About 20 percent of children suffer from “internalizing disorders” commonly referred to as anxiety and depression. And adults are tasked with recognizing potential mental health problems in children by inferring to their mental state because it’s difficult for children under the age of eight to articulate their emotional suffering. But the process is always difficult and most times result in late treatments. By using the speech patterns of children, a newly developed Artificial Intelligence (AI) algorithm can detect signs of internalizing disorders, claims a research published in the Journal of Biomedical and Health Informatics.

Children miss out on vital early depression treatment due to diagnosis challenges which include their parents’ failure to recognize the symptoms, assurance issues, and long queues for appointments with psychologists.

“We need quick, objective tests to catch kids when they are suffering,” says the study’s lead author Ellen McGinnis who is a child psychiatry clinician at Vermont Center for Children, Youth and Families, University of Vermont Medical Center. “The majority of kids under eight are undiagnosed.”

A crucial part of internalizing disorder treatment is early diagnosis because children respond very well at the development stage of their brains. Late treatment exposes the children to the risk of substance abuse and greater chances of committing suicide later in life.

Behavioral characteristics of patients with internalizing disorders include loneliness, anxiety, withdrawal, and depression. Standard diagnosis lasts for about 90 minutes and involves a semi-structured interview with the child’s primary caregiver and a trained clinician. McGinnis has been researching with the study senior author Ryan McGinnis, an engineer at University of Vermont biomedical, for ways to make the diagnosis more reliable and faster using machine learning and artificial intelligence.

Trier Social Stress Task investigative protocol

The team used the Trier Social Stress Test Protocol – an adapted version of a mood induction task which is engineered to force the subject into feelings of stress and anxiety. The study involved 71 children aged 3 – 8 who were asked to improvise a 3-minute story. The participants were told they would be judged based on their creativity – how interested their version was. As the judge, the researcher remained strict while the children deliver their stories, and gave only negative feedback. A buzzer would sound after every 90 seconds and again after 30 seconds, with the judge having to remind the children how much time they have left.

“The task is designed to be stressful,” says Ellen McGinnis. The idea is to put them into pressure and into a mindset that they are being judged by someone.

Unlike the standard diagnosis pattern, a structured clinical interview and a questionnaire were used to diagnose the children, as well-established ways of identifying internalizing disorders in children.

The algorithm worked with 80 percent accuracy

The researchers analyzed statistical features of each child’s story (audio recordings) using a machine learning algorithm and related them to results from the child’s diagnosis. The team found that the most predictive diagnosis was the phase of the recordings, between the 90 seconds and 30 seconds buzzers. And that the algorithm (artificial intelligence) was highly successful at diagnosing internalizing disorders in children.

“The algorithm was able to identify children with a diagnosis of an internalizing disorder with 80 percent accuracy,” says Ryan McGinnis. “In most cases, the result compared really well to the accuracy of the parent checklist.”

The algorithm, which requires just a few seconds of processing time, identified eight different audio features, with three specifically indicating internalizing disorders. They include a higher-pitched response to the buzzers, low-pitched voices and repeatable speech content. These features are a perfect match for someone suffering from depression, says Ellen McGinnis

 “A low-pitched voice and repeatable speech elements mirror what we think about when we think about depression: speaking in a monotone voice, repeating what you’re saying”

The researchers found something similar to the higher-pitched response to the buzzer in their previous work, where children with internalizing disorders completing a fear induction task were found responding to the fearful stimulus by largely turning-away.

The fear task is more cumbersome. It requires a toy snake, a darkened room, a guide and motion sensor attached to the child. But the voice task requires only a way to record speech, a judge and a buzzer to interrupt them which makes it easier to use in a clinical setting, says Ryan McGinnis.

Developing the algorithm for speech analysis into a universal screening tool for clinical use is the next step, says Ellen McGinnis. It could available as a smartphone app that could record and analyze results immediately. The technology could also be integrated into motion analysis to help identify anxiety and depression in children at a very early stage.