A research team led by Professor Sang Wan Lee from the Department of Bio and Brain Engineering and the Center for Neuroscience-inspired Artificial Intelligence collaborated with researchers from the Yonsei University Severance Hospital to develop a deep-learning model that analyzes brain scans to predict the symptoms and severity of autistic spectrum disorders (ASD). This research, which was published in the IEEE Access journal on August 14, introduces a novel improved method to diagnose ASD. 

Due to the complexity of ASD, it has primarily been diagnosed in an ambiguous manner, simply by observing children’s behaviors and development. However, the current study focused on using deep-learning AI to more precisely diagnose ASD. Five different models were used to train and test two sets of MRI scan data from child patients with autism admitted in the Severance Hospital and from the Autism Brain Imaging Data Exchange initiative (ABIDE), an international collection of brain imaging data. Among the models are convolutional networks, recurrent networks, and spatial transformation networks. Together, these models found structural and strategic evidence that specific brain structures below the cortical cortex may serve as possible indicators for the presence of autistic disorders in patients. One example of such a biomarker is the basal ganglia, which is thought to cause repetitive behaviors typical of ASD.

“Our findings provide both structural and strategic pieces of evidence for characterizing ASD”, commented Professor Lee. “We note that revealing such strategies for diagnosis does not only suggest which parts of the brain regions need to be observed, but can also be significantly more economical and time-efficient.” Overall, this research strives not only to improve the classification performance of AI models in detecting ASD, but also attempts to provide reasoning on how exactly the data was classified. In terms of clinical applications, the research team believes that their model can assist in the understanding of other psychiatric disorders like OCD and depression by pinpointing structural abnormalities associated with these disorders within the brain.