AI4Life

AI4Life_Logo
Menu

Author: Pasi Kankaanpää

Categories
News

AI4Life Hackathon & Solvathon

AI4Life Hackathon & Solvathon Summary Report

by Beatriz Serrano-Solano

After a highly successful 2nd General Assembly in Heidelberg, AI4Life took advantage of the gathered members and organized a Hackathon & Solvathon. The event was held from October 10th to 13th, 2023, with a total of 25 participants.

Two parallel events ran during this period:

  • Hackathon: Focused on building and deploying AI tools for bioimage analysis, enhancing scalability and FAIRness. Participants engaged in discussions, formed interest groups around projects, and collaborated on various web & cloud bioimage analysis topics.
    The Hackathon members participated in a number of projects:
    • EDAM & bio.tools collaboration with AI4Life
    • Segment Anything in the BioEngine
    • Segment Anything in the BioImage Model Zoo
    • bioimageio Python packages
    • bioimageio uploader
    • ChatBot for the BioImage Model Zoo
    • Documentation improvement
    • Kubernetes BioEngine on Google Cloud
    • Semantic Segmentation of cartilaginous tissue
  • Solvathon: Experts from AI4Life and beyond worked together on the 8 selected open-call projects. Detailed progress updates on these projects will be shared soon.

Both tracks achieved remarkable outcomes. Stay tuned for the next event!

Pictures by Ayoub El Ghadraoui

Categories
News

AI4Life 2nd General Assembly: Summary & Actions

AI4Life 2nd General Assembly: Summary & Actions

by Beatriz Serrano-Solano

The AI4Life project held its 2nd General Assembly on October 9th and 10th, 2023, bringing together computational experts and life scientists from various fields to discuss the project’s progress and future goals. 

The assembly opened with a welcome from the Euro-BioImaging director, John Eriksson and Euro-BioImaging Bio-Hub director, Antje Keppler.

Anna Kreshuk and Florian Jug, the AI4Life Scientific Coordinators, discussed AI4Life’s objectives, including democratizing AI-based methods, establishing standards for submission, storage, and FAIR (Findable, Accessible, Interoperable, and Reusable) data and models. They also highlighted the need for open calls and empowering common platforms with AI integration. 

The keynote speaker was Gergely Sipos, who presented “Computing and AI in EOSC”, discussing opportunities for collaboration. Sipos emphasized the role of EGI, an international e-infrastructure for research and innovation, and its mission to provide computing power, data storage, and training services to the scientific community. He also introduced iMagine, another Horizon Europe-funded project that makes heavy use of EGI resources. Together, AI4Life, iMagine, and EGI decided to explore ways to team up and re-use each others’ technology stacks.

The rest of the meeting was structured from two different angles:

1) Updates from partners: describing the interaction with each other, analysing dependencies within the project, summarising the activities during the first year, and the outlook for the one to come;

2) Updates from Work Packages: describing the goals, deliverables and milestones (submitted and upcoming), and the interactions between work packages. Special focus was given to the question of whether needs have changed since the grant for AI4Life was written. Work Packages also updated the audience with success stories, achievements, goal blockers, pain points, and challenges.

We would like to thank all the participants for their contributions and we are looking forward to an exciting second year of FAIR AI with AI4Life!

Categories
News

The BIA launches a collection of explorable AI-ready image datasets

The BioImage Archive launches a collection of explorable AI-ready image datasets

by Teresa Zulueta-Coarasa

Artificial Intelligence (AI) methods have revolutionised the analysis of biological images, but their performance depends on the data the models are trained with. Therefore, to develop, benchmark, and reproduce the results of AI methods, developers need access to high-quality annotated data.

One of the missions of AI4Life is to democratise the access to well-annotated datasets which are standardised to facilitate their reuse, and presented in a manner that is useful to the community. As part of this effort the BioImage Archive has launched a gallery of datasets that can be explored in-browser without the need to download the images and annotations. Each dataset is presented in a consistent way, following community metadata standards that include information such as the biological application of a dataset, what type of annotations a dataset contains, the licence the data are under or what models have been trained using this dataset. Furthermore, because all images are converted from different formats into the cloud-ready file format OME-Zarr, there is potential for analysing these datasets in the cloud. 

The BioImage Archive team plans to keep enriching this collection with more datasets over time, with the aim of establishing a community resource that can empower the development of new AI methods for biological image analysis.

