Artificial intelligence can be used to provide a more precise time of death, which can be crucial in e.g. murder investigations. The method was developed by researchers at Linköping University and the Swedish National Board of Forensic Medicine who have trained the AI model on so-called metabolites in thousands of blood samples from real deaths.

“Death is a strong biological signal,” says Rasmus Magnusson, postdoctoral fellow at the Department of Biomedical Engineering, IMT, at Linköping University, who led a study published in Nature Communications where AI is used to determine the time of death.

When the body dies, a number of biological processes set in. Organs and tissues begin to break down, leading to changes in small molecules in the blood called metabolites. They are broken down in a predictable way that correlates with how much time has elapsed since the time of death.

“This enables us to assess the actual time of death of an individual, which is very important in forensic investigations, but also to the work of the police. For example, they need to spend their resources on the right witnesses in the right period of time in the deceased person’s life,” says Henrik Green, professor of forensic sciences at LiU and researcher at the National Board of Forensic Medicine, RMV.

The methods currently used to determine the time of death, also known as the post-mortem interval, include body temperature, rigor mortis and the amount of potassium in the vitreous of the eye. However, these methods yield less accurate results when a few days have passed since the time of death.

The method now developed by researchers at LiU together with RMV instead uses artificial intelligence to analyse the metabolites in blood samples collected at autopsy.

Blood samples from more than 45,000 autopsies have been collected by RMV over a period of almost ten years, resulting in a world-unique database. The samples are used to find various chemical substances such as drugs, pharmaceuticals or toxins. But body metabolites too can be found in the blood samples.

Of these 45,000 samples, 4,876 with known post-mortem interval were used to train the AI model.

“This is a gold mine of data at the National Board of Forensic Medicine. But we were also able to show that there is no need for such large amounts of data that was perhaps previously thought. A few hundred individuals are enough to build corresponding models, which makes our method useful even in laboratories worldwide that don’t have access to as much data,” says Rasmus Magnusson.

The researchers showed that their new model could predict the time from death to autopsy with a precision of about one day even for those deceased for up to 13 days. A clear improvement on current methods. According to Elin Nyman, docent in systems biology at IMT, this was a high-risk project that was not necessarily expected to work.

“We knew that many external factors affect body decomposition and were surprised that the signal from the body’s metabolites was so strong when it comes to predicting the post-mortem interval. The data set we have today provides information on the date of death, but we don’t know the time,” says Elin Nyman.

So the researchers’ next step is to produce a data set with more precise information about the time of death, and then train models that will provide more reliable estimates of the post-mortal interval as well as be able to determine during which part of the day a death occurred.

“Forensic assessments often involve puzzle-like detective work. This new tool gives us better opportunities to assess how long someone has been deceased even when a long time has passed since their death, which is of great importance especially in more complex cases. We’re now working on developing even more accurate models,” says Carl Söderberg, forensic pathologist and researcher at RMV.

The study was funded mainly by the Swedish Research Council, the foundation Forska utan djurförsök (Research without animal experiments) and the Strategic Research Area in Forensic Sciences at LiU and RMV.