A Severance-KAIST research team has affirmed that deep learning using brain image big data can predict symptoms and severity of people with autism spectrum disorders (ASD).

ASD is one of the disorders caused by abnormal brain development. ASD patients show defects in social communications, limited range of interests, and repetitive behaviors. The U.S. Centers for Disease Control and Prevention (CDC) said that one in 54 people develop ASD, and the number is increasing every year. The prevalence of ASD is also about 2 percent in Korea.

A joint research team, led by Professor Cheon Keun-ah of Severance Hospital and Lee Sang-wan of Korea Advanced Institute of Science and Technology, has developed a deep learning model to predict symptoms and severity of autism patients.

The diagnosis of the disorder is based on observing children’s behavior and counseling and the Diagnostic and Statistical Manual of Mental Disorders (DSM-5). However, an accurate diagnosis of ASD and prognosis is difficult due to wide individual differences.

The joint research team developed a deep-learning model that can diagnose ASD and predict prognosis with magnetic resonance imaging (MRI) by using MRI big data of 84 ASD patients aged 3 to 11 collected at Severance Hospital and 1,000 autism cases piled up at an international consortium.

The diagram shows how the artificial intelligence model finds the relationship between major brain regions related to the severity of autism through magnetic resonance imaging (MRI) big data of the patients with autism spectrum disorders. (Severance)

Researchers built the model with a spatial transformer network and a three-dimensional convolutional neural network and trained with the MRI big data.

The analysis of the study result showed that the subcortical structures, including the basal ganglia of the brain, are correlated with the severity of autism.

“We have confirmed that there is a difference between the brain images of sub-symptoms and severity of autism,” Severance Hospital Professor Cheon Keun-Ah said. “The significance of this study is the personalized diagnosis and the prediction of prognosis for autistic patients with various clinical phenotypes and severity.”

Professor Lee Sang-wan of the Korea Advanced Institute of Science and Technology said, “We can now provide structurally associated candidates for diagnosing and studying autism in the medical field. The study results will allow doctors and related experts to understand complex disorders with artificial intelligence and diagnose autism.”

The study results were published in the online edition of the Institute of Electrical and Electronics Engineers (IEEE) Access.

Copyright © KBR Unauthorized reproduction, redistribution prohibited