of processing modules using the
clearCLBuffer class has been tested and validated using the CLIJ system
8.
Operation
TAPAS is java-based and works with the ImageJ/Fiji system, the use of a OMERO database is optional. TAPAS is designed to work with a database; either OMERO or a local database. A local database is a folder organized, not unlike OMERO, as projects, datasets, and finally images. We also add an attachments' folder to store the results tables. Having a completely similar organization between OMERO and a local database allows to have an exact same protocol to run using an OMERO database or a local database. A typical workflow of the system is presented in
Figure 1.
Flowchart of the TAPAS system.
1) The data to be processed is
input into the processing pipeline from the Database (either OMERO or local).
?image? is a keyword used to refer to the image name. The names in boxes refer to the module names.
2) The necessary data to be used later is saved locally, in a temporary folder (home folder, ImageJ/Fiji folder or system temporary folder). Here we saved the raw data for channel 1.
3) The data is processed, here a classical pipeline consisting of filtering, thresholding and labelling.
4) The resulting labelled data is
output to the Database, here the labelled structure for channel 1 is the nucleus.
5) The previously saved raw data is used as parameter to quantify intensity inside the labelled nuclei. The results table is saved first in a file locally.
6) The results table file is then
attached to the original processed image. The temporary saved data (raw data for channel 1 and results table file) can be then deleted within the pipeline or manually.
Use case
The system separates the data to be processed from the processing pipeline. Firstly, the list of image data to be processed is built, each image data to be processed is identified by its project, dataset and name (either on OMERO or on a local DB), and by the channel and frame to be processed. Second, the processing pipeline file is to be selected. After clicking
run, the system will process the images sequentially, displaying information for each module, and the final processing time per image. Raw data will be
pulled from the database and processed and analysed data will be
pushed back to the database.
We propose a simple TAPAS
menu that will display in an organized manner the list of available modules with their corresponding category and documentation. After selecting a module, the list of parameters will be displayed, the user can then manually enter the parameters values, and the corresponding processing pipeline text will be created.
Conclusion
TAPAS is a comprehensive system for data processing automation, relying on an extensive set of more than 60 modules for processing and analysis of multi-dimensional image data. An extensive documentation including the list of modules, various tutorials and links to the source code is available at
https://imagej.net/TAPAS.
Data availability
All data underlying the results are available as part of the article and no additional source data are required.
Software availability
Software available from:
https://imagej.net/TAPAS.
Source code available from:
https://github.com/mcib3d/tapas-core/.
Archived source code at time of publication:
http://doi.org/10.5281/zenodo.40911779.
License: GPL 3.0
Acknowledgements
This publication was supported by COST Action NEUBIAS (CA15124), funded by COST (European Cooperation in Science and Technology)
BrayMACarpenterAE:
CellProfiler Tracer: exploring and validating high-throughput, time-lapse microscopy image data.BMC Bioinformatics.2015;16(1):368.
2653730010.1186/s12859-015-0759-x4634901de ChaumontFDallongevilleSChenouardN:
Icy: an open bioimage informatics platform for extended reproducible research.Nat Methods.2012;9(7):690–6.
2274377410.1038/nmeth.2075SchneiderCARasbandWSEliceiriKW:
NIH Image to ImageJ: 25 years of image analysis.Nat Methods.2012;9(7):671–5.
2293083410.1038/nmeth.20895554542Arganda-CarrerasIAndreyP:
Designing Image Analysis Pipelines in Light Microscopy: A Rational Approach.Methods Mol Biol.2017;1563:185–207.
2832461010.1007/978-1-4939-6810-7_13AllanCBurelJMMooreJ:
OMERO: flexible, model-driven data management for experimental biology.Nat Methods.2012;9(3):245–53.
2237391110.1038/nmeth.1896 3437820BoudierT:
3D ImageJ Suite [Internet]. 2012; [cited 2016 May 23].
Reference SourceBoudierT:
Introduction to 3D Analysis with 3D ImageJ Suite.F1000Res.2020; [cited 2020 Jul 17]9.
10.7490/f1000research.1118016.1HaaseRRoyerLASteinbachP:
CLIJ: GPU-accelerated image processing for everyone.Nat Methods.2019;17(1):1–2.
3174082310.1038/s41592-019-0650-1BoudierT:
mcib3d/tapas-core 0.6.4 (Version 0.6.4).Zenodo. 2020.
http://www.doi.org/10.5281/zenodo.409117710.5256/f1000research.29799.r73800Reviewer response for version 1LooLit-Hsin1Refereehttps://orcid.org/0000-0001-6303-9840
Bioinformatics Institute (BII), Agency for Science, Technology, and Research (A*STAR), Singapore, Singapore
Competing interests: No competing interests were disclosed.
Gilles
et al. report a new software tool called TAPAS for batch processing and analysis of microscopy images. The tool is based on ImageJ/Fiji and can load images from the local file system or remote OMERO database. A set of 60 analysis modules or commands are currently available, mostly come from a previously developed package, 3D ImageJ Suite.
The ability to batch process images using a standardized system is critical for ensuring the reproducibility of published results, and also the continuity of the development and applications of established analysis pipelines. The authors correctly point out that there are many existing tools that can do batch processing, such as CellProfiler, Icy, or ImageJ macros. However, one of the key advantages of TAPAS is the simplification of the specifications of an analysis workflow using text files (which can be edited using any text editor) and intuitive command syntax (which require minimum effort to learn). Thus, TAPAS can still fill a gap currently unaddressed by these other tools in the field.
