Teeth often survive when little else does, which makes them valuable evidence when unidentified human remains are found. Reading them well takes specialized training in dental morphology, the shape and features of the teeth, and relatively few forensic anthropologists get it during their education, according to the authors of a new paper.
Their answer is FoRDent, a tool co-developed by Dr. Joe Hefner, associate professor in the Department of Anthropology at Michigan State University. As a practitioner scores each dental trait, FoRDent shows a written description, drawings of how the trait varies and, for most traits, two photographs of real examples, all on the same screen.
Hefner and Dr. Heather J.H. Edgar of the University of New Mexico described the tool in a technical note published Oct. 18 in Forensic Science International. Hefner developed the software and ran the statistical analysis.
FoRDent estimates population affinity, one of four parts of a biological profile along with age, sex and stature. Population affinity is an estimate of which reference populations a person’s traits most resemble. It does not identify anyone, but it helps investigators narrow the pool of missing persons an unknown individual should be compared against.
What the tool does
To use FoRDent, a forensic anthropologist examines an unidentified person’s teeth and records up to 65 traits, such as whether the front teeth are shovel-shaped or whether a tooth never developed. For each trait, the tool shows a written description, drawings of the range of expression and, for most traits, two photographs of real examples. An earlier tool, rASUDAS, relies on published references for trait descriptions and images. FoRDent was designed to be as accessible as possible for users with minimal training in dental anthropology.
The tool compares cases against reference data from 6,052 individuals. Missing data is common in forensic casework, so FoRDent builds a new model for each case using only the traits the user could actually score.
Three levels of comparison
FoRDent returns results from three independent models: a Continental model with four groups; a U.S. model with six groups based on U.S. census categories plus probable migrants found deceased along the U.S. Southwest border; and a Finer resolution model with 23 regional groups in the United States and Mexico.
The paper is direct about the tradeoff. Measured by balanced accuracy on individuals held out from training, the Continental model scored 64%, the U.S. model 58% and the Finer resolution model 26%. Accuracy drops as the comparison gets more specific, and the authors describe a Finer resolution result as a contextual lead rather than a conclusion of equal weight. Practitioners decide which level fits the case.
How the reference groups are defined
Edgar and Hefner treat population affinity as locally defined: the categories should mean something to the community a practitioner serves. Where possible, group labels in FoRDent’s reference data come from how people described themselves, or from next of kin or community members.
“FoRDent compares teeth, not racial identities,” Hefner said. “Some of our U.S. reference groups have names readers will recognize as racial or ethnic categories. Others describe where a sample came from, such as Mexico City or Yucatán. Those labels only show who is represented in the comparison; they don’t divide people into biological races and they can’t tell us how a person identified in life or ever establish who they were.”
Funding came from Congressionally Directed Spending through the National Institute of Standards and Technology, the National Science Foundation, the American Museum of Natural History and the University of Tennessee Forensic Anthropology Center.
FoRDent is free to use at fordent.streamlit.app.