My research project was on generating a 3D reconstruction of ancient fossils (565 million years old) from Siberia. Working with over 2000 cross-sectional images of the fossil sample, which is encased in rock, I developed a machine learning-based model to distinguish the fossils from the non-fossils. My methodology was designed to take advantage of 3D relationships, as I iterated it over three orthogonal axes of the rock sample; it also had the advantage of being automated and quantitatively consistent. Ultimately, this allowed me to produce a full 3D model of the fossils.
These fossils are highly significant due to their age. The ancestors of most modern-day animals evolved about 540 million years ago, and we have very scant evidence of animals before this time. If my sample represented an ancient animal, we would expect a regularity of shape. We found one particular shape—an elliptical “rolled sheet”—to consistently appear in our 3D reconstruction. We quantitatively analyzed these “rolled sheets” to determine that their interior edges were consistently rougher than their exterior edges.
This Siberian sample could very well be an ancient animal. However, we need more evidence to say so for sure. This could come in the form of comparing to a visually similar sample from Australia, which could be an ancient sponge. If sponges truly did exist this long ago, they could have dramatically affected global climate, biodiversity, and evolution. In today’s world of unprecedented change, these potential ancient animals could offer us insight into our own future.