An Overview of Artificial Intelligence Methods in Zebrafish-based Tests

Background and aims: Zebrafish (Danio rerio) is an efficient animal model for conducting wholeorganism drug testing and toxicological evaluation of chemicals. They are frequently used forhigh-throughput screening owing to their high fecundity. Peripheral experimental equipment andanalytical software are required for zebrafish screening, which need to be further developed. Machinelearning has emerged as a powerful tool for large-scale image analysis and has been appliedin zebrafish research as well. This study aims to attempt to present various helpful and efficienttools that use artificial intelligence and machine learning in zebrafish-based tests and examine theadvantages and disadvantages of each method used previously.Method: This study is a review of published articles since ۲۰۱۰ in the field of artificial intelligenceapplication in the zebrafish-based tests. In order to collect the articles, the keywords of“Zebrafish”, “Artificial intelligence”, “Imaging” and “Machine learning” were used in databasessuch as Google Scholar, Science Direct, PubMed and etc. Also the way of using Pixel Classifier,Deep Fish, ML Classifiers, Athena Zebrafish software that uses artificial intelligence or machinelearning has been explored in various articles.Results: The results are classified in the fields of automated sample handling, imaging, and dataanalysis with zebrafish during early developmental stages. Furthermore, advances in orientingthe embryos, including the use of robots, microfluidics, and creative multi-well plate solutionshave been highlighted. Analyzing the images in a fast, reliable fashion that maintains the detailsis a crucial step; which is a main feature of AI. another benefit of machine learning approachesis ease of use once trained. Users feed input images into the algorithm and the software returnsa result output without further interaction. This flexibility of input image type (grayscale, RGB,etc.) allows for a variety of specific staining methods to visualize zebrafish. After examining softwarethat uses artificial intelligence and machine learning in the zebrafish field, a comparison ofthis software was made in terms of functionality, availability, quality of results and user interface.Conclusion: The main problem of the zebrafish-based test is the manual inspection of thousandsof embryo’s images in different phases and this is not workable enough for the analysis, also it isslow and may be an inaccurate process. In summary, there are many options in terms of artificialintelligence tools available to zebrafish researchers to screen zebrafish embryos and larvae. Thebiggest challenge is choosing the most efficient tools available or finding a bridge to use multipleefficient tools at the same time.