Using technology to detect autism sooner

UM psychology and engineering professors are collaborating to create a program that evaluates children for autism using digital data.

While doctors agree that children can be diagnosed with autism as early as age two, the average age of diagnosis is about 4 years old. And for minority children, research indicates that age is much older.

Yet the earlier children are diagnosed, the quicker they can take advantage of interventions that can improve their lives dramatically.

This glaring disparity prompted University of Miami psychology professors Lynn Perry and Daniel Messinger to team up with electrical and computer engineering professor Mei-Ling Shyu, physics associate professor Chaoming Song, psychology assistant professor Sierra Bainter and professor Michael Alessandri, to create a more effective way to detect autism symptoms using technology.

“One of the motivators for this line of work is to improve access to an [autism] diagnosis for everyone,” Perry said, noting that a 2018 study showed that white children were accurately diagnosed 7 percent more often than black children and 22 percent more often than Hispanic children. “We know with autism spectrum disorder, having access to early intervention is key for promoting healthy developmental outcomes in language, cognition, and social interaction.” 


Today, if a parent suspects their child may be on the autism spectrum, they must take them to a licensed psychologist and undergo an assessment that includes a series of structured activities, such as a pretend birthday party, popping bubbles, and blowing up a balloon. The psychologist then scores the child’s behavior to determine a diagnosis.

“Right now, the state of the art is that highly trained clinical experts make a diagnosis, but we do not have an objective measure of autism symptoms,” Messinger said. “For example, one problem that children with autism have is they often engage in less frequent eye contact, so with this project we ascertain how often they make eye contact.”

According to the Centers for Disease Control, one in 59 children are diagnosed with autism spectrum disorder (ASD) in the United States. Autism is a developmental disability that can cause social, communicative, and behavioral disabilities, but the condition affects each person in different ways. Autism affects boys three times more often than girls, research indicates.

Supported by UM’s Clinical and Translational Research Institute, the study, named BIAS, for the Behavioral


Imaging of Autism for Science, focuses on measuring social-communicative behaviors. These behaviors including language, facial expressions like eye contact and smiling, restricted interaction with others, and repetitive actions, are often key indicators that an individual may be on the autism spectrum, Perry and Messinger said.

Researchers are measuring these symptoms using special glasses, which record videos of the child’s facial expressions and movements, as well as audio recorders that capture language while each child undergoes the typical autism assessment. Computer vision software is then used to calculate and analyze the prevalence of certain behaviors, which helps diagnose the severity of autism symptoms, Perry said.

So far, 43 children between the ages of 2 and 5 have participated in the study through the Center for Autism and Related Disabilities (CARD) at UM, which is currently completing autism assessments for families throughout Miami-Dade County with funding from The Children’s Trust.

Initial trials have shown promising results, Messinger said. One program — which Shyu’s team of engineering graduate students Saad Sadiq and Yudong Tao developed — uses machine learning to analyze a child’s language skills. So far, the computer’s scores were able to match the psychologist’s diagnosis 70 percent of the time, which is nearly twice as effective as existing technology. 

Shyu, Sadiq and Tao
Shyu (center), Sadiq and Tao (left to right).

“This is a step toward developing technology that can be used to assist experts in making more informed assessments,” Shyu said. “We could simply record audio when they are at home and then produce a symptom severity score.”

The professors are now looking for more participants to expand their research. They would like to have 150 families participate before the study is completed, Messinger said.

If they succeed in crafting an effective program, it could remove obstacles for families whose children often are misdiagnosed, or not diagnosed until they are much older, Perry said. 

“If people live in rural or remote communities where they might not have access to expert clinicians easily, but they do have access to video, they can record a parent and child interacting, and that video can be transmitted to someone who can use this computer vision program to better recognize kids who should be getting clinical help and attention,” Perry said.

Beyond the convenience this technology could offer, UM researchers want to examine whether children diagnosed with autism in a clinic behave similarly in other settings.

“It’s assumed that across lots of different contexts – the clinic, preschool, at home, on the playground – that children’s symptoms should be similar, but we don’t actually know that the severity of behaviors or symptoms would be the same,” Perry said. “Knowing about how children’s behaviors manifest in different contexts is really important, so that interventions can help children succeed in their everyday activities.”