Home Videos May Fast-Forward Autism Diagnosis

Deborah Brauser

April 16, 2014

Short home videos of young children, such as those commonly posted online by proud families, may help in the diagnosis of autism spectrum disorders (ASDs), potentially paving the way for faster diagnosis and intervention, new research suggests.

A proof-of-concept study of 100 public videos posted to YouTube of children between the ages of 1 and 15 years showed that for trained classifiers, the videos had 97% accuracy, 94% sensitivity, and 100% specificity in distinguishing between children who did have an ASD and those who did not.

"These rates were incredibly reassuring, although also surprising," senior author Dennis P. Wall, PhD, associate professor of pediatrics in the Division of Systems Medicine and associate professor of psychiatry by courtesy at Stanford University School of Medicine in California, told Medscape Medical News.

"These numbers are high, which is exciting. But it's also encouraging because it provides us with some great support to move forward," he added.

The study was published online April 16 in PLoS One.

Time-Consuming Process

The investigators note that these data suggest that moving the process of detecting and monitoring autism out of the clinical setting may eventually be feasible ― especially because video tracking has the potential to improve the speed and availability of a diagnosis.

Dr. Dennis Wall

"These findings give us a great deal of hope that we will be able to make diagnosis and follow-up much more widely available, not only in the US but across the globe, so that children get recognized as early as possible," said Dr. Wall in a release.

"If they have an Internet connection and the capability of uploading a video, we should be able to score it and provide immediate understanding of risk that they can then communicate to their clinical practitioners."

The most recent statistics from the Centers for Disease Control and Prevention show that 1 in 88 children in the United States are diagnosed with autism, yet, the researchers note, getting a diagnosis is a cumbersome and time-consuming process.

Currently, ASDs are diagnosed through lengthy observations of a child's possible lack of eye contact, poor language skills, and repetitive behaviors. These diagnoses occur, on average, when a child is around the age of 4 years.

However, previous studies have shown that behavioral interventions work best when initiated by age 2 or 3 years ― the time when parents begin to suspect there is a problem.

In addition, families who want a formal evaluation often have to travel long distances or wait for a long period.

"Today's standard of care is quite substantial and lengthy. They're diagnostically useful, but they take hours to complete and must be administered by clinical professionals in medical facilities," said Dr. Wall.

"I wanted to look at this and see if we could apply methods from my background. My skill set has been developed in the areas of systems biology and signal detection. And we wanted to see if we could use all of this to reduce the complex diagnostic process with the same level of accuracy," he explained.

For this study, the researchers chose 100 videos posted publicly on the YouTube channel. All were 5 minutes or less in length and featured children engaged in continuous activity.

Of these, 45 were tagged by the uploaders with the phrases "autism," "ASD," "Asperger's," or "hand-flapping/stimming" and were classified as ASD videos. The other 55 were classified as non-ASD videos and showed age-matched kids without the disorder.

Dr. Wall noted that although the people involved in the videos were contacted and informed about the study, they did not comprehensively confirm diagnoses.

The investigators provided very brief training to 4 nonclinical research assistants to be able to score autistic-type behaviors in the children showcased within the home videos using questions from the Autism Diagnostic Observation Schedule (ADOS).

Through this scoring system, the raters assessed a child's eye contact, whether social cues were picked up from other people in the video, whether there were any repetitive behaviors demonstrated, and whether toys were played with appropriately.

High Sensitivity, Specificity

Results showed that children in the videos were correctly classified by the raters as having or not having ASD 96.8% of the time.

In addition, sensitivity and specificity rates were 94.1% and 100%, respectively.

"The inter-rater correlation for the behavioral domains on the ADOS was 0.88, and the diagnoses matched a trained clinician in all but 3 of 22 randomly selected video cases," report the researchers.

"In each of the 3 cases, the classification provided by the clinician did not agree with the self-reported diagnosis of ASD," they add.

Overall, "our new paper supports the hypothesis that we can detect autism quickly in very short home videos with high accuracy," said Dr. Wall.

He added that the results may also have future clinical implications. Although the video-based evaluations should not completely replace standard diagnostic methods from a trained clinician, they could be used to augment the process.

"For instance, we could use this system for clinical triage, as a way to channel traffic so that children can get the kind of attention they need as early as possible," explained Dr. Wall.

He noted that videos could be especially helpful for children who clearly have ASD so that they can get started on therapy sooner. This would "free up" clinicians for longer evaluations of the children whose diagnoses are not so clear-cut.

In addition, Dr. Wall noted that videos could also be used to track how well a child with ASD is doing after starting behavioral therapy, or to track development in very young children (such as those around 18 months of age) who are suspected of having the disorder to better help during later assessments.

Videos can also catch behaviors that differ between home and a physician's office.

"Clinical settings are often stark, artificial, and can elicit behaviors that are abnormal. The odds are stacked against the diagnostic professional because the child is in an unknown environment with strangers," said Dr. Wall.

He added that it was also important for the investigators to use raters in this study who were not highly trained.

"It would be great if we could take advantage of a population of individuals who are nonclinically educated to be trained to be video analysts with high accuracy at predicting the presence or absence of autism. This suggests the possibility of a virtual workforce," he said.

"This could also allow us to create a mobile paradigm for rapid, early detection of autism that enables us to drive down the average age from where it is today, at 4 or 4-and-a-half, to something much, much lower."

"Fascinating" Research

"I'm really fascinated by this work; it raises a lot of questions we need to flesh out," Kari Berquist, PhD, a pediatric psychologist at Lucile Packard Children's Hospital in Palo Alto, California, said in a release.

Dr. Berquist, who was not involved in this research, agreed with the investigators that evaluating home videos should not necessarily replace traditional diagnostic approaches. However, she added that it could be a good way for families to start getting the help they need.

"Hopefully it will also prevent the 'watch and wait' approach for parents who are really concerned and feel like they are not being heard," she said.

Dr. Wall reported his involvement with Cognoa, a new company about to launch that is dedicated to refining, testing, and delivering mobile solutions for the detection and monitoring of autism. The other investigators report no relevant financial relationships.

PLoS One. Published online April 16, 2014. Full article


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.