Technology is catching up with professional clinicians at accurately identifying traits associated with autism spectrum disorders (ASD). There are already programs that can identify patterns in gait or gaze. New research from a collaboration between Duke University and other institutions built on existing technology to develop an algorithm that identifies autism traits using video. This method of identifying autism may help clinicians identify children with ASD who can benefit from early interventions.
The research team developed algorithms that can evaluate children’s behavior according to components of the Autism Observation Scale for Infants (AOSI) evaluation, working as a complement to a trained evaluator. The algorithms measure children’s up-and-down and side-to-side motions by tracking their facial features.
To test the system, the researchers recorded evaluations of 12 children between five and 18 months of age. A trained evaluator performed tests like shaking a toy in one hand and then shaking a toy in her other hand, observing how long it took for the child to focus on the second toy. Evaluators use tests like these to determine if a child has problems tracking objects in real-time, which is an indicator of autism.
The research team’s algorithms agreed with the trained evaluator’s assessment of a child 90% of the time. In contrast, untrained evaluators agreed with the trained evaluator only 53% to 78% of the time. Because the children in the study were too young to be diagnosed with autism, the research team is unsure how effective their system performs as a tool for diagnosis.
The researchers intend to continue working on their system. Their next goal is to develop software that analyzes the behavior of infants interacting with a tablet computer. These automated evaluations could help screen for autism spectrum disorders, especially in rural areas.
This research is published in the journal Autism Research and Treatment.
Previous news in autism:
