A recent study led by the University of Plymouth has introduced a new deep learning AI model called Dev-ResNet that can accurately identify key developmental events during embryonic development from video footage. The research, published in the Journal of Experimental Biology, focuses on pond snails and shows how the model can detect crucial events such as heart function, crawling, hatching, and even death. One of the key innovations of this study is the use of a 3D model that analyzes changes between frames of the video, allowing the AI to learn from these features rather than still images.
The use of video enables Dev-ResNet to reliably detect various features such as the first heartbeat, crawling behavior, shell formation, and hatching. The model has also revealed sensitivities of different features to temperature that were previously unknown. While the study was conducted on pond snail embryos, the researchers believe that Dev-ResNet has the potential to be applied across all species. They have provided comprehensive scripts and documentation for using the model in different biological systems. The ultimate goal of this research is to accelerate the understanding of how climate change and other external factors impact both humans and animals.
The project was spearheaded by PhD candidate Ziad Ibbini, who initially studied BSc Conservation Biology at the University of Plymouth before pursuing further education in software development and embarking on his PhD journey. Ibbini designed, trained, and tested Dev-ResNet independently, emphasizing the importance of delineating developmental events in early animal development to better comprehend changes in event timing between species and environments. He highlights that while the model is small and efficient, the key challenge lies in creating the necessary data to train the deep learning model effectively.
Dr. Oli Tills, the senior author of the paper and a UKRI Future Leaders Research Fellow, underscores the significance of this research for advancing our understanding of organismal development. He notes that the University of Plymouth’s Ecophysiology and Development research group has a rich history of studying this field. Tills expresses excitement about the potential of deep learning technology and the impact it could have on studying animals during their most dynamic developmental period. The research team believes that Dev-ResNet represents a significant step forward in equipping the scientific community with tools to understand how various factors influence a species’ development and, consequently, how they can be protected.
In conclusion, the successful development of Dev-ResNet has opened up new possibilities for studying embryonic development in different species using video analysis. The model’s ability to detect key functional events with high accuracy and efficiency showcases its potential impact on various fields, including conservation biology and climate change research. By providing accessible scripts and documentation, the researchers aim to empower the scientific community to apply Dev-ResNet in diverse biological systems, furthering our understanding of how developmental processes are influenced by environmental factors and external stressors. This groundbreaking research not only highlights the technological advancements in deep learning but also underscores the importance of studying organismal development for broader scientific and conservation efforts.