Overall, the work is technically sound and the article is well written. However, insufficient details and examples or use cases have been given to demonstrate the applications of TAPAS. Also, the targeted applications and the current limitations of the software are not well explained or discussed.
Major comments:
For this reviewer, the syntax of TAPAS is one of the most interesting and novel parts of the software. However, insufficient details and examples have been provided to explain the syntax and rules of the “TAPAS language”. One suggestion is to provide an example for a typical image analysis pipeline (e.g., image import, preprocessing, segmentation, feature extraction, result export). The authors may include pseudo codes (or a block diagram) and the corresponding TAPAS code that can accomplish the tasks.
It is not very clear what types of microscopy images can TAPAS process. The authors kind of alluded that the tools is meant for 3D images, but this has not been clearly presented. For example, can TAPAS process bright-field or RGB images? Time-lapse images? Or 3D-stack images?
Can TAPAS run any ImageJ commands/functions? Or is it restricted only to functions from the ~60 modules?
The authors may present some simple examples of the results and screenshots that can be obtained from TAPAS. For example, counting of cells or measurement of protein expression levels in 3D images. These examples will be very helpful to let the readers understand how TAPAS may be used.
Another major advantage of using text-file based configurations is that the configurations may be version-controlled using standard software such as Git. Perhaps the authors may consider adding this function in the future.
Minor comments
In the abstract, “Once the user has designed her protocol, she can … “ may be changed to “he/she”.
Is the work clearly and accurately presented and does it cite the current literature?
Yes
If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
Are all the source data underlying the results available to ensure full reproducibility?
Yes
Is the study design appropriate and is the work technically sound?
Yes
Are the conclusions drawn adequately supported by the results?
Partly
Are sufficient details of methods and analysis provided to allow replication by others?
Partly
Reviewer Expertise:
computational and system biology, image analysis, high-throughput screening and imaging, machine learning, in vitro cell models
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
10.5256/f1000research.29799.r73799Reviewer response for version 1HaaseRobert1Refereehttps://orcid.org/0000-0001-5949-2327
Cluster of Excellence "Physics of Life", Technische Universität Dresden, Dresden, Germany
Competing interests: I do work on software solutions with similar purpose as TAPAS. I hope anyway that my review is appropriate and my feedback allows the authors to improve the manuscript.
The authors present TAPAS which is an approach Towards an Automated Processing and Analysis System. One can see it as a workflow-design tool, primarily, but not exclusively, to the ImageJ 3D Suite.
The authors' efforts are appreciated because TAPAS fills a gap in the ImageJ/Fiji ecosystem. Furthermore, TAPAS is extensible and user friendly. TAPAS allows the removal of technical implementation details from the process of designing an image processing workflow. Furthermore, users are suggested to organize their data in a structure similar to OMERO databases. Standardization in image data organization is also a field where every step forward is deeply acknowledged.
For the presented manuscript, I would suggest minor additions: First of all, a more detailed report of the state-of-the-art would be interesting. Differentiation of TAPAS against Icy Protocols
1 and Knime
2 could give new insights to the reader. Furthermore, also commercial systems such as Zen and Apeer (Zeiss, Jena, Germany) and Arivis (Arivis, Munich, Germany) come nowadays with workflow design user interfaces which might be worth exploring and distinguishing.
From the user's perspective, I would love to see an example workflow and an explanation of how the user benefits from exploiting the capabilities of TAPAS in a practical context. The TAPAS website contains examples and tutorials and the manuscript could benefit from this amazing documentation.
On the implementation side, I would happy to read some insights about how plugins for TAPAS can be implemented and distributed to collaborators. I think if the involved procedures would be explained a bit more in detail, developers could estimate efforts to implement TAPAS wrappers for their tools and algorithms, e.g. for denoising and segmentation. This increases the chance of other developers picking up TAPAS as a distribution system of their tools.
Furthermore, on the technical side, I wonder if it is possible to implement processing steps that take two images from two former processing steps as input. Figure 1 suggests TAPAS is a linear system where each operation has just one input and one output. I'm not sure if I interpret this correctly.
The fact that TAPAS is developed fully in the open endorses the authors as fantastic contributors to the bio-image analysis open-source community. The audience knows from former projects like the authors' famous ImageJ 3D Suite that long-term support is available for many years. I think potential future plans for TAPAS could be mentioned in the manuscript. To my experience, communicating long-term perspectives clearly to the community pays off, because if users know that long-term support is given, the chance is higher that they pick up a new tool and build their workflows with it.
To conclude, TAPAS is a great extension for ImageJ/Fiji and eases integration of workflows in the OMERO ecosystem. It is extensible and flexible in use giving a powerful toolbox in the hands of end-users.
Is the work clearly and accurately presented and does it cite the current literature?
Partly
If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
Are all the source data underlying the results available to ensure full reproducibility?
No source data required
Is the study design appropriate and is the work technically sound?
Yes
Are the conclusions drawn adequately supported by the results?
Yes
Are sufficient details of methods and analysis provided to allow replication by others?
Partly
Reviewer Expertise:
Bio-Image Analysis, Computer Science
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
References:
Icy: an open bioimage informatics platform for extended reproducible research.Nat Methods.2012;9(7) :
10.1038/nmeth.2075690-62274377410.1038/nmeth.2075:
KNIME: The Konstanz Information Miner.2008;
10.1007/978-3-540-78246-9_38319-32610.1007/978-3-540-78246-9_38