Evrim Yazgin
Cosmos science journalist
Researchers at the CSIRO, Australia’s national science agency, have developed a new artificial intelligence (AI) tool for determining the biological sex of individuals – from their skulls.
The tool has the potential to speed up investigations in criminal analysis or after severe natural disasters.
Results published in Scientific Reports show the AI tool achieved an accuracy of 97%. This is significantly better than the 82% accuracy of human assessments. And it is much quicker.
The researchers note that current methods – such as that developed in 2008 by leading physical anthropologist Philip Walker – are often limited because they were derived from specific ethnic groups and from skulls obtained in certain time periods.
“For instance, the Walker method was derived from English/American and Native American population groups, and its application outside of the US reported classification accuracies that were lower,” the authors write.
The CSIRO team turned to digital data to produce a deep-learning AI model which can sex skulls.
They collaborated with forensic anthropology researchers from the University of Western Australia who provided labelled data and expertise in forensics to help produce the model.
A dataset of 200 CT scanned skulls was analysed by the AI algorithm and compared to human assessors. The scans were collected at Dr. Wahidin Sudirohusodo General Hospital at Hasanuddin University, Indonesia.
“Our AI tool produces its results approximately 5 times faster than humans can, meaning families waiting for results of investigations can receive news about their loved ones more quickly,” says joint first author of the study, Hollie Min.
“This AI tool has the potential to support forensic anthropologists to enhance the accuracy of sex estimations, while reducing the potential impact of human bias.”
“This collaborative study allowed us to address some of the perceived limitations of traditional methods and better account for diversity in forensic data,” she adds.
“Future research is needed, especially around expanding datasets to include diverse populations, enhancing the robustness and generalizability of the AI framework. Our goal is to provide forensic anthropologists with a reliable, interpretable tool to support their critical work, especially in cases involving individuals of unknown population backgrounds,” Min says.
“Our team is currently looking for industry collaborators to develop and translate this technology for real-life applications.”
