Seeing into the future of autism detection

According to the CDC, about 1 in 36 children in the U.S. has autism. An early and accurate diagnosis is hugely important for children and their parents as they learn how to navigate life on the spectrum. A 2023 study in Nature describes a new method to facilitate early detection of autism using a smartphone app.

Hosts: Andrew Saintsing, PhD, Intern, and Stephanie DeMarco, PhD, Managing Editor

Guests:

Geraldine Dawson, Duke University

Josephine Barbaro, La Trobe University

More on this podcast:

DDN Dialogues is a new podcast from Drug Discovery News. Join us as we explore the stories behind the latest advances in drug discovery research.

Click here to be notified when new episodes of DDN Dialogues are posted!

Transcript

Stephanie DeMarco: Hello everyone! Welcome back to a new episode of DDN Dialogues! I’m your host Stephanie DeMarco. Today, we have a very special episode. We’re joined by Andrew Saintsing, who was one of our 2023 Drug Discovery News interns.  He’s going to tell us about new developments that could help neurodivergent children and their parents navigate the world.

Andrew Saintsing: That’s right, Stephanie. A team of researchers from Duke University has developed a smartphone app to detect signs of autism in toddlers.

DeMarco: Wow, that sounds super interesting! I can’t wait to learn more about this app, but we should start from the beginning. What exactly is autism?

Saintsing: That’s a great question, and one that’s not so easy to answer. Individuals with autism are neurodivergent, so they display atypical thoughts and behaviors. But according to Duke University psychologist Geraldine Dawson, these differences can manifest in a variety of ways.

Geraldine Dawson: Autism is very heterogenous. So, the saying goes: if you’ve met one person with autism, you’ve met one person with autism. But there are common features that are part of the diagnosis, that are common across all the individuals who are diagnosed with autism.

Saintsing: First of all, people with autism tend to have trouble with social interactions. In a conversation, most people use a lot of nonverbal cues to convey meaning. They make facial expressions and manual gestures, and they change the tone of their voice and the cadence of the conversation. Certain areas of the brain help neurotypical people naturally pick up on these cues, but these brain areas work differently in autistic people.

Dawson: That makes it just difficult to understand and interpret those behaviors. Most people on the spectrum do learn to do that, but they learn to do it in a much more intentional way. 

Saintsing: At the same time, people with autism often experience sensory information differently. For some, this means that they’re extra sensitive to certain stimuli, like light, sound, even touch. 

Dawson: Like putting on a shirt might feel uncomfortable, or having a tag on the inside of your shirt. Things like that.

Saintsing: Dawson places these sensory features of autism in the same category as other nonsocial aspects of the condition. People with autism often engage in repetitive behaviors and maintain a restricted range of interests. While these might sound like limitations, Dawson said that the different ways that people with autism experience the world can give them unique perspectives and extraordinary talents. 

Dawson: And so, it’s important to understand autism as, yes, coming with challenges, but also bringing in differences in the way of experiencing the world that make our world more interesting and more diverse. A focus now is to try to make accommodations so that a person on the spectrum can experience the world in a way that is a better fit for them, whether it’s the work environment or the school environment, so that they can develop their natural talents and experience in the world in a way that gives them confidence and joy.

Saintsing: That’s where detection and diagnosis become important. Once a person knows that they are autistic, they can understand themselves better and start accessing the resources that are available to them. According to La Trobe University psychologist Josephine Barbaro, the earlier detection occurs, the better an experience an autistic individual will have.

Josephine Barbaro: In the past and still currently, a lot of professionals say, ‘Don’t label children because we’re pigeonholing them.’ But what autistic people tell me is that they all still get labeled in the schoolyard, but it’ll be these really harmful labels like freak or weirdo, instead of autistic. It’s not just about providing supports and services for the child, but it’s also about helping the family to understand autism and how to communicate that to their child so that they feel a sense of pride in their identity.

DeMarco: How do doctors currently diagnose kids with autism? 

