AI Identifies Autism Spectrum Disorder at the Initial Stage

Is it possible to employ artificial intelligence to help in the early diagnosis of autism spectrum disorder? Researchers from the University of Arkansas are attempting to provide an answer to that query. 

Khoa Luu, an assistant professor of computer science and computer engineering, and Han-Seok Seo, an associate professor with joint appointments in food science and the UA System Division of Agriculture, will recognize sensory cues in both neurotypical children and those who are known to be on the spectrum.

Then, to identify signs of autism, machine learning technology will be utilized to examine physiological data and behavioral responses to certain tastes and odors.

ASD is characterized by a variety of behaviors, such as communication problems, issues with social interaction, or repetitive behaviors. As well as exhibiting some atypical eating behaviors, people with ASD are also found to reject some or even many meals, have particular mealtime needs, and engage in non-social eating. Avoiding food can result in poor nutrition, including vitamin and mineral shortages, which is very alarming.

In light of this, the team wants to find sensory clues from food that cause unusual perceptions or actions when ingested. For example, people with ASD are known to react more strongly to the scents of peppermint, lemons, and cloves than people without them, potentially evoking higher sensations of rage, surprise, or disgust.

Seo is an expert in the fields of biometric data, eating behavior, behavioral neuroscience, and sensory research. This initiative is being organized and led by him, and as part of it, sensory clues that can distinguish autistic children from non-autistic children in terms of perception and behavior are being screened for.

Luu is a leading authority on artificial intelligence with expertise in computer vision, machine learning, deep learning, and biometric signal processing. Based on distinct patterns of perception and behavior in reaction to particular test samples, he will create machine learning algorithms for identifying ASD in young people.

Their ultimate objective is to develop an algorithm that performs as well as or better than conventional diagnostic methods for the early detection of autism in children. These methods typically involve trained medical and psychological professionals conducting evaluations, longer assessment times, caregiver-submitted questionnaires, and higher medical costs. They hope to be able to verify a less expensive method to help with an autism diagnosis.

Even while their method is unlikely to be the last word in a diagnosis, it might offer parents a preliminary screening tool, ideally excluding kids who are not prospects for ASD while ensuring that the most likely candidates undergo a more thorough screening procedure.

Seo claimed that after the birth of his daughter and after starting to work with a graduate student named Asmita Singh, who had experience working with autistic pupils, he became intrigued by the idea of employing multi-sensory processing to assess ASD.

Video Credit: University of Arkansas