Categories
News

BioEngine: Unveiling cloud-powered AI for simplified Bioimage Analysis

BioEngine: Unveiling cloud-powered AI for simplified Bioimage Analysis

by Jeremy Metz, Beatriz Serrano-Solano and Wei Ouyang

In a major step toward democratizing AI in life sciences, the AI4Life consortium announces the launch of BioEngine—a scalable, cloud-based infrastructure that powers the BioImage Model Zoo. Designed to be accessible to both experts and novices, BioEngine aims to revolutionize the way bioimage analysis is conducted.

The challenge

The escalating growth of data in life sciences has revealed the limitations of conventional desktop applications used for bioimage analysis. These local solutions are increasingly inadequate to handle high-throughput data and sophisticated applications like AI-driven image analysis. Users often face challenges with large data sets, complex hardware requirements, and intricate software dependencies that make these tools cumbersome to use and difficult to deploy. Additionally, existing machine learning model zoos often necessitate a level of expertise in programming and model selection, making them inaccessible to a wider audience. All these challenges collectively point to the need for a more efficient, scalable, and user-friendly solution for AI model serving.

The solution

Enter BioEngine, a state-of-the-art cloud infrastructure designed to simplify the complex landscape of bioimage analysis. BioEngine powers the BioImage Model Zoo, allowing users to test-run pre-trained AI models on their own images without requiring any local installation. Its cloud-based approach means you can easily connect BioEngine to existing software platforms like Fiji, Icy, and napari, thereby eliminating the need to install multiple dependencies.

For the developers in our community, BioEngine serves as a cloud platform designed to cater to many users while judiciously using limited GPU resources. It features a simple API that can be accessed via HTTP or WebSocket, offering a seamless experience for running models in the cloud. This API can be effortlessly integrated into Python scripts, Jupyter notebooks, or web-based applications.

Key Features
  • User-friendly API for quick integration into existing workflows.
  • Diverse array of supported model formats including TensorFlow, Torch, and ONNX.
  • Cloud-based system for simplifying complex setups, ideal for connecting with existing bioimaging software.
  • Upcoming toolkit for on-premise deployment, providing versatile installation options from Kubernetes clusters to individual workstations or laptops.

By employing BioEngine, both experts and non-experts can overcome the challenges associated with traditional desktop-based solutions and enter an era of streamlined, accessible and scalable bioimage analysis.

Technical challenges

Efficiently serving a variety of models on limited GPU resources required the development of a unique framework that can manage dynamic model execution and scheduling.

 Our approach

BioEngine is the result of a concerted collaboration primarily between KTH Royal Institute of Technology and the BioImage Archive Team at EMBL-EBI, under the auspices of the AI4Life consortium. The platform is built as an extension of the Hypha framework, a robust RPC-based communication hub that orchestrates containerized components for a seamless and efficient operation.

Technical Underpinnings
  • Hypha Framework: The bedrock of our BioEngine, Hypha facilitates the orchestration of containerized modules, enabling a robust system architecture. It supports key functionalities like user authentication and virtual workspaces.
  • Nvidia Triton Inference Server: For handling AI models, BioEngine leverages the Nvidia Triton Inference Server. This allows us to manage and dynamically schedule GPU workloads based on incoming user requests, ensuring optimal resource allocation and efficiency.
  • S3-Based Object Storage: All data is securely and efficiently stored using S3 object storage, ensuring both speed and reliability.

Developer-friendly API

BioEngine provides both HTTP and WebSocket-based RPC APIs. Developers can effortlessly integrate these APIs into web and desktop apps, thereby streamlining the interaction between software and the cloud-based AI models.

Key Features
  • Simple API for model execution
  • Scalability across various AI frameworks like TensorFlow, Torch, and ONNX
  • Support for multiple types of image analysis tasks
  • A variety of data storage and retrieval options
Deployment Options

We are actively working on broadening our deployment offerings:

  • Kubernetes (K8s) Cluster: For large-scale, enterprise-level applications, we are developing support for auto-scaling within a K8s cluster environment.
  • Docker-Compose: For smaller-scale or local deployments, Docker-Compose options are also in the pipeline.
Cost and Accessibility

With the support the EU Horizon research infrastructure grant, we are committed to promoting accessibility and lowering barriers to advanced bioimage analysis:

  • Free and Open-Source: For testing and evaluating, BioEngine will offer a free and open-source option.
  • On-Premise Deployment: For institutions seeking to integrate BioEngine into their existing IT infrastructures, we are developing versatile on-premise deployment solutions.

BioEngine integration examples

Piximi

Source code for integrating BioEngine in web app (Typescript)

KNIME
Icy

Example source code for BioEngine integration in Java

napari

Source code for demo napari plugin for BioEngine integration

Invitation to the community

We encourage everyone to try BioEngine and provide feedback (via image.sc forum, GitHub issues, or our contact form). Your input is crucial for the continual refinement of this revolutionary platform.

