Autism Diagnosis Through a Home Video Analyzed by a Machine Learning Model

Machine learning models have reduced the load for diagnosis. Instead of diagnosing through time-consuming tests (ADOS + ADI) performed by qualified experts, it is possible to diagnose autism in a child with a high sensitivity level.
Outstanding striking finding that unprofessional people trained to identify only eight child behaviors is enough to diagnose autism.
Furthermore, watching a home video is shorter than 5 minutes.
Trained nonexpert people score only 8 child behaviors. Their scoring results diagnosed autism from watching child videos at great accuracy (90% sensitivity and 75% specificity) . “The primary motivation for such research”,
says prof. Dennis Wall, pediatrics and biomedical data science professor at Stanford University and a lead researcher in this study,
“is the long wait times for diagnosing children with suspected autism,” which is a bottleneck for treatment that can benefit them.
Currently, the standard of diagnosis requires time and expertise, burdens the diagnostic child development centers,
and prolongs parents’ waiting for services. This burden is evident in all countries. The research team used machine learning,
which learned from scoring behaviors of children with autism and children with typical development.
The machine has identified a minimal set of eight behaviors sufficient to the clinical diagnosis of autism.
Learning has relied on the hypothesis that has emerged from previous studies, that models that use a few behavioral characteristics
are reliable and accurate in diagnosing autism. The videos were 1-5 minutes long and demonstrated the child’s face
and hands while showing him various social opportunities and filming him using toys and colors. To read the article.

Tariq, Q., Daniels, J., Schwartz, J. N., Washington, P., Kalantarian, H., & Wall, D. P. (2018). Mobile detection of autism through machine learning on home video: A development and prospective validation study. PLoS medicine15(11), e1002705.