Artificial Intelligence May Help Diagnose Autism in Children
Artificial intelligence could help experts identify young children who may have autism, researchers say, after developing a screening system called AutMedAI that they say has nearly 80% accuracy in identifying children who may have the condition.
The researchers say their model, which is based on a type of artificial intelligence called machine learning , could bring many benefits in the field of autism spectrum diagnosis.
“Using the AI model, we can use the available information and identify individuals who are at high risk of developing autism early so they can get early help,” said study co-author Dr. Kristiina Tammimies from the Karolinska Institute in Sweden.
But she added: “I want to stress that the model cannot fully diagnose autism, because that has to be done using standard clinical methods.”
Study details:
In an article written by the research team in the journal Jama Network Open, Kristiina Tammimies and her co-authors report how they took advantage of data collected in a US research initiative called Spark that includes information on 15,330 children with autism and 15,330 children without the condition.
The team describes how they zeroed in on 28 traits that could be easily detected in children with autism before they were 24 months old, based on information reported by parents on medical questionnaires. The researchers then developed machine learning models that looked for different patterns in these traits between children with and without autism.
After using the data to build, tune, and test four different models, the team selected the AutMedAI model that performed best. They then tested the model on another dataset of 11,936 participants whose data included the same traits found in people with autism. The model indicated that 10,476 of those participants had autism.
The results revealed that the model correctly identified 78.9% of all participants with and without autism spectrum disorder. Its accuracy was 78.5% in identifying children with autism aged 2 years and up, 84.2% in those aged 2 to 4 years, and 79.2% in those aged 4 to 10 years.
Another test of the same model using another data set of 2,854 individuals with autism revealed that the model correctly identified 68% of those with the condition.
“This dataset was another research cohort with families with only one child with autism and some parameters were missing, so the model performed a little less well, which indicates that we need to improve further,” said Kristiina Tammimies .
The researchers said that the most important criteria that the model relies on to predict the possibility of a child having autism include: problems eating, age at first forming long sentences, age at first being able to use the toilet, and age at first smile.
The research team conducted an additional analysis, comparing participants who were correctly identified as having autism with those who were incorrectly identified as not having autism, and the analysis indicated that the model tended to diagnose autism in individuals with the most severe symptoms.
However, some experts urged caution, noting that the model's ability to correctly identify non-autistic people was only 80%, meaning that 20% were incorrectly classified as having autism.
Conclusion:
This system opens the way for further research and development in the field of early autism spectrum diagnosis using artificial intelligence.
This means that in the future we may see more innovations in this field, which will help in providing appropriate assistance to people with autism early on, which will help in controlling the symptoms they may suffer from and reducing their severity.
This isn’t the first time researchers have tried to use AI to help diagnose autism. Among other studies, scientists have previously used AI with retinal scans of children to help identify children who might have autism.
This was in new research from the University of South Australia and Flinders University, which showed that artificial intelligence could provide a faster and more accurate way to diagnose autism spectrum disorder in children using a light stimulus that is shone on the eye.
The researchers then measured the response of the retina to the light stimulus, to identify certain features that could help identify people who may have autism spectrum disorder.
The researchers found that the retina generates a different response in children with autism spectrum disorder compared to those without.