For Developers and Bioimage Analysts

BioEngine offers an easy-to-use API for running models, simplifying software design and deployment. 

This API can be integrated into Python scripts, Jupyter notebooks, or web-based applications, giving you the flexibility to adapt BioEngine to your own projects.

Looking ahead

We are actively working on an easily deployable package for institutional use and aim to improve stability and scalability.

For more information, please refer to our API documentation.

With BioEngine, we are democratizing advanced bioimage analysis by offering cloud-based, plug-and-play AI solutions that can be effortlessly integrated into existing software ecosystems, thereby accelerating both research and real-world applications in life sciences.

Acknowledgements

  • EU: Funded by the EU’s Horizon Europe program, grant no. 101057970.
  • Nvidia: Supported our development, notably with Triton Inference Server and computing credit.
  • Bioimaging Community: Thanks to testers and AI4Life Stockholm Hackathon 2023 participants for vital feedback.
  • SciLifeLab & KAW: Supported by SciLifeLab DDLS program, funded by Knut and Alice Wallenberg Foundation.
  • The authors utilized the language model ChatGPT developed by OpenAI for assistance in the structuring and drafting of this news article.
Categories
News

New poster alert: Explore the AI4Life project

New poster alert: Explore the AI4Life project

AI4Life is delighted to introduce a brand-new poster, now available on Zenodo.

Discover the AI4Life journey, its goals, structure and opportunities in this informative poster. This resource is freely accessible and open for anyone to use. Download it, share it, print it, and let the knowledge flow!

Happy exploring!

Categories
News

Introducing the AI4Life Help Desk: Your resource hub for support and information

Introducing the AI4Life Help Desk: Your resource hub for support and information

by Caterina Fuster-Barceló

We’re excited to introduce a brand-new addition to the AI4Life project website that will make your experience even more enriching and informative. We’ve just launched the AI4Life Help Desk, a comprehensive resource hub designed to cater to all your queries, support needs, and information requirements.

What is the AI4Life Help Desk?

The AI4Life Help Desk is your go-to destination for accessing a wealth of resources that will enhance your understanding of the AI4Life project and the BioImage Model Zoo. Whether you’re a seasoned user or just getting started, our Help Desk is here to assist you every step of the way.

What’s inside the Help Desk?

  • Contact Us: Reach out through Image.sc/bioimageio, GitHub, or our user-friendly Typeform for tailored support.
  • Training and Documentation: Access BioImage Model Zoo guides, Galaxy Training Network resources, and AI4Life tutorials.
  • FAQs: Get quick answers to common questions about AI4Life and the BioImage Model Zoo.

Discover the AI4Life Help Desk today at https://ai4life.eurobioimaging.eu/help!

Categories
News

New section on the AI4Life Website: Explore our latest outputs!

New section on the AI4Life Website: Explore Our Latest Outputs!

We are excited to introduce a fresh addition to the AI4Life website dedicated to showcasing our project’s accomplishments in a more accessible manner.

Milestones and Deliverables
AI4Life is a very prolific project, we will publish 25 deliverables and 16 milestones throughout the project’s duration. This newly launched section serves as a hub where you can dig into the specifics of each deliverable and milestone. An interactive timeline provides insight into the projected completion dates of our deliverables, offering a clear roadmap for our ongoing endeavours.

Highlighting Our Scientific Contributions
In this same section, we’ve also gathered our scientific publications. By centralizing these contributions, our aim is to foster knowledge-sharing and encourage collaboration within the wider scientific community.

Your Window into AI4Life’s Advancements
We invite you to explore this new section of our website to gain insight into the work that has defined AI4Life’s journey thus far. By making our outputs easily accessible, we hope to engage you with our work. And, for those that want to stay closely connected to AI4Life, we offer the option to subscribe to our newsletter.

Categories
News

Models developed in AI4Life and Bioimage.io now made available in AIVIA

Models developed in AI4Life and Bioimage.io now made available in AIVIA

AI4Life and Leica are announcing their collaboration to make deep-learning models developed by the bioimage community available to a wider user community through integration into Leica’s AIVIA software.

In a commitment to the scientific community, AI4Life and Leica Microsystems join forces to help researchers to leverage AI in complex experiments. AI4Life, coordinated by Euro-BioImaging, is a Horizon Europe-funded project that brings together the computational and life science communities. Its goal is to empower life science researchers to harness the full potential of Artificial Intelligence (AI) methods for bioimage analysis – and in particular microscopy image analysis, by providing services, and developing standards aimed at both developers and users.

