A newly published study in SLAS Technology describes a novel artificial intelligence platform, FINS (Fish Imaging and Neural Scoring), that could accelerate zebrafish embryo toxicity screening while reducing the need for labor-intensive manual scoring. Zebrafish embryos are widely used in toxicology and drug development because they develop rapidly, share genetic similarities with humans, and allow researchers to observe organ formation in real time. Traditionally, however, embryo evaluation has relied on manual image review, a process that can be slow, subjective, and difficult to scale.
FINS addresses this challenge by combining automated deep learning image classification with a human review interface. Trained on thousands of annotated embryo images and validated using an independent dataset, the platform analyzes multi-well plate images and classifies embryos as normal, abnormal, or dead across developmental stages. A built-in quality-control module also flags unusable images, such as blurry or obscured samples. The AI model achieved strong performance across classification tasks, including an F1 score of 0.94 for identifying normal embryos and 0.96 for detecting high-quality images suitable for analysis. In practice, researchers report that the web-based system reduced annotation time by more than 70% while improving consistency and scalability in high-throughput screening workflows.
Developed by scientists at Sciome in collaboration with researchers affiliated with Oregon State University’s Sinnhuber Aquatic Research Laboratory, FINS is designed for both research and potential regulatory toxicology applications. Unlike many earlier zebrafish image-analysis tools, the platform operates as an interactive web application that allows scientists to review predictions, correct errors, and retrain the model over time. Future versions will integrate with laboratory information management systems and support automated dose-response and benchmark dose analyses for chemical safety testing.
Read the publication here.
