An international team of scientists, including Vladimír Ulman from IT4Innovations, has presented the MIFA recommendations (Metadata, Incentives, Formats, Accessibility). These aim to accelerate the development of artificial intelligence (AI) in bioimage analysis by making large, well-annotated datasets more accessible, allowing AI to learn more effectively and analyse microscopic images of cells, molecules, and tissues faster and more accurately.

AI-based analysis of biological images has become essential — for example, it enables the segmentation of cells and molecular structures and the processing of the enormous volumes of data generated by modern microscopes. However, the development of such methods requires extensive, diverse, and well-annotated datasets, which are often fragmented, inconsistently formatted, and difficult to access.

The MIFA recommendations address this challenge through four key principles:

  • Reducing format diversity – promoting the use of open and easily accessible formats such as OME-Zarr, GeoJSON, EMDB-SFF, or COCO;
  • Standardising metadata – consistent descriptions of data and annotations improve understanding and enable more efficient use of datasets;
  • Increasing data accessibility – supporting open repositories, such as the BioImage Archive or the newly established Czech Bioimaging repository within EOSC, to facilitate data discovery and sharing;
  • Incentivising dataset creation – promoting systematic sharing and annotation of data.

Vladimír Ulman from IT4Innovations, one of the MIFA co-authors, also presented these recommendations at the EOSC National Tripartite Event in Brno, Czechia. He emphasised the importance of computational infrastructure: "To develop AI for bioimage analysis, we need large, high-quality annotated datasets. This requires time, effort, and investment. Without FAIR repositories and high-performance computing resources, available for example through EOSC, it is difficult to develop these methods."

Researchers from the Czech Republic, the UK, Italy, Germany, Spain, France, Sweden, and the USA contributed to the MIFA recommendations, published in the prestigious journal Nature Methods. This collaborative outcome represents an important step towards more open, accessible, and effective AI tools for biomedical research.

 

Research article:
MIFA: Metadata, Incentives, Formats, and Accessibility guidelines to improve the reuse of AI datasets for bioimage analysis
https://www.nature.com/articles/s41592-025-02835-8


* Illustrative image adapted from: Zulueta-Coarasa, T., Jug, F., Mathur, A. et al. MIFA: Metadata, Incentives, Formats and Accessibility guidelines to improve the reuse of AI datasets for bioimage analysis. Nat Methods (2025). https://doi.org/10.1038/s41592-025-02835-8, Fig. 1.