Saintsing: Pediatricians commonly screen children for autism between 18 and 24 months. Many doctors in the United States have a child’s parent answer a questionnaire called the Modified Checklist for Autism in Toddlers, Revised, or M-CHAT-R. The questions prompt parents to reflect on their observations of their child’s behavior. If a child scores high, then he or she may be eligible for an autism diagnostic evaluation.

Dawson: As clinicians we sit down with the child and play with them and probe different skills. So, we’re very interested in how the child is using gaze. Typically, when we interact with people, and we’re playing with things, we’re constantly alternating our gaze between the toy and the person, because we’re interested in what the person is looking at and whether they’re enjoying the interaction. Often, for a child with, with a diagnosis of autism, they’re not going to be alternating gaze in that way. And then we’re also looking at how they play with toys and their motor behaviors. We point at things and see whether they’re following it. We look to see whether they’re showing us things. We’ll look at whether the child is making eye contact, call the child’s name and see whether they turn to listen. In a very standardized way, during this play interaction, the psychologist or other clinician is writing down all the observations, and then, if they have enough of these features, they’ll qualify for a diagnosis of autism. So, it does take training, and it takes a skill of observation.

Saintsing: If there aren’t enough trained professionals available, a child may not be able to receive a timely diagnosis. On top of that, the diagnostic process by questionnaires like M-CHAT-R may screen out many children who are actually autistic. Barbaro found that when she started in the field in 2005, many screening tools were highly inaccurate and failed to identify the majority of autistic kids.

Barbaro: Screening tools were either developed for just one time point in development, so either 12 months or 18 months or 24 months, and not through a developmental surveillance lens whereby you’re continuously monitoring that child for differences in their social communication. And also, if screening tools were used across time, they tended to be the same screening tool and didn’t change the types of behaviors that they looked at. And then the third reason was this heavy reliance on parental report. Now parents are absolutely the expert when it comes to their child, but not necessarily expert when it comes to social communication development or autistic development.

Saintsing: Regardless of how much training a person has, reliance on human observation introduces subjectivity into the process. To make it more objective, Dawson and her team designed an app called SenseToKnow, which uses machine learning to diagnose kids with autism. Their app relies on a technique called computer vision.

Dawson: This is a way in which a computer maps all the different movements of the face and body. This takes us out of the realm of just clinical judgement into these very accurate and high-resolution measurements of things like eyebrow raises and blink rate and the degree to which the muscles around your mouth are moving, whether you’re moving your head back and forth. All of these things can be measured very accurately with a computer.

Saintsing: The app has access to a phone or tablet’s camera, and it records footage of the child as they interact with it.

Dawson: There’s a series of very brief movies — under a minute — that are shown on the phone or the tablet. So, these movies, for example, might be a woman who’s playing with a toy. In one case, she would be blowing a pinwheel. 

And we design the movie so that the pinwheel’s on one side, and the woman is on the other side. We’ve separated the social elements and the nonsocial elements in the movie in a very distinct way. Now when a neurotypical baby or toddler watches that movie, their eye gaze is going to be going back and forth. As you can imagine, they're interested in watching that spin wheel, but they're also interested in watching the woman who's smiling and blowing, whereas a child who has a high likelihood of a diagnosis of autism is going to be mostly focused on that pinwheel and not really paying attention to the social elements. And one of the behaviors that you can map using computer vision is gaze. Eye tracking is one of our key measures, and we can do this without any kind of special equipment, just basically using an iPhone or an iPad. 

Saintsing: The app also records a child’s blink rate.

Dawson: So, it turns out that when we're really interested in something, we actually suppress our blink rate. And you can think about this as the brain unconsciously is saying, ‘I want to take in as much visual information as possible.’ So, when you're talking with someone, and they get to a really interesting part of the conversation, you're automatically suppressing your blink rate. Well, that's another thing that can be measured readily with computer vision. We have movies that are primarily social in content, say, a woman that is saying nursery rhymes. 

Woman: The itsy-bitsy spider went up the waterspout.

Dawson: And then another movie that is nonsocial, different kinds of dynamic toys. 

