AI4Life at the 5th NEUBIAS Conference

AI4Life recently participated in the 5th NEUBIAS Conference in Porto, Portugal, a two-part event comprising the Defragmentation Training School and the Open Symposium. This conference brought together experts and enthusiasts in the field of BioImage Analysis (BIA) to explore cutting-edge techniques and advancements in the integration of cloud computing. Under the umbrella of AI4Life, there were significant contributions in various sessions and presentations during the conference showcasing the project’s expertise and involvement in the BIA community. 

Defragmentation Training School 2: Bringing BioImage Analysts to the Cloud!

Under the guidance of Daniel Sage and Estibaliz Gómez de Mariscal, one of the sessions of the Defragmentation Training School delivered insights into the application of Zero Code Deep Learning tools for Bioimage Analysis. Attendees were introduced to the Bioimage Model Zoo, a valuable resource for accessing trained models, and its community partners ZeroCostDL4Mic and deepImageJ. Additionally, Félix Mercier and Ignacio Arganda-Carreras provided comprehensive training on machine learning techniques for classification and segmentation. This training school proved to be a pivotal opportunity for participants to enhance their AI skills for BioImage Analysis.

AI4Life session at the Open Symposium

AI4Life had a dedicated session during the Open Symposium, where Ignacio Arganda-Carreras presented BiaPy, a new library designed to train different networks for BioImage Analysis, which is on the roadmap to be included in the BioImage Model Zoo. Constantin Pape showcased the integration of the Segment Anything Model (SAM) from Meta AI in Napari to enable the interactive segmentation of microscopy images. Estibaliz Gómez de Mariscal discussed the BioImage Model Zoo and its potential to standardise the use of deep learning-based image analysis and promote user-friendly tools. Overall, the session highlighted AI4Life’s dedication to training and facilitating the transfer of skills to life scientists.

During the second half of the session, Ko Sugawara discussed deep learning-based cell segmentation with sparse annotations, while Cristina Lozano Izquierdo focused on data analysis for super-resolution microscopy in nanomedicine. Johannes Roos introduced Arkitekt, a platform for streaming analysis and real-time workflows in microscopy. 

Open Source Software Lounge

In the Open Source Lounge, attendees had the opportunity to explore tools such as Ilastik, deepImageJ and the Bioimage Model Zoo, which received considerable attention, both from the most naive users and bioimage analysis developers, and positive feedback.

Community Concerns and Future Prospects

The conference provided a platform for addressing community concerns and discussing the future of BioImage Analysis. A general topic was the need for collaboration and multidisciplinarity in the field involving all the STEM areas of expertise. As Martin Maska indicated, “the choice of the best approach to implement is application dependent”. This highlights the value of the synergies between life sciences and engineering that can push the optimisation of such approaches. He also emphasized that methods, benchmarks and challenges are of high interest to the BIA community as they enable the dissemination of robustly evaluated tools. Volker Baecker, representing BIAFLOWS, raised concerns about reproducibility, reusability, replicability, and reliability, underscoring the importance of maintaining functional and open software environments. Participants engaged in fruitful discussions on documenting manual pipelines and workflows to ensure transparency and reproducibility.

Yet an important point for discussion was the environmental price of AI technologies, raised by Daniel Sage. While AI is bringing uncountable benefits to society, it is also a significant source of carbon dioxide (CO2) emissions due to the high-performance computing behind it. As expressed by Daniel, “regarding climate change, we are in an extreme urgency, but this matter is still actionable. We have to act in our personal life, in the school, in our working place, …”. A good start to becoming more conscious and careful in these matters, can be the estimation of CO2 emissions of our models, which in Python is now possible thanks to the recommended CodeCarbon package.

Regarding the usage of deep learning methods for image analysis, the conference showed a shift in mindset within the imaging community. Unlike the discussions within the community in previous editions of NEUBIAS, there were no concerns this time about the risks or trustworthiness of these methods. Instead, the focus shifted to implementing and exploiting these powerful research tools effectively. Attendees showed great curiosity and interest in novel models, such as SAM, the scope of fine-tuning models and the integration of interpretable machine learning in user-friendly ecosystems. Reusability and accessibility throughout the entire research process, from sample preparation to data analysis, were also prominent themes.

All in all, during this beautiful week of May 2023, the 5th NEUBIAS Conference provided a platform for advancing BioImage Analysis. AI4Life’s participation brought the expertise present in the project closer to the vibrant BioImage Analysis community and opened the door for collaborations between its players. We are already looking forward to the next one!

Contributors to the post

This post was written by Beatriz Serrano-Solano and Estibaliz Gómez-de-Mariscal.

(1 votes, average: 1.00 out of 1)