2023-12-20 06:18:56
[메디칼업저버 신형주 기자] An AI model that can predict autism spectrum disorder from fundus examination retinal photographs has been developed.
The research team of Yonsei University Severance Hospital’s pediatric psychiatry department professors Cheon Geun-ah and Choi Hang-nyeong, ophthalmology professor Kang Hyeon-gu, medical school biomedical system information department professor Park Yu-rang, student Kim Jae-han, and researcher Hong Jae-seong uses AI to screen for autism spectrum disorder and predict the severity of symptoms using fundus examination retinal photographs. It was announced on the 20th that a model had been developed.
The results of this study were published in the latest issue of the international academic journal ‘JAMA Network Open (IF 13.8)’.
Early diagnosis of autism spectrum disorder is important. However, the number of cases of delayed diagnosis is increasing due to limitations in screening tests and lack of social resources.
In particular, early diagnosis and treatment are the most important factors in a positive prognosis, so early diagnosis and medical support tailored to individual characteristics are essential.
The retina develops from the same tissue as the brain, so its composition and structure of nerve cells have characteristics similar to those of the brain.
Recently, research results have been published showing that changes in retinal structure are observed not only in autism spectrum disorder but also in various central nervous system diseases such as attention deficit hyperactivity disorder (ADHD), Parkinson’s disease, Alzheimer’s disease, and schizophrenia.
From April to October 2022, the research team collected 945 fundus retinal photographs of 479 patients with autism spectrum disorder who visited the Severance Hospital Pediatric Psychiatry Department and 945 retinal photographs of normal controls who visited the Severance Hospital Ophthalmology Department.
The collected fundus retinal photograph data was trained on an artificial intelligence model to build an ‘AI model to distinguish between autism spectrum disorder and normal controls’ and an ‘AI model to distinguish between severe autism spectrum disorder and mild to severe autism spectrum disorder’.
The research team analyzed the AI model’s predictive performance using four indