KEY POINTS
- Current autism screening methods rely solely on parent report, which is not always accurate;
- Studies suggest that children of color and girls are more often missed during routine autism screening; and
- An app based on AI, administered on a tablet during a well-child visit, can detect early signs of autism.
Researchers at Duke University have been exploring whether Artificial Intelligence (AI) can be used for autism screening in real-world settings, such as primary care. Their goal is to develop an objective, scalable tool for early autism detection that does not rely on parent report. The American Academy of Pediatrics recommends that all children be screened for autism at their 18- and 24-month well-child checkups. The most common way children are screened is with a 20-question parent survey which asks parents questions, such as “Does your child respond when you call his or her name?” Based on the parent’s responses, the pediatrician may follow up with an interview to clarify the parent’s responses. If the parent endorses enough signs of autism, the pediatrician should refer the child for a diagnostic evaluation.
Although screening for autism using a parent questionnaire is helpful, it has some limitations. Not all parents understand the questions well enough to answer them accurately. The questionnaire might not be offered in the parent’s primary language. Pediatricians do not always follow up with an interview when it is needed. This reduces the accuracy of the questionnaire. In real-world settings, such as pediatric primary care where screening should occur, large-scale studies have found that a questionnaire is not as accurate as it should be. Children of color and girls, in particular, are often missed.
In other areas of health, doctors rely on multiple sources of information to determine whether a person is more likely to have a medical condition. Suppose, for example, you were concerned that you have a heart problem. Your doctor would inquire about your symptoms, such as asking you if you feel more tired than usual, get out of breath, or are experiencing chest pain. Your self-report would be an important part of your doctor’s screening for heart problems. Your doctor likely would then follow up with several objective measures of heart function. An electrocardiogram (EKG) and blood pressure test would be ordered, as well as a blood sample to measure your cholesterol. Your doctor would integrate multiple sources of information to make an informed decision regarding your risk for heart problems. We do not have any objective tests for autism, such as a blood test. Autism is a behavioral condition and is diagnosed through behavioral observation. Can AI, in other words, a computer rather than a human being, detect and objectively measure early behavioral signs of autism?
The research team at Duke University has developed a digital app that can be downloaded on a smartphone or tablet and used to screen for autism. Here is how it works:
The app displays a series of brief, fun movies that have been strategically designed to elicit autism-related behaviors, such as gaze, orienting to a name call, and facial expressions. While the child watches the movies, the camera embedded in the smartphone or tablet records the child’s behavioral responses. A technique called “computer vision analysis” then automatically and precisely measures the child’s behavioral responses. The computer can measure whether the child noticed the social elements of the movie, such as people, or was mostly paying attention to the nonsocial elements, such as toys or other objects. It can measure the child’s facial expressions, blink rate, orienting to their name, and other body movements. The app then uses machine learning to integrate all these behavioral signals and determine how likely it is that the child is autistic. It is all done in less than 10 minutes using only a smartphone or tablet.
One of the advantages of using a computer to measure behavior (called “digital phenotyping”) is its high degree of resolution and accuracy. The computer can measure subtle but distinctive behaviors that are impossible for the human eye to detect. For example, one of the early signs is a child’s blink rate. Neurotypical children suppress their blink rate when they look at people whereas autistic children suppress theirs when they look at objects. The computer can measure whether a child orients when their name is called; however, it can also tell how quickly the child turns their head. Autistic children who do orient to their name do so about a second more slowly compared to neurotypical children. The clinician likely would have missed this early sign.
In a recently published study, a digital autism screening app was administered to 475 children during a pediatric well-child visit, 49 of whom were subsequently diagnosed with autism and 98 with developmental delay without autism. The app showed 87.8% sensitivity for detecting autism, meaning it correctly identified most children with the condition. Its specificity — the percentage of children without autism who screened negative — was 80.8%. The app was equally accurate in its ability to identify autism in children of color and girls. Thus, this more objective approach may help reduce disparities in early detection, which is the first step to accessing early intervention services.
Using computer vision to detect early signs of autism is just one example of how AI is transforming healthcare. A computer will never fully replace a human when it comes to providing high-quality, compassionate medical care. When used ethically and responsibly; however, AI offers many benefits and has the potential to increase access to health services and deliver them more efficiently and equitably.
References
- Guthrie W, Wallis K, Bennett A, Brooks E, Dudley J, Gerdes M, Pandey J, Levy SE, Schultz RT, Miller JS. Accuracy of Autism Screening in a Large Pediatric Network. Pediatrics. 2019;144(4).
- Dawson G, Sapiro G. Potential for Digital Behavioral Measurement Tools to Transform the Detection and Diagnosis of Autism Spectrum Disorder. JAMA Pediatrics, 2019;173(4):305-6.
- Perochon S, Di Martino JM, Carpenter KLH, Compton S, Davis N, et al. 2023a. Early detection of autism using digital behavioral phenotyping. Nature Medicine, 29: 2489-97.
REFERENCE: Psychology Today; 20 DEC 2023; Geraldine Dawson, Ph.D