A fascinating study has demonstrated a new technique that can identify children with anxiety and depression just by analyzing their movements. Using a machine learning algorithm that examines movement tracked by a wearable motion sensor, the system is claimed to identify children with psychological disorders better, and faster, than many current methods.

It is estimated that around 20 percent of young children suffer from what are known as "internalizing disorders." These conditions can include anxiety and depression, but are notoriously difficult to identify due to the difficulties in children being able to reliably self-report symptoms and the often unobservable nature of the disorders. Early development internalizing disorders in children often precede later health problems such as substance abuse and suicide.

"Because of the scale of the problem, this begs for a screening technology to identify kids early enough so they can be directed to the care they need," says Ryan McGinnis, explaining the motivations behind the research.

The study focused on training a machine learning algorithm to distinguish children with anxiety and depression based on small physical movements. To do this the researchers recruited 63 children between the ages of three and seven, about a third of whom had previously been diagnosed with an internalizing disorder.

The children were fitted with wearable movement sensors and then subjected to a mood induction task designed to induce certain feelings such as anxiety. Traditionally, highly trained therapists would observe these behavioral tests and subsequently generate a diagnosis but the researchers suspected a trained algorithm could do the same job, and they were correct.

Using just 20 seconds of movement data from an early stage in the mood induction task, the algorithm was ultimately able to discern the children with internalizing disorders from those without, at an accuracy rate of 81 percent. The algorithm was more accurate at identifying internalizing disorders than a diagnosis generated from what is called the Child Behavior Checklist, a parent-completed questionnaire with 120 items related to behavioral issues.

"Something that we usually do with weeks of training and months of coding can be done in a few minutes of processing with these instruments," says Ellen McGinnis, a clinical psychologist working on the project.

The researchers are planning on further refining the algorithm with larger volumes of subjects, as well as incorporating other data such as voice analysis, to increase the specificity of the results. Ideally, the system will ultimately be able to distinguish between behaviors such as anxiety and depression. In the long term the researchers suggest this technology could be introduced into schools to help quickly identify children that need special assistance, or even used in doctor's clinics as standard developmental assessments.

The research was published in the journal PLOS ONE.