When assessing mental health problems: artificial intelligence really wants to beat the doctor

According to a survey by the National Institutes of Health, about 20% of adolescents in the United States have mental health problems. But the tools used by mental health professionals are becoming more intelligent, and they use artificial intelligence to diagnose patients, and the results are usually more accurate than human diagnosis. The author Peter Rejcek has decades of experience in science journalism.

A study published in the journal Suicide and Life-threatening Behavior showed that machine learning accuracy was as high as 93% when identifying suicidal tendencies in patients. The study was led by John Pestian, a professor at the Cincinnati Children's Hospital Medical Center, and involved 379 adolescents from three regional hospitals. According to the university's press release, each patient completed a standard behavioral level assessment and participated in a semi-structured interview, answering five open questions. The researchers analyzed the tester's voiced and nonverbal behavior, and then used machine learning algorithms to get accurate results, to determine whether the tester had suicidal tendencies, whether there was mental illness but no suicidal tendency, and other possibilities.

In his press release, Pentian said that these calculations are technically indispensable for the protection and prevention of suicidal behavior. According to the American Society of Suicide Studies, suicide was the tenth leading cause of death in the United States in 2014, but it is the second leading cause of death among people aged 15 to 24.

A recent study published in the Psychological Bulletin further highlights the need for social tools to prevent suicide. A meta-analysis of 365 studies conducted over the past 50 years found that mental health experts predict that a person's probability of attempting suicide is almost a random probability.

Harvard University writer Joseph Franklin said in an email to Singularity Hub that one of the main reasons for this is that researchers almost always use a single factor (such as a diagnosis of depression) to predict such events. The complex nature behind this patient's thoughts and behaviors requires consideration of dozens or even hundreds of factors to make accurate predictions.

In a submission to Psychological Medicine earlier this year, Franklin et al. stated that machine learning and related technologies are ideal for mental health treatment. Search engines that use only one factor to return results are ineffective, as are attempts to predict suicidal behavior.

Including his colleague Matthew K. at Harvard University. Researchers in Boston, Nock, use machine learning to predict suicidal behavior with an accuracy of 70%-85%. However, the research is still in its infancy and the sample size is small.

Franklin added:

The work of the Pestian team is also very interesting, and the audio mode/natural language processing methods they use are unique in this area. Although there are some limitations, their research is a radically different innovation from what the researchers have been doing for the past 50 years.

According to Franklin, machine learning has not yet been used in treatment, and most conventional methods of treating suicidal mental illness are lacking. Even if several cutting-edge organizations are about to master AI technology that accurately predicts suicidal behavior across the healthcare system, we don't know how to help those who put themselves at risk to reduce risk.