Duke Center for Autism Awarded $12M Research Grant to Use Artificial Intelligence to Detect Autism

The Duke Center for Autism and Brain Development has been awarded a $12 million federal grant to develop artificial intelligence tools for detecting autism during infancy and identifying brain-based biomarkers of autism. The grant, from the National Institute of Child Health and Human Development, extends the Duke Autism Center of Excellence research program for an additional 5 years. Geraldine Dawson, PhD, director of the Duke Center for Autism and Brain Development and professor of psychiatry and behavioral sciences, will lead a team of researchers that includes Duke faculty from psychiatry, pediatrics, biostatistics and bioinformatics, computer and electrical engineering, and civil and environmental engineering.

“We are thrilled to receive this award, which allows Duke to remain at the forefront of autism research,” Dawson said. “Our goal is to use advanced computational techniques to develop better methods for autism screening that will reduce known disparities in access to early diagnosis and intervention.”

In a project led by Dawson and Guillermo Sapiro, PhD, professor of electrical and computer engineering, researchers will extend earlier work in which they developed a digital app, deployed on a smart phone or tablet, to videotape young children’s behavior. Artificial intelligence based on computer vision analysis automatically codes the videotapes to identify behavioral characteristics of infants and toddlers who are later diagnosed with autism and track their development. Using a computer, the app can detect differences in the child’s facial expressions, vocalizations, and gaze that are early signs of autism.

The new award will extend this work in two ways: First, the team has developed a version of the app that parents can download and use at home. The previous version was administered in the clinic. The goal is to provide access to early detection of autism signs to families who might not be able to travel to a medical center for an assessment, including families from rural locations and those who lack the resources to travel to Duke. Second, in the newly funded study, families will use the app over time to track changes in their child’s development and assess responses to early intervention.

A second project, led by Benjamin Goldstein, PhD, associate professor of biostatistics and bioinformatics, and Gary Maslow, MD, MPH, associate professor of psychiatry and behavioral sciences and co-director of the Division of Child and Family Mental Health & Community Psychiatry, will extend earlier work led by Matthew Engelhard, MD, PhD, which developed an algorithm to screen for autism based on information obtained from an infant’s early medical records. In the newly funded project, the team will use artificial intelligence to analyze a much larger data set with the goal of developing an early detection algorithm that could be used by health systems outside of Duke. The team will examine 260,000 health insurance claims, including those from 6,000 children diagnosed with autism, from birth to 18 months.

That data also will be used to identify the nature of early medical conditions associated with a later diagnosis of autism. Based on the algorithm, a team led by Lauren Franz, MBChB, MPH, assistant professor of psychiatry and behavioral sciences, will develop support tools to help primary care providers screen and guide patients. The goal is to develop an objective way to alert pediatricians and other providers that an infant has a higher likelihood of autism and then to automatically provide advice for how to link parents with appropriate services for their child.

The third project, led by Kimberly Carpenter, PhD, assistant professor of psychiatry and behavioral sciences, and David Carlson, PhD, assistant professor of civil and environmental engineering, will use artificial intelligence to monitor brain wave activity, which is synchronized with videotaped behavior of three- to six-year-old children diagnosed with autism. The data will be used to identify specific brain networks associated with behaviors characteristic of autism. The team will use this information to validate a brain-based biomarker, or neural signature, that is characteristic of autism which can be used in clinical trials to better understand how therapies influence brain function in autism.

______