A team of researchers from CISPA Helmholtz Centre for Information Security and the Helmholtz Institute for Pharmaceutical Research Saarland (HIPS) has presented an AI model that automatically detects developmental disorders in embryonic development in zebrafish. The work will be presented at the 28th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in South Korea.

The study is based on a comprehensive image dataset of more than 185,000 microscopic images, which was created specifically for the analysis of zebrafish development. It includes both sequences for assessing fertility and series of images for investigating toxicity effects. According to the researchers, this has created a basis that overcomes the previous lack of high-resolution, time-detailed data.

Based on this dataset, the scientists developed a transformer-based neural network that can take into account both individual images and temporal sequences. The system achieved 98 percent accuracy in detecting fertilized embryos and 92 percent in identifying developmental abnormalities resulting from toxic exposure. This brings the process closer to human expertise, but enables faster and scalable evaluation.

Zebrafish are considered an indispensable model in biomedical research due to their transparency, rapid development and genetic proximity to humans. Until now, however, the analysis of their development has been heavily dependent on time-consuming manual processes. The new method is intended to overcome this bottleneck and serve as a benchmark for future models.

Both the image dataset and the trained model are made publicly available. The researchers hope that this will accelerate the development of new methods for toxicity screening and advance research into reproducible, efficient and ethically responsible AI-based analyses.

Original Paper:

Sarath Sivaprasad, Hui-Po Wang, Anna-Lisa Jäckel, Jonas Baumann, Carole Baumann, Jennifer Herrmann, and Mario Fritz: Automated Detection of Abnormalities in Zebrafish Development, In: 28th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2025.

Editor: X-Press Journalistenbüro GbR

Gender Notice. The personal designations used in this text always refer equally to female, male and diverse persons. Double/triple naming and gendered designations are used for better readability. ected.