BIAFLOWS: A collaborative framework to reproducibly deploy and benchmark bioimage analysis workflows

Rubens, Ulysse; Mormont, Romain; Baecker, Volker; Michiels, Gino; Paavolainen, Lassi; Ball, Graeme; Ünay, Devrim; Pavie, Benjamin; Chessel, Anatole; Scholz, Leandro Aluisio; Maška, Martin; Hoyoux, Renaud; Vandaele, Rémy; Stanciu, Stefan G.; Golani, Ofra; Sladoje, Nataša; Paul‐Gilloteaux, Perrine; Marée, Raphaël; Tosi, Sébastien

doi:10.1101/707489

Cited by 6 publications

(6 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More practically, we plan to integrate newly released datasets into our pool to reinforce its versatility and hopefully the transferability of the pre-trained models. We also plan to integrate the pre-trained models to the Cytomine open-source tool [39] and the BIAFLOWS benchmarking platform [51].…”

Section: Discussion and Future Workmentioning

confidence: 99%

Multi-Task Pre-Training of Deep Neural Networks for Digital Pathology

Mormont

Geurts

Marée

2021

IEEE J. Biomed. Health Inform.

Self Cite

View full text Add to dashboard Cite

In this work, we investigate multi-task learning as a way of pre-training models for classification tasks in digital pathology. It is motivated by the fact that many small and medium-size datasets have been released by the community over the years whereas there is no large scale dataset similar to ImageNet in the domain. We first assemble and transform many digital pathology datasets into a pool of 22 classification tasks and almost 900k images. Then, we propose a simple architecture and training scheme for creating a transferable model and a robust evaluation and selection protocol in order to evaluate our method. Depending on the target task, we show that our models used as feature extractors either improve significantly over ImageNet pre-trained models or provide comparable performance. Finetuning improves performance over feature extraction and is able to recover the lack of specificity of ImageNet features, as both pre-training sources yield comparable performance.

show abstract

Section: Discussion and Future Workmentioning

confidence: 99%

Multi-Task Pre-Training of Deep Neural Networks for Digital Pathology

Mormont

Geurts

Marée

2021

IEEE J. Biomed. Health Inform.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Apeer is another service recently available promoted by microscope company Zeiss. NEUBIAS has started to provide a web-service for benchmarking bioimage analysis workflow called BIAFLOW, and this service can also be used as a place to share workflows for others to test (Rubens et al, 2020). Though these services are quite useful as there is no problem to occur in terms of the infrastructure, we still cannot be sure how long these services will be kept in the future, sustainable enough for scientific publications.…”

Section: Box 3 Shared-infrastructures For the Reproducible Analysismentioning

confidence: 99%

Reproducible image handling and analysis

Miura

Nørrelykke

2021

The EMBO Journal

View full text Add to dashboard Cite

Image data are universal in life sciences research. Their proper handling is not. A significant proportion of image data in research papers show signs of mishandling that undermine their interpretation. We propose that a precise description of the image processing and analysis applied is required to address this problem. A new norm for reporting reproducible image analyses will diminish mishandling, as it will alert co-authors, referees, and journals to aberrant image data processing or, if published nonetheless, it will document it to the reader. To promote this norm, we discuss the effectiveness of this approach and give some step-by-step instructions for publishing reproducible image data processing and analysis workflows.

show abstract

“…They can be invoked from other SciJava‐compatible applications including KNIME Analytics Platform, CellProfiler, OMERO, and others, as well as from user scripts, including for executing in headless mode from the command line, without any graphical user interface (Allan et al., 2012; Dietz et al., 2020; Kamentsky et al., 2011; Möller et al., 2016; Ouyang, Mueller, Hjelmare, Lundberg, & Zimmer, 2019). This is useful in several scenarios, including: for batch analysis across large quantities of data; when combining ImageJ functionality with other systems via standard interprocess interoperability approaches (files and/or pipes); for use within modules of container‐based workflow systems; and for distributed execution of analysis pipelines on server clusters (Krumnikl et al., 2018; Rubens et al., 2020; also see Apeer website in Internet Resources). See Support Protocol 2 for an example of invoking ImageJ headless from the command line.…”

Section: Commentarymentioning

confidence: 99%

New Extensibility and Scripting Tools in the ImageJ Ecosystem

et al. 2021

View full text Add to dashboard Cite

ImageJ provides a framework for image processing across scientific domains while being fully open source. Over the years ImageJ has been substantially extended to support novel applications in scientific imaging as they emerge, particularly in the area of biological microscopy, with functionality made more accessible via the Fiji distribution of ImageJ. Within this software ecosystem, work has been done to extend the accessibility of ImageJ to utilize scripting, macros, and plugins in a variety of programming scenarios, e.g., from Groovy and Python and in Jupyter notebooks and cloud computing. We provide five protocols that demonstrate the extensibility of ImageJ for various workflows in image processing. We focus first on Fluorescence Lifetime Imaging Microscopy (FLIM) data, since this requires significant processing to provide quantitative insights into the microenvironments of cells. Second, we show how ImageJ can now be utilized for common image processing techniques, specifically image deconvolution and inversion, while highlighting the new, built‐in features of ImageJ—particularly its capacity to run completely headless and the Ops matching feature that selects the optimal algorithm for a given function and data input, thereby enabling processing speedup. Collectively, these protocols can be used as a basis for automating biological image processing workflows. © 2021 Wiley Periodicals LLC. Basic Protocol 1: Using PyImageJ for FLIM data processing Alternate Protocol: Groovy FLIMJ in Jupyter Notebooks Basic Protocol 2: Using ImageJ Ops for image deconvolution Support Protocol 1: Using ImageJ Ops matching feature for image inversion Support Protocol 2: Headless ImageJ deconvolution

show abstract

BIAFLOWS: A collaborative framework to reproducibly deploy and benchmark bioimage analysis workflows

Cited by 6 publications

References 36 publications

Multi-Task Pre-Training of Deep Neural Networks for Digital Pathology

Multi-Task Pre-Training of Deep Neural Networks for Digital Pathology

Reproducible image handling and analysis

New Extensibility and Scripting Tools in the ImageJ Ecosystem

Contact Info

Product

Resources

About