Students at UC Berkeley have created an application to help parents determine if their child under 2 years old is at risk of developing autism spectrum disorder. Currently, the standard official screening of autism in children is done at 2 years of age, when a majority of the symptoms have already manifested. While obtaining an official diagnosis earlier is possible, it is an extensive, high-cost process that involves genetic and neurological testing as well as consultation with multiple specialists. This extensive diagnostic process is unfeasible to administer to all infants under 2 years old. By 2 years of age, a lot of the neural circuitry relating to autism has already solidified, which limits the impact of therapy administered after this point. The unfortunate case is that most children with autism are diagnosed too late for therapy to be fully effective.
This is where the Early Autism Detection Team from Data-X at UC Berkeley comes in. Using a state-of-the-art machine learning model, they made an application which parents can use to determine a risk score. This score would allow parents to determine well before age 2 if their child is at risk of autism, so they can decide to pursue the expensive medical tests associated with an early official diagnosis and ultimately early intervention therapy. This easy-to-use application asks the parents to answer 11 behavioral questions about their infant, and also provides a simple guide to aid them in this process. Using data from 9,000 patients with autism and 3,000 control patients they trained their machine learning model to predict the autism diagnosis of the infant. With their model, they were able to predict the diagnosis on a test set of infant data to an accuracy of 90% showing the large potential of this product solving the current diagnostic problem and affecting the lives of thousands of children diagnosed with autism each year.