In a revolutionary breakthrough, researchers have harnessed the power of deep learning AI to achieve a 100% accurate diagnosis of autism in children by analyzing retinal images. The study, conducted by experts from Yonsei University College of Medicine in South Korea, highlights the potential of artificial intelligence as a reliable and objective screening tool for early autism detection, particularly in regions facing a scarcity of specialist child psychiatrists. At the core of the research is the examination of the retina, where the optic nerve connects at the optic disc, providing a non-invasive gateway to vital brain-related information. This approach capitalizes on the ease of access to the retina, making it a feasible and efficient method for obtaining crucial insights into neurological conditions. The study involved 958 participants, with an average age of 7.8 years, whose retinas were photographed, resulting in a total of 1,890 images. Half of the participants had been previously diagnosed with autism spectrum disorder (ASD), while the other half served as age- and sex-matched controls. The severity of ASD symptoms was assessed using standardized measures, including the Autism Diagnostic Observation Schedule – Second Edition (ADOS-2) and the Social Responsiveness Scale – Second Edition (SRS-2). A sophisticated convolutional neural network, a deep learning algorithm, was trained using 85% of the retinal images and symptom severity scores to construct models for ASD screening and symptom severity assessment. The remaining 15% of images were reserved for testing. The AI exhibited remarkable performance in identifying children with ASD in the test set, achieving a mean area under the receiver operating characteristic (AUROC) curve of 1.00, indicating a perfect accuracy rate. Notably, even when 95% of less crucial areas of the image were removed, there was no significant decrease in the AUROC, emphasizing the robustness of the AI model. The researchers emphasized the potential of retinal alterations as biomarkers for ASD, stating, “Our models had promising performance in differentiating between ASD and TD [children with typical development] using retinal photographs.” The study further revealed that the optic disc, comprising only 10% of the image, played a crucial role in distinguishing ASD from typical development. While the mean AUROC value for symptom severity was 0.74, falling within the ‘acceptable’ range, the researchers highlighted the significance of the findings, suggesting that retinal photographs could provide additional insights into symptom severity. They noted the distinction between assessment tools, stating that the AI model demonstrated better feasibility in reflecting severity based on ADOS-2 scores compared to SRS-2 scores. The study included participants as young as four, paving the way for the potential use of the AI-based model as an objective screening tool from that age onward. However, the researchers acknowledged the need for further research to validate its accuracy in participants younger than four, considering the ongoing growth of the newborn retina. In conclusion, the researchers emphasized the pivotal role of their AI-based model in addressing urgent challenges, particularly the limited accessibility to specialized child psychiatry assessments. They expressed optimism about the future potential of their study in establishing generalizability and paving the way for objective screening tools for ASD. By Impact Lab
This content was originally published here.