NEWS  Science  December 18, 2023  Reading time: 2 Minute(s)

Max (RS editor)

Researchers from Yonsei University College of Medicine in South Korea have successfully harnessed the power of deep learning AI algorithms to diagnose autism spectrum disorder (ASD) in children with unprecedented accuracy. The study not only demonstrates the potential of artificial intelligence as an objective screening tool but also highlights the crucial role of retinal imaging in providing valuable insights into neurodevelopmental disorders.The Retina as a Window into the Brain

At the intersection of the retina and optic nerve lies the optic disc, offering a non-invasive gateway to gather essential brain-related information. This study builds upon recent innovations, such as the non-invasive diagnosis of concussion through retinal imaging, showcasing the potential of leveraging the eye as a diagnostic window into neurological conditions.

Methodology and Findings

The research involved 958 participants, with half diagnosed with ASD and the remaining half serving as age and sex matched controls. Retinal images were obtained and subsequently screened by a convolutional neural network, a deep learning algorithm. The algorithm achieved a remarkable 100% accuracy in distinguishing children with ASD from those with typical development, as indicated by the mean area under the receiver operating characteristic (AUROC) curve.

Furthermore, the study assessed ASD symptom severity using established diagnostic tools. The AI model demonstrated a mean AUROC value of 0.74, classifying symptom severity at a level considered 'acceptable' in the diagnostic realm.

Implications for Early Diagnosis and Access to Care

The researchers emphasize the potential of retinal alterations as biomarkers for ASD, with promising implications for early diagnosis. Given the limitations in access to specialized child psychiatry assessments, particularly in resource-constrained settings, the AI-based model offers a viable and objective screening tool.

Age Considerations and Future Directions

The study included participants as young as four years old, suggesting that the AI model could be employed as a screening tool from that age onward. However, further research is necessary to validate its accuracy for participants younger than four, considering the ongoing growth of the newborn retina.

The successful application of deep learning AI algorithms in screening retinal images for ASD marks a significant stride toward addressing the pressing need for accessible and objective diagnostic tools. While future studies are warranted to establish generalizability, this research lays a foundation for the development of innovative solutions that could revolutionize the early detection and management of autism spectrum disorder.

 COVER IMAGE BY FREEPIK