Researchers in South Korea developed a deep learning algorithm to objectively screen for ASD and assess symptom severity
Artificial intelligence models could be used to diagnose Autism Spectrum Disorder (ASD) and assess the severity of symptoms using retinal photographs, according to a new study.
Previous studies have shown that individuals with ASD exhibit structural changes in the retina, which may reflect underlying brain alterations, including abnormalities in the visual pathway arising from embryonic and anatomical connections. ASD is accompanied by two main categories of symptoms: impaired social communication and restricted and repetitive behaviors or interests. As of 2020, the US Centers for Disease Control and Prevention projected the prevalence of ASD to be 1 in 36 people. This rate is increasing, possibly due to increased awareness among the public, physicians, and researchers.
Now, researchers in South Korea have developed a convolutional neural network—a deep learning algorithm—to objectively screen for ASD and assess symptom severity. First, they trained the model on retinal photographs where the AI told whether the subject had autism or not. The researchers then asked the system to analyze retinal photographs of 958 participants under the age of 19, half of whom had been diagnosed with autism. Participants were recruited from Yonsei University School of Medicine in Korea between April and October 2022.
Encouraging results
The AI tool identified children with autism as well as those without, with 100% accuracy. The researchers noted that their findings show that retinal photographs can reveal additional information about the severity of symptoms and could potentially be used as biomarkers.
While further research is needed to confirm the findings, the researchers note that their study marks an important advance in creating objective tools for diagnosing ASD. These tools could alleviate concerns such as the limited accessibility of specialized child psychiatric assessments due to limited resources.
The study was published in the scientific journal “Jama Network Open”.