The neurotypical children will suppress their blink rate more when they're watching the woman saying the nursery rhymes, whereas a child who has a diagnosis of autism will be suppressing their blink rate when they're watching those dynamic toys. And there's many other measures. We were able to derive about 23 different phenotypes, and then those are combined using machine learning into a single algorithm that then predicts the likelihood that this child will have a diagnosis of autism.

Saintsing: Dawson and her team have administered the app to thousands of toddlers so far.

Dawson: One of the things we did from the beginning is to not work in a lab setting, but actually work out in the real world. Our first studies were all done in primary care settings, working with pediatricians where the app was delivered actually in the exam room. So that was really important because often scientists will develop something in the lab, and then when they try to use it in the real world, there’s a lot of modifications that need to be made. But we worked with pediatricians right from the beginning and administered this when a child came in for their well child visit.

Saintsing: Dawson’s team administered the SenseToKnow app in parallel to the standard M-CHAT-R protocol and monitored children until they were four years old to confirm whether they received an official diagnosis of autism or not. Then, they trained a series of algorithms to classify children based on footage that the app had recorded. Statistical analyses revealed that one algorithm in particular was adept at distinguishing between autistic children and children without autism, and now Dawson and her team are building on their initial results. They’ve just completed a study to determine the accuracy of the app when parents administer it in their own homes. Barbaro is eager to see what the future holds for SenseToKnow and machine learning-based detection of autism in general, but she was a little concerned that the children Dawson’s team sampled in their initial study weren’t entirely representative of the general population.

Barbaro: Over 10 percent of the children in this sample were autistic. Now we know that the prevalence of autism is around three percent and likely to probably be upwards of five percent. So, we know that 10 percent is a huge overrepresentation, which tells me that the type of person that put their hand up for the study or who agreed to participate in this study was more likely to have an autistic kid. So, they either already had concerns, or they were primed to take part because they were maybe thinking about it.

Saintsing: A major focus of Barbaro’s career has been implementing community-wide surveillance for autism. She also tries to compare the effectiveness of her tools in distinguishing between autism and other developmental language delays.

Barbaro: Because when you're comparing autistic to neurotypical, you will find differences across all behaviors. But when you're comparing to developmental delay, that's much more helpful for professionals, because often they know when a child is not developing typically, and they need to know, ‘Okay, is it autism? Or is it more developmental language delay?’ And same with this app here, with the overall results, they were comparing typical development to autistic development. But when you look at the table, when it actually breaks it down, and it compares autism to developmental language delay the results are not as robust.

Dawson: Now our vision of this is still that the data would come back to a pediatrician or other provider so that that person can provide the information to the parent and guide them into next steps. So, we’re not at the point yet where this is direct-to-consumer or parents can just download this to find out, you know, ‘Does my child possibly have a diagnosis of autism?’

Saintsing: Researchers hope that apps like SenseToKnow point to a future where it’s easier and more comfortable for parents to ask questions about their child’s development and for providers to set families up to succeed. 

Barbaro: Let’s empower our professionals. Let’s empower our parents. Normalize developmental surveillance for autism and other developmental conditions. And that really takes away the stigma from autism.

DeMarco: That’s it for this episode of DDN Dialogues. Thank you so much, Andrew, for bringing us this great story. I’d also like to thank Geraldine Dawson and Josephine Barbaro for speaking with Andrew, and thanks to all of you for listening! Until next time, I’m your host Stephanie DeMarco.

This episode of DDN Dialogues was reported, written, and produced by Andrew Saintsing with additional production support by me. To never miss an episode, subscribe to DDN Dialogues wherever you get your podcasts. And if you like the show, please rate us five stars and leave a review on your favorite podcasting platform. If you’d like to get in touch, you can send me an email at sdemarco@drugdiscoverynews.com.

And I just have to say, it’s so exciting to see a smart phone app that can really improve the lives of kids as well as their families.

Audio credits:

At 4:41-4:43, 01700 baby spill wooden blocks.wav by Robinhood76 at  https://freesound.org/s/95830/

License: Attribution NonCommercial 4.0