Scientists use AI to evaluate dental anthropology

Mario Modesto Mata is a researcher in the Dental Anthropology Group at the Center for Human Evolution and Human Research (CENIEH) and the lead author of a paper published in 2006. Anatomy recordon the use of an artificial neural network to reconstruct the number of perifossae, which are enamel growth lines but are absent in worn teeth.

Teeth are an almost inexhaustible source of information from both a biological and taxonomic point of view. From the way it grows, you can directly count its lineages and estimate the formation time in days. However, counting the periodontal margin can be difficult depending on the condition of the tooth. This is because as the periodontal margins wear down through normal use, some of the margins are lost along with the loss of enamel.

“Solving this problem is very important, as it allows us to increase the number of teeth suitable for evolutionary studies, leading to more reliable conclusions,” asserts Modesto Mata from the European project Tied2Teeth. (ERC-2021) -ADG) is led by researcher Leslea Hlusko.

According to the paper, once we know how much enamel has been lost (which is measured as a percentage of the crown height lost), artificial intelligence techniques can be applied to improve modern human teeth. It is said that the number of missing circumferential processes can be predicted.

Specifically, an artificial neural network was developed to predict the number of circumferential projections when teeth are missing up to 30% of their crown height. Validation of the neural network showed that in 86% of cases, when 30% of the enamel is missing, the maximum error was only a total of three circumferential projections.

“These data on growth lines are so accurate that we can predict when enamel will be fully formed very close to reality, showing that neural networks can be used to investigate paleontological questions. “We're working hard,” says Modesto Mata.

To get the most out of the use and application of these neural networks, the authors of this study developed software in the form of an R package called thesR (from “teeth aRe excellent”), which can be freely distributed and installed. No AI training is required to know how to use it; a very basic knowledge of R is enough. The functions developed within the package allow you to make predictions very quickly.