Artificial intelligence (AI) is changing the landscape of bioimage analysis. Today, these same researchers can save many hours and instantly spot subtle patterns in millions of microscopy images in seconds instead. This technological advancement has tremendous potential to improve the speed and accuracy of diagnostics and research in life sciences. Joshua Talks, a Ph.D. student at the European Molecular Biology Laboratory (EMBL), recently shared insights on AI analysis for bioimages, emphasizing the significance of guidelines aimed at improving dataset reuse.
The new MIFA guidelines – an acronym for Metadata, Incentives, Formats and Accessibility – were released today. Specifically, they were designed to enhance the interoperability of AI datasets across modalities in bioimage analysis. In 2023, our workshop of 45 participants, from different sectors, went even further. Our participants were data producers, AI scientists, and bioimage analysts, all of whom are involved in the AI4Life project and primarily responsible for shaping these policy recommendations. These MIFA guidelines are an attempt to bridge that gap between life scientists and AI developers. Life scientists work for years producing deeply-annotated datasets AI developers can’t interpret or worse, mis-interpret.
The Role of AI in Bioimage Analysis
AI excels in analyzing complex biological data. Within a few seconds, it runs an individualized patient scan against thousands of others to find similarity matches. This process discovers trends that would be missed by a human analyst. This level of speed and efficiency can have tremendous impact in accelerating research results. The outcomes of AI interpretation for bioimages are detailed in the publication available here publication.
ReSCU-Nets is a prime example of the innovative work happening in this field. This groundbreaking technology was created by University of Toronto Professor Rodrigo Fernandez-Gonzalez. These deep learning recurrent neural networks combine segmentation and tracking of complex cellular, subcellular and supracellular structures derived from multifactorial multidimensional confocal microscopy time series.
“We hope that, by sharing our images and annotations according to the MIFA guidelines, we will maximize the reusability of our datasets for training novel AI tools, and increase the visibility of the AI tool that we trained using those datasets. Our ReSCU-Nets are recurrent neural networks that integrate segmentation and tracking of cellular, subcellular, and supracellular structures from multidimensional confocal microscopy sequences.” – Rodrigo Fernandez-Gonzalez
The promise of fast, robust, accurate large-scale data analysis to accelerate the pace of scientific discovery and medical diagnosis remains largely unfulfilled. Without widely accepted best practices for sharing and interpreting these data, the promise of these AI tools goes unrealized.
The Importance of MIFA Guidelines
The MIFA guidelines aim to maximize the reusability of AI datasets to facilitate their use in bioimage analysis. They outline four key components: Metadata, Incentives, Formats, and Accessibility. Through following these principles, researchers can make the datasets we develop more easily discoverable and usable.
The first one, Metadata, to focus on the need for rich descriptive information to go with datasets. This is how we can get AI developers to appreciate the context, use cases, and limitations of data they are using. The Incentives section promotes data-sharing to researchers by outlining the rewards that they will receive for working together.
Datasets are rendered highly usable by advanced algorithms through standardization of formats. Equity Accessibility is primarily concerned with removing barriers and ensuring quality data is accessible to a broader audience. Fostering collaboration across disciplines and stakeholders is essential to ensuring we realize AI’s transformative power across bioimage analysis.
These recommendations were published in Nature Methods and can be found via DOI: 10.1038/s41592-025-02835-8. Lead authors are Teresa Zulueta-Coarasa et al. They provided a 3D render built with 3D Slicer software as a result of their ongoing research practices.
Real-World Implications
Even in the short time between MIFA release and today, we’ve seen the real-world impact of implementing MIFA guidelines in active research projects. Joshua Talks highlighted how these guidelines have facilitated his work in accessing appropriate datasets for investigating pre-trained image segmentation models.
“Thanks to the MIFA guidelines and the BioImage Archive, I could easily locate appropriate new datasets for a project that investigated the transferability and suitable selection of pre-trained image segmentation models. Access to well-structured metadata made working with multiple datasets for training and evaluating neural networks much simpler and time-efficient. Our results are available here.” – Joshua Talks
Aligning with MIFA guidelines is an important step towards creating deeper, more sustained collaborations between life scientists and AI developers. As researchers continue to generate vast amounts of data, adhering to standardized practices will be vital for ensuring that these resources can be effectively utilized.