MOTIVATION: Zebrafish embryo assays are increasingly recognized as a robust and scalable model for developmental toxicity screening due to embryos' optical transparency, rapid organogenesis, and high genetic and physiological conservation with humans. Despite these advantages, conventional phenotypic assessment relies heavily on manual scoring of morphological abnormalities, which is time-consuming, prone to inter-observer variability, and challenging to scale for high-throughput studies. The lack of reproducible and automated scoring tools hampers the adoption of zebrafish assays in large-scale toxicology pipelines and regulatory workflows. METHOD: We present FINS (Fish Imaging and Neural Scoring), an interactive, web-based platform that integrates deep learning-based classification with intuitive visualization and quality control tools for zebrafish morphological screening. FINS leverages convolutional neural network (CNN) models to automatically identify normal and abnormal phenotypes across multi-well plate images. It allows users to visualize the model's confidence scores, manually correct annotations, and track quality control metrics, providing a seamless workflow for both automated and expert-reviewed scoring. RESULTS: In benchmark experiments, FINS showed high classification accuracy across classes, with high F1 for both normal and abnormal groups. In addition to phenotype prediction, FINS incorporates a quality control (QC) module that automatically flags poor-quality or unusable embryo images (e.g., out-of-focus, occluded, or containing water droplets). This ensures that only valid images are used for downstream morphological assessment and reduces annotation time by over 70% compared to conventional scoring. By combining speed, accuracy, and interpretability, FINS enables reproducible, scalable, and standardized zebrafish image analysis, facilitating adoption into both research and regulatory applications. AVAILABILITY: FINS is accessible at https://apps.sciome.com/fins/login/. For access, please contact us at software.support@sciome.com, Contact: software.support@sciome.com.
Oktay et al. (Wed,) studied this question.