A machine learning tool for speech analysis has been shown capable of diagnosing childhood depression and anxiety with 80% accuracy.
This is better than the 67% to 77% accuracy achieved by standard clinician evaluation of pediatric patients and their symptoms as reported by parents.
The study behind the findings, jointly authored by researchers at the Universities of Vermont and Michigan, is running online in the Journal of Biomedical and Health Informatics.
The team had 71 children between 3 and 8 years old tell a little story, challenging them to make it as interesting as they could. One of the researchers played the role of a poker-faced judge to keep things moving. Recordings of the storytelling sessions were analyzed by the experimental algorithm.
On analysis, the speech features most indicative of childhood anxiety or depression, or of the absence thereof, were low-pitched voices with repeatable speech inflections and content—i.e., speaking monotone and repeating points—along with high-pitched vocalizations following a surprising stimuli (a buzzer).
The tool scored similarly to the traditional methods when it came to ruling out depression and/or anxiety in healthy children.
“These results point toward the future use of this approach for screening children for internalizing disorders so that interventions can be deployed when they have the highest chance for long-term success,” the authors concluded.
In a related news release, lead author Ellen McGinnis, a PhD candidate in clinical psychology at Vermont, suggests the algorithm may augment current practices.
“We need quick, objective tests to catch kids when they are suffering,” she says. “The majority of kids under 8 are undiagnosed.”
The team points out that anxiety and depression disorders in childhood are risk factors for later substance abuse and other behavioral problems, including suicide.
The work follows the April publication of a study in which an AI voice-analysis tool diagnosed PTSD in military veterans to the tune of 90% accuracy.