Suicide is a topic that hits home for Rafael Zamora-Resendiz.
His friends and family members experienced depression and opioid use, said Zamora-Resendiz, a computational engineer at the Lawrence Berkeley National Laboratory.
Zamora-Resendiz graduated from Hood College in 2017, and while in his undergraduate years, he worked with professor Xinlian Liu. At Hood, the two began working on a project that involved using algorithms to identify if someone might kill themselves.
“I’ve always been interested in applying my knowledge of computer science to help real world problems,” Zamora-Resendiz said.
After graduating, he went to the Lawrence Berkeley National Laboratory to continue work on the project. Now, he and Liu work together to try and address suicidal rates.
The pair are part of a project that looks to use computer algorithms to identify when someone might be suicidal.
The project is still in the early stages, Zamora-Resendiz said. It will use data from the Department of Veterans Affairs that has been defragmented, a process that protects a personal identity, to build and train the algorithm to identify suicidal patterns. From there, the tool could be used by doctors outside of the VA to identify patients that might be thinking about suicide, he said.
“Dr. Liu always tells us, ‘Yeah, we won’t be able to catch all of the suicide attempts, but as long as we are able to catch maybe five or 10 that’s going to make a big difference to a lot of people,” Zamora-Resendiz said.
Artificial intelligence — the broad subject under which machine and deep learning falls — has been around since the 1950s, Liu said. Liu, Zamora-Resendiz and their colleagues are working with deep learning, which involves computer neural networks identifying features.
Instead of looking at pictures of people or animals, the computer networks look at pictures of medical records, Zamora-Resendiz said.
“The specific algorithms that we’re using are kind of trendy right now,” he said.
Part of the pattern is looking at unexpected visits to the hospital, Liu said. The approach is to identify patients before something might happen because there is a short window between suicidal ideation and acting on it.
Rates of suicide in the U.S. are increasing with a 33 percent increase since 1999, making rates the highest they have been since World War II, according to a recent TIME article. Historically, veterans have higher rates of suicide compared to those who did not serve.
“That makes us feel like what we’re doing is pretty meaningful for human lives, something we really care about,” Liu said.
There are challenges to identifying suicidal patients, Liu said, including opioid use. An overdose may or may not be suicidal in nature.
Some people say they might not want to live, but that does not mean they’re thinking about suicide, he said.
Another challenge is keeping veteran data private. The project was approved by research and ethics boards, Liu said, and the project researchers underwent bioethics training.
The data can only be handled by the supercomputers the team uses for the project, he said.
Right now the team is still waiting on the data from the VA. Instead, they are using publicly available data sets like data from the intensive care units in two Boston hospitals, Zamora-Resendiz said.
Although they are not helpful for predicting suicide, the data sets can help the researchers practice training algorithms so they are ready once they have access to the data.
Then they can start preparing an algorithm that might save lives.
The study was presented at the Society for Industrial and Applied Mathematics, or SIAM. SIAM is headquartered in Philadelphia, Pennsylvania, is an international society of more than 14,000 individual, academic and corporate members from 85 countries. SIAM helps build cooperation between mathematics and the worlds of science and technology to solve real-world problems through publications, conferences, and communities like chapters, sections and activity groups. For more information go to: siam.org