AI analyzes chest radiographs to quickly shortlist potential matches in forensic cases

An interdisciplinary team comprising faculty and doctoral students from the Department of Anthropology and Computer Science and Engineering has found a way to use artificial intelligence (AI) to help forensic anthropologists identify individuals faster and more efficiently.

Members of the Michigan State University Forensic Anthropology Lab, including Dr. Carolyn Isaac, Dr. Todd Fenton, Dr. Joseph Hefner, and doctoral student Alexis VanBaarle, co-authored a new study published in IEEE Access that analyzed more than 5,000 chest radiographs, identifying different regions of interest that aid in identifying a person. The study used deep neural networks, a type of AI program, that allowed large numbers of radiographs to be analyzed in a fraction of the time.

“In mass fatality situations when a large number of individuals require identification, this system can assist by short-listing potential matches for a practitioner to visually assess,” Isaac said. “It can do this for more than 1,800 radiographs in 17 seconds rather than the 30 to 60 hours it would take a human practitioner.”

Isaac shared that this research could also be used in unidentified or missing person databases to propose potential matches for consideration, which helps reduce practitioner bias.

“These (deep neural networks) compare target radiographs to thousands of others to find the most likely matches,” Isaac said. “This research shows how AI can be used to enhance forensic casework by making tasks more efficient.”

This AI approach is the first of its kind to evaluate how different ROIs within radiographs can be used for human identification in forensic contexts.

“There has not been this type of application previously, so it is showing the computer science world how forensics uses radiographs differently than the medical field, which primarily uses them to diagnose disease,” she said.

Isaac said she has enjoyed collaborating with the team of researchers to develop this approach, which includes Dr. Arun Ross and Redwan Sony of the iPROBE Lab in Computer Science and Engineering.

“I love when we are brainstorming on the project and get to see the unique perspectives of computer science versus the domain experts in forensic anthropology,” Isaac said.