With the help of CT scans, researchers can see dinosaur fossils without having to remove the surrounding stone.
However, the large datasets of X-ray images – often over 2,000 images per sample – need to be manually segmented in a computer to see what is fossil and what is stone. Then, they are assembled into a 3D image, a task that can take months.
With the help of AI expert Dmitry Konovalov, paleontologist Espen Knutsen at Australian James Cook University segmented around 40 of 2,000 images in a set and used them to train a deep learning model. The AI could then complete the analysis on its own.
"We obtained a high-precision 3D representation of a small Triassic reptile from Queensland that was around 240 million years old. It took days, instead of months," says Knutsen in a press release.
Now, researchers hope that the technique can be used to quickly bring forth new fossils, which can help paleontologists move forward. The study is published in Scientific Reports.
