Itamar Yahav
Research Mentor(s): Anne Kort
Mentor Department: Earth and Environmental Sciences
Authors: Itamar Yahav, Anne Kort
Session: Session 1 (9:00am – 9:50am)
Presentation Type: Poster 66
Abstract
Advancing technology in 3D imaging has revolutionized the study of morphological variation in biology and generated detailed 3D data of numerous organisms. However, due to the complexity of this data, current methods for comparing morphology between organisms and classifying into ecological categories have failed to fully capture the subtle links between morphology and ecology. With the rapid advancement of Machine Learning (ML) technologies, new classification and clustering algorithms can be applied to this complex morphological data. Our research seeks to apply clustering algorithms to 3D mesh models using unsupervised ML algorithms, such as K-Means, DBSCAN, and Hierarchical Clustering, to identify potential ecological groupings that may otherwise be missed. We are building off of an existing supervised ML model for 3D data called Meshnet. We are applying this method to a dataset of mammalian lumbar vertebrae. Lumbar vertebrae, located between the thoracic vertebrae (chest) and sacrum (pelvis) in the lower back of mammals, represent a region that has undergone diverse morphological evolution. These changes have been documented by previous research, but the functional and ecological implications of these changes remain unclear. Researchers interested in the evolutionary development of the lumbar vertebrae and in gaining a deeper understanding of the lumbar regions across various mammalian classes could find this research helpful. Once we develop and test the ML models on this dataset, we expect to find that the lumbar vertebrae groupings were largely consistent with existing classifications, with some exceptions. This highlights the potential of unsupervised ML clustering algorithms to advance research in paleontology and related fields.
