Development of an AI-Assisted Embryo Selection System Using Iberian Ribbed Newts for Embryo–Fetal Development Toxicity Testing

Abstract

Background The 3Rs (Reduct ion, Ref i nement, Replacement) principle is driving the need for alternative methods in animal testing. Despite advancements in in vitro testing, complex systemic toxicity tests still ne-cessitate in vivo approaches. The aim of this study was to develop a developmental toxicity test protocol using the Iberian ribbed newt (Pleurodeles waltl) as a model organism, integrating AI image analysis for embryo selection to improve test accuracy and reproducibility. Methods We established a developmental toxicity test protocol based on the zebrafish test. Gonadotropin was administered to induce ovulation, and in vitro fertilization was performed. Embryos were imaged at 5–6 and 6–7 h post-fertilization. AI image analysis was utilized to assess embryo viability. The test chemi-cal was administered 24–48 h post-fertilization, and morphological changes were observed daily until day 8. Additionally, a time-lapse photography system was constructed to monitor embryonic development. Results Out of 24 cultured embryos, 75% developed normally to the late tail bud stage or initial hatching stage, whereas 25% experienced developmental arrest or death. AI image analysis achieved high accuracy in classifying embryos, with overall accuracies of 92.0% and 92.9% for two learning models. The AI system demonstrated higher precision in the selection of viable embryos compared to visual inspection. Conclusion The Iberian ribbed newt presents a viable alternative model for developmental toxicity testing, adhering to the 3Rs principles. The integration of AI image analysis substantially enhances the accuracy and reproducibility of embryo selection, providing a reliable method for evaluating developmental toxicity in pharmaceuticals.