Researchers from the University of Wisconsin-Madison have developed a new, state-of-the-art artificial intelligence model to automate the detection of developmental abnormalities in zebrafish embryos. This significant advancement was presented at the 28th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2025) in Korea. This monumental effort was produced by Sarath Sivaprasad and his colleagues on Sarath’s CISPA Helmholtz Center for Information Security academic research team. They have prioritized environmental improvements greatly enhancing fertility classification and toxicity assessment utilizing advanced imaging modalities.
The study focuses on two critical experimental tasks: Fertility classification and Toxicity assessment. The Fertility classification task includes a large scale dataset of 130368 images taken in eight hour long sequences. This final task measures the general health and survivability of zebrafish eggs. It provides critical information relevant to reproductive health and developmental biology. The Toxicity assessment consists of a different dataset of 55,296 images collected over a 48-hour period. Its main goal is to recognize the impact of these harmful substances on embryonic development.
Fertility Classification Insights
Fertility classification decides the fate of embryonic development. To better understand the effect of egg viability, researchers created a dataset of more than 130,000 images to determine egg viability accurately. This huge collection provides an unprecedented view of the fertilization process. As such, it provides groundbreaking information that will change the field of reproductive studies in zebrafish.
As vertebrates with transparent embryos and rapid development, zebrafish are well-established as an ideal model organism for biomedical research. Conventional approaches to modeling their growth tend to involve the intensive manual review by experts. This process, which is both time-consuming and subjective, can add a level of variance in results.
Sivaprasad noted the challenges faced by researchers in this field: “Analyzing their development still relies heavily on expert manual inspection—a time-consuming and subjective process.” To address the above challenges head-on, this study created a new automated system to solve these problems at scale. This simplifies the analysis and provides more uniform outcomes.
Toxicity Assessment Advancements
The toxicity assessment component aims to identify the effects of multiple different chemicals on developing zebrafish embryos. This assignment includes a detailed dataset with well over 55,000 images. These photos were taken during a sub-continuous monitoring session that lasted 48 hours. Because this approach uses high resolution imaging and automated analysis, researchers can detect very fine changes during embryo development that indicate a potential toxicity.
Rapid and accurate detection of toxic effects is important across the biomedical research spectrum. This process is particularly important when determining the safety of new pharmaceuticals and environmental chemicals. Sivaprasad emphasized the importance of this aspect: “It’s a valuable resource for the machine learning community to benchmark their methods and for biomedical research to better understand the effects of different drugs.”
By using machine learning methods specifically designed for anomaly detection, researchers can automatically flag where development is occurring at odds with expected patterns. This approach greatly improves the ability to identify the potential toxic effects early in the developmental process.
Machine Learning and Anomaly Detection
The application of machine learning in this research highlights a shift in how scientists approach data analysis in developmental biology. Anomaly detection, as defined by Sivaprasad, is “the process of identifying data points, events, or patterns that deviate significantly from the expected behavior.” Specifically, during training, the AI model determines what abnormal development looks like in relation to normal development. It then scores each sample according to how far it strays from that starting point.
During a training period, the system learns what normal should look like. At inference, it rates each sample according to how far it strays from that ideal of normal, added Sivaprasad. This advanced methodology lets scientists shift their lens from simply classifying images into fixed buckets to spotting anomalies.
Sivaprasad further illustrated the significance of this method: “Unlike traditional classification, which assigns inputs to specific categories (e.g., cat, dog, or car), anomaly detection focuses on distinguishing between ‘A’ and ‘not A’.” In this research framework, it allows a more detailed and nuanced appreciation of the developing zebrafish embryo.
The potential implications of this work go beyond near-term research applications. As Sivaprasad remarked: “Right now, we evaluate only one chemical to understand how anomalies develop. Our goal is to scale this up to an entire library of chemicals.” This ambition represents the promise of a future where high-throughput, detailed chemical assessments could be the standard practice in developmental biology.