One of the consortium’s objectives is to build an open, accessible, community-driven repository (the BioImage Model Zoo) of FAIR pre-trained AI models and develop services to deliver these models to life scientists. Together with their community partners, the consortium ensures that models and tools are interoperable with Fiji, ImageJ, Ilastik and other open-source software tools.

In the meantime, Leica Microsystems has cultivated a valuable connection with the AI4Life project and the BioImage Model Zoo. Most recently, the Leica team had the opportunity to meet the people behind the AI4Life project at a workshop organized by the Euro-BioImaging Industry Board (EBIB). It was there, through their participation, that they realized the power of their combined resources.

Widely recognized for optical precision and innovative technology, Leica Microsystems supports the imaging needs of the scientific community with AIVIA, their advanced Al-powered image analysis software. AIVIA is a complete 2-to-5D image visualization and analysis platform designed to allow researchers to unlock insights previously out of reach. Through the joined effort, BioImage Model Zoo models can now also be easily integrated into Leica’s commercial software AIVIA.

“We see this initiative as an opportunity to tie in AIVIA with the scientific community and to make work done by the community accessible in a convenient way to AIVIA users.” said Constantin Kappel, Manager AI Microscopy and Insights at Leica Microsystems.

To achieve this, Leica is taking the models that are published in BioImage.io and converting them to their own AIVIA Model repository format for interoperability with AIVIA. From a small number of models, the offer will gradually be increased.

This further supports the aim of AI4Life to lower the barriers for using pre-trained models in image data analysis for users without substantial computational expertise or those with existing licences, who might rely on commercial solutions.

This is a great example of how resources curated by academia and available in open access can be of interest to the imaging industry. The source of the models are recognized directly in AIVIA, a nice testimony to industry-academia collaboration and a nice endorsement of the models curated by the BioImage Model Zoo.

“We much appreciate the interest of Leica. Models relevant to AIVIA will be directly available to users, which is a testimony to fruitful industry-academia collaboration and a great endorsement of the utility of the BioImage Model Zoo. We hope many other software tools will follow Leica’s lead and also start benefitting from this community resource we are currently building” said Florian Jug, one of the scientific coordinators of AI4Life.

“Bioimage.io is supported by AI4Life. AI4Life has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement number 101057970.”

 

More information:

https://www.aivia-software.com/post/pancreatic-phase-contrast-cell-segmentation-bioimage-io

https://www.aivia-software.com/post/hpa-cell-segmentation-bioimage-io

https://www.aivia-software.com/post/hpa-nucleus-segmentation-bioimage-io

https://www.aivia-software.com/post/b-subtilist-bacteria-segmentation-bioimage-io

Categories
News

First AI4Life Open Call: Announcement of selected projects

First AI4Life Open Call: 
Announcement of selected projects

by Florian Jug & Beatriz Serrano-Solano

The first AI4Life Open Call received an impressive response, with a total of seventy-two applications. It proved to be an incredible opportunity for both life scientists seeking image analysis support and computational scientists eager to explore the evolving landscape of AI methodologies. In this blog post, we announce the awarded projects and invite you to join us behind the scenes as we explore the selection process that determined which projects have been selected.

Awarded projects

First things first, here is the list of titles of the selected projects (in alphabetical order):

  • Analysis of the fiber profile of skeletal muscle.
  • Atlas of Symbiotic partnerships in plankton revealed by 3D electron microscopy.
  • Automated and integrated cilia profiling.
  • Identifying senescent cells through fluorescent microscopy.
  • Image-guided gating strategy for image-enabled cell sorting of phytoplankton.
  • Leaf tracker plant species poof.
  • SGEF, a RhoG-specific GEF, regulates lumen formation and collective cell migration in 3D epithelial cysts.
  • Treat CKD.

The projects are diverse, covering scientific topics ranging from Plant Biology, Physiology, Metabolism, Cell Biology, Molecular Biology, Marine Biology, Flow Cytometry, Medical Biology, Regenerative Biology, Neuroscience, etc. The researchers who have proposed the projects come from the following countries: France (2x), Germany, Italy, Netherlands, Portugal, and the USA (2x).

How did the review procedure work?

1. Eligibility checks

The selection procedure started with internal eligibility checks. Is the project submitted completely? Is the information complete and telling a complete story that is fit for external reviews? At this stage, we only had to drop 10 projects of a grant total of 72 submitted projects. Our intention was to only filter projects that drew an incomplete picture and leave the judgement of the scientific aspects to our reviewers.

2. Reviewing procedure

After assembling a panel of 16 international reviewers (see list below), we distributed anonymized projects among them. All personal and institutional information was removed, only leaving project-relevant data to be reviewed. We aimed at receiving 3 independent reviews per project, requiring each review to review about 11 projects total.

Here is the list of questions we asked our reviewers via an electronic form:

  1. Please rank the following statements from 1 (Likely not) to 5 (Likely):
    1. The proposed project is amenable to Deep Learning methods/approaches/tools.
    2. Does the project have well-defined goals (and are those goals the correct ones)?
    3. A complete solution to the proposed project will require additional classical routines to be developed.
    4. The project, once completed, will be useful for a broader scientific user base.
    5. The project will likely require the generation of significant amounts of training data.
    6. This project likely boils down to finding and using the right (existing) tool.
    7. Approaches/scripts/models developed to solve this project will likely be reusable for other, similar projects.
    8. The project, once completed, will be interesting to computational researchers (e.g. within a public challenge).
    9. The applicant(s) might have a problematic attitude about sharing their data.
    10. Data looks as if the proposed project might be feasible (results good enough to make users happy).
    11. Do you expect that we can (within reasonable effort) improve on the existing analysis pipeline?
  2. What are the key sub-tasks the project needs us to improve?
  3. What would you expect will it take (in person-days again) to generate sufficient training data?
  4. Do suitable tools for this exist? What would you use?
  5. Once sufficient training data exists, what would you expect is the workload for AI4Life to come up with a reasonable solution for the proposed project? Please answer first in words and then (further below) with the minimum and maximum number of days you expect this project to take.
  6. What is your estimated minimum number of days for successfully working on this project?
  7. What is your estimated maximum number of days for successfully working on this project?
  8. On a scale from 1 to 10, how enthusiastic are you about this project?

Due to the unforeseen unavailability of some reviewers, we ended up with about 2.7 reviews per project, with some projects receiving 2 but most projects receiving all 3 desired reviews.

 
3. Scoring projects according to reviewer verdicts

We first aggregated all reviews per project by averaging numerical values and concatenating textual evaluations. We then developed three project scores: a quality score (main metric), a total effort score, and a slightly more subjective excitingness score.

  1. The quality score was computed by taking a weighted average of the evaluations we received. I.e., questions (1-a) to (1-k) from above.  (Note: not for all questions higher values are better. We have of course first inverted the “low-is-better” ones to make all values compatible.)
  2. The effort score was taking the (minimum) time estimates for label data generation and successfully completing the project, and computing a value corresponding to the estimated total person-month to completion.
  3. The excitingness score simply is the average of the values received as answers to question 8.

The final score was computed by: 0.75*(quality/effort) + 0.25*excitingness

This formula favors projects that are estimated to be conducted in less time, which is in line with our aim to help more individuals through the AI4Life Open Calls. 

 4. Final decisions by the Open Call Selection Committee 
  1. After anonymized scoring of all projects, we have added the applicants’ identities and institutions back into the final decision matrix.
  2. We have prepared ourselves to break ties and potentially remove better-ranked projects for the sake of having a higher diversity. To our surprise, the top-ranked projects showed a wonderful diversity, making this step unnecessary.
  3. The final decision was taken by the Open Call Selection Committee. With the members of the committee (see below), we have re-lived all steps of the Open Call process, from the application, and reviewing, to the final grading stage. After some stability analysis (i.e., after changing the weights for the weighted sums in the procedure outlines above and noticing that the best projects remained rather stably top-ranked), the Committee decided to simply select as many of the best-evaluated projects as we could fit into the AI4Life time budget for this round of Open Calls. This led to a total number of 8 selected projects.
  4. Seeing the extraordinary quality of many of the submitted projects, it was clear to us that many more than 8 projects would deserve to receive support. We have therefore decided to put a sizeable number of additional projects on a waiting list, hoping that we can engage more helping hands. 

Who was involved in the review process?

And now? What’s next?

The selected projects will be assigned to our AI4Life experts waiting to support them. All other projects are offered a space in the AI4Life Bartering Corner, a new section soon to appear on our website, where projects will be showcased to computational experts who can reach out to the proposing parties and engage in a fruitful collaboration. 

If you did not apply to the first Open Call, we invite you to do so at the beginning of 2024. Subscribe to our newsletter, we will inform you when the next call opens.

Additionally, if you are interested to put any open analysis problem you have on our Bartering Corner, please fill out this form.

If you need help quicker, we recommend Euro-BioImaging’s Web Portal, where you can access a network of experts in the field of image analysis. Please note that this service may involve associated costs, but access funds for certain research topics are available through initiatives such as ISIDORe.