Korean researchers have developed a deep learning algorithm to screen for childhood autism with 100% accuracy. It checks retinas for patterns related to this mental illness, facilitating the identification of autism spectrum disorder (ASD) worldwide. Believe it or not, it boasts 100% accuracy, even if you remove parts of its sample data!
Mental health awareness has been expanding worldwide, enabling those needing it to receive proper care. However, many parts of the world lack trained professionals and resources to properly diagnose patients. Fortunately, this artificial intelligence system could assist those with autism.
This article will discuss how Korean researchers created their AI autism detection program. Then, I will cover a similar AI health tool to illustrate this tech healthcare trend.
How does the AI autism tool work?
The Korean researchers started developing their autism detector by taking retinal photographs of 1,890 eyes of 958 participants under 19 years of age. They selected participants from the Severance Hospital, the Department of Child and Adolescent Psychiatry, and Yonsei University College of Medicine in Korea.
They also gathered retinal photographs from people with typical development (TD) matching the autism participants’ ages and sexes. In other words, they took eye photos of those without autism as a separate control group.
Then, the team created a convolutional neural network. It is a deep learning algorithm that trains AI models for ASD to assess and screen symptom severity.
AI training involved 85% of the retinal images and corresponding scores from symptom severity tests. Conversely, they reserved the 15% for testing purposes.
The AI model can accurately identify children with an ASD diagnosis. It yielded a mean area under the receiver operating characteristic (AUROC) curve of 1.00. A “1” means the model predicts autism with 100% accuracy.
You may also like: ChatGPT detector spots AI-generated papers
Surprisingly, the team claims removing 95% of the least important parts of the retinal pictures shows a negligible decline in accuracy. “Our models had promising performance in differentiating between ASD and TD using retinal photographs, implying that retinal alterations in ASD may have potential value as biomarkers,” said the study.
“Interestingly, these models retained a mean AUROC of 1.00 using only 10% of the image containing the optic disc, indicating that this area is crucial for distinguishing ASD from TD,” it added.
The team admits they need more research and development to verify the tool works for most people. Nevertheless, Interesting Engineering says it could address the limited accessibility of specialized child psychiatry assessments.
Other AI health tests
Retinas seem to provide a wealth of medical information because scientists have used it to screen for various illnesses. Aside from detecting autism, experts used AI to detect heart disease risks with eye scans.
The British Journal of Opthalmology created this tool and called it the QUARTZ (QUantitative Analysis of Retinal vessels Topology and siZe). It uses a database of retinal images from 88,052 UK Biobank participants and 7,411 people from a European Prospective Investigation into Cancer (EPIC)-Norfolk study.
You may also like: AI eye test detects brain injuries in minutes
The database classified these photos according to the risks of heart attack and other cardiovascular diseases. QUARTZ compares a patient’s eye scans to these photos so that it can diagnose potential issues. For example, if your eye scan is similar to a stroke patient, it would say you have a high risk for it.
The study applauds its potential to make heart tests more accessible by making them less expensive. Also, it is non-invasive as it does not require a blood test.
Pearse Keane, a researcher in ophthalmology and AI analysis said that QUARTZ has many issues. More importantly, it must tackle the biggest challenge of going from “code to clinic.”
Conclusion
Korean researchers created an AI autism tool that spots this mental condition with 100% accuracy. As a result, it could facilitate diagnosing this illness worldwide.
It could compensate for the lack of resources and trained professionals in other countries. However, they admitted they need more research and development to verify its real-world applications.
Learn more about this AI autism program by reading its JAMA Network webpage. Moreover, check out the latest digital tips and trends at Inquirer Tech.