Monitoring Chicken Embryological Development with Machine Learning
Overview
Teratogens are agents that cause malformations in embryos during development. With 3% of live births having a malformation and the cause of 65-75% of malformations in human newborns being multifactorial or unknown, it is a common yet complex problem. There is currently no teratogenic screening test, so determining what drugs and other products are safe for a developing fetus can be difficult. Our goal is to create a teratogenic assessment method for clinical and research use. In this study, chicken embryos were used since they are an ideal model for embryonic and developmental studies. They can be observed during their development outside of the mother and, with modern practices, be removed from the eggs to better visualize development over time. In this study, we extracted about 15 embryos from their eggs at approximately 50 hours of development and maintained their development in vitro on individual nutrient agar plates. All embryos were photographed twice a day for the subsequent five days using a Nikon SMZ-U at a magnification of 75x. Using the machine learning program MakeML, boxes were drawn on the images by hand around five prominent anatomical parts of embryos: the spine, the end of spine, somite 16, the telencephalon, and the eye cup. The data annotated by hand was transferred to the CreateML program to train the computer to identify those objects. After 25 pictures were annotated and the first training session was completed, the program had a loss of 1.4. Loss is a measure of how often the computer is incorrect when it checks itself against the answer key of the human-annotated boxes, and a value of 2 or below is considered acceptable. The algorithm also had 100% confidence in all areas except somite 16. This was understandable since all of the somites appear identical. As we build this database of healthy embryo images, the computer will learn what is normal at various stages of development. In the future, we will train the computer to recognize developmental defects caused by known teratogens. This will lay the foundation for a teratogen screening system that may one day be used in clinical and research applications.
Competition history
- AJAS 2020
Related projects
ISEF · 2017
A Morphokinetic and Machine Learning Model for Aneuploidy Screening of Human Preimplantation Embryos
AJAS · 2020
Using Object Detection Machine Learning to Count and Categorize C. Elegans
JSHS · 2024
Minimizing In-Vitro Fertilization Failures by Utilizing Artificial Intelligence to Evaluate the Health of Human Embryos
AJAS · 2022
Using Machine Learning to Count Microscopic Worms
CWSF · 2026
A Multi-Omic “Digital Embryo” Framework to Model Early Human Preimplantation Development In Silico
CSEF · 2003
Glucose as a Teratogen Affecting Cranial Neural Crest Migration in the Developing Chick Embryo
ISEF · 2023
Novel Approach to Creating Hyper-Customizable Egg Models for Embryonic Analysis
ISEF · 2026
Novel Machine Learning Model Developed to Optimize Embryo Selection for Frozen Embryo Transfer (FET) and to Calculate Appropriate Number of Embryos Recommended for Embryo Banking
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: AAAS Annual Meeting (Confex) / American Junior Academy of Science