“My recommendation is below [two years] is too early to start applying psychiatric labels based on a few signifiers like eating behaviour,” she said.
Key Takeaways
Artificial intelligence is showing promise in helping experts identify toddlers who may be autistic, according to recent research from Karolinska Institutet in Sweden.
The study introduces an AI-based screening system that boasts an accuracy of approximately 80% for children under the age of two. This development could enable earlier identification of children at risk for autism, allowing for earlier intervention and support.
Dr. Kristiina Tammimies, a co-author of the study, explained that the AI model leverages available information to identify individuals with an elevated likelihood of autism at an earlier stage.
She further added, “I want to stress that the algorithm cannot diagnose autism as this should [still] be done with gold standard clinical methods.”
“Using [the] AI model, it can be possible to use available information and earlier identify individuals with elevated likelihood for autism so that they can get earlier diagnosis and help.”
The AI model, known as AutMedAI, was developed using data from the SPARK study, a large U.S. research initiative that included information from around 30,000 children—half with an autism diagnosis and half without.
The model focuses on 28 specific measures that can be collected before a child reaches 24 months of age. These measures include parent-reported data on the child’s early behaviors, such as when they first smiled, constructed short sentences, and faced eating difficulties.
So the problem with this is that the algorithm that was used would have been designed by Psychiatry to enable imprisonment for on average 5.5 years in U.K. but some patients are imprisoned and tortured indefinitely. #DefundPunitivePsychiatry
— Dr Humera Shaikh BSc(Hons) MBBS with Merits MRCP (@HumeraShaikh101) August 20, 2024
After developing and testing four different machine learning models, the research team selected the most effective one, which they then tested on a separate dataset of nearly 12,000 participants.
Among these participants, the model correctly identified nearly 80% of those who were later diagnosed with autism. The accuracy varied slightly by age group, with the model being most accurate for children aged two to four years.
Additional testing on another dataset, which included 2,854 autistic individuals, showed that the model correctly identified 68% of these cases.
Good to see predictive technology in this space that relies on a diverse set of signals.
— Hadas Bitran (@hadasbitran) August 20, 2024
Tammimies noted that this particular dataset had certain limitations, such as missing data points, which may have affected the model’s performance.
Despite these challenges, the model demonstrated great potential by identifying autism in children with more pronounced social communication and developmental issues.
The research highlights specific predictors that were particularly major in the model’s analysis, including difficulties with eating, the age at which children began constructing longer sentences, the age at potty training, and the age at which they first smiled.
This is a terrible rate. Over 90% of positive diagnoses would be false
— 🫎 Moose 🥥🌴 (@MooseHB) August 20, 2024
These early-life markers were crucial in helping the AI model differentiate between autistic and non-autistic children.
However, the introduction of this artificial intelligence (AI) tool has not been without its critics. Some experts have expressed concern over the potential for misidentification, as the model correctly identified non-autistic children in only 80% of cases.
This could result in some children being incorrectly flagged as at risk for autism, leading to unnecessary stress for families and possibly unwarranted further testing.
Comment
byu/kevdautie from discussion
inautismpolitics
Moreover, there is caution about the implications of early diagnosis based on such predictive tools. Prof Ginny Russell of the University of Exeter pointed out that it is difficult to determine which toddlers may have developmental impairments and which may “catch up” despite initial delays.
She advised against applying psychiatric labels to children under two years old based on limited behavioral indicators, such as eating habits.
The research team is aware of these limitations and is working on refining the AI model further. They are exploring including genetic data to enhance the model’s predictive accuracy and reliability.
Comment
byu/kevdautie from discussion
inautismpolitics
Their goal is to ensure that the model can be a valuable tool for early screening in clinical settings while stressing that it should complement, not replace, thorough clinical assessments.
This AI tool represents a step forward in the ongoing effort to provide earlier support to children at risk for autism, potentially improving outcomes by enabling earlier intervention.
Comment
byu/kevdautie from discussion
inautismpolitics
However, as with any new technology, careful consideration and further development are necessary to ensure its effectiveness and reliability in real-world applications.
For more news and insights, visit AI News on our website.
Was this article helpful?
YesNo
