When studying fossils, we rely on morphology, unpinning its geological past using methods that can sometimes damage specimens. How else can we open doors to new discoveries in palaeontology without using our conventional modus operandi?
Computed tomography (CT) is a leading technique to observe fossil interior structures – paving ways for pioneering breakthroughs without damaging specimens. 3D graphics of fossils are produced from thousands of X-ray images taken at various angles. These graphics are useful for research and education, and can be reconstructed into physical models using 3D printing. CT scans can uncover previously concealed structures, providing insights into various unknowns, including a specimen’s inner ears, brain cavity and body mass. Although now an established fixture in vertebrate palaeontology labs, using this ingenious method is not as plain sailing as may appear – much time is needed to analyse the data and many regard it as taxing.
Congyu Yu, from the American Museum of Natural History, New York, USA, and colleagues use CT scanning to harness greater understanding of Protoceratops embryonic skull fossils from the Gobi Desert, Mongolia. The team trained a deep neural network (an artificial intelligence, AI, method) using over 10,000 scans of these embryonic skulls. Their results show that the AI model can differentiate between the fossils and surrounding rock matrix within the CT images. The results don’t boast the same accuracy as those conducted by human-controlled methods – the benefit of the AI approach is the remarkable speed of comparison, reducing the time needed for analyses from days/weeks to minutes. But should we compromise on accuracy to accommodate quicker processes?
CT scanning combined with AI provides a promising approach to this new era of research. When partnered with proficient excavation and an aptitude for anatomical description, one can only imagine the profound advances that lie ahead for palaeontology.
Amelia Jane Piper
DETAILS: Front. Earth Sci., 27, 2022; https://doi.org/10.3389/feart.2021.805271
