According to a study, artificial intelligence can make a diagnosis of autism spectrum disorder.

‘Scientific Reports’ is a journal that published the study.

Due to the intricacy of the condition, diagnosing autism spectrum disorder (ASD) still presents a difficult issue that calls for highly qualified clinicians. Multifactorial neurodevelopmental disorder autism has a wide range of
symptoms.

The study was based on brain imaging data for 500 people, about half of whom (242) had been diagnosed with ASD. Machine learning techniques were applied to the data. “We began developing our methodology by
collecting functional magnetic resonance imaging [fMRI] and electroencephalogram [EEG] data,” said Francisco Rodrigues, last author of the article. He is a professor at the University of São Paulo’s Institute of
Mathematics and Computer Science (ICMC-USP) in São Carlos, Brazil, and his contribution to the research was supported by FAPESP.

“We compared maps of people with and without ASD and found that diagnosis was possible using this methodology,” Rodrigues said.

The researchers fed a machine learning algorithm with these maps. Based on the learned examples, the system was able to determine which brain alterations were associated with ASD with above 95% accuracy.

Much recent research proposes methods for diagnosing ASD based on machine learning but uses a single statistical parameter, ignoring brain network organization, which is the innovation featured by this study, the
article notes. Brain maps or cortical networks show how brain regions are connected. Research on these networks began about 20 years ago and has offered a new vision of neuroscience. “Just as a road with
interruptions alters the traffic in a region, a brain with alterations leads to changes in behavior,” Rodrigues said.

The analysis of fMRI data highlighted changes in certain brain regions associated with cognitive, emotional, learning and memory processes. The cortical networks of ASD patients displayed more segregation, less
distribution of information and less connectivity compared to controls.