As part of the ESTIV meeting, Dr. Ruchir Shah will present recent work using artificial intelligence (AI) to evaluate developmental toxicity in zebrafish embryos. Evaluating toxicity using the zebrafish embryo developmental toxicity bioassay requires assessing larval morphological changes by experienced screeners using imaging data. The process is time-consuming and prone to subjectivity. However, deep learning models trained using labeled image data can help classify subtle changes in images more efficiently and objectively. Sciome and NIEHS collaborators are developing a deep-learning model using labeled image data from Systematic Evaluation of the Application of Zebrafish in Toxicology (SEAZIT) to classify over 20 different types of larval morphological changes collected from embryos exposed to various test substances for 5 days. We developed a customized multi-view convolutional neural network (MVCNN) using EfficientNet architecture which uses the distinct dorsal and lateral views of each embryo. Additionally, we are developing a deep-learning segmentation model to identify and quantify larval organ structures: eye, bladder, notochord, yolk, otolith, otic vesicle, brain, pigment, and fish body. We are building segmentation models using UNet++ architecture with encoders from the EfficientNet family. Together, these models facilitate rapid review of larval morphological changes in images obtained from zebrafish. Such models can provide more efficient and standardized output than manual visual assessment of zebrafish images alone, thereby enhancing data reliability.
The 22nd International Congress of the European Society of Toxicology In Vitro (ESTIV2024) is being held in Prague, Czech Republic June 3-6. ESTIV brings together researchers, toxicologists, and students engaged in the promotion of development of in vitro and in silico approaches from around the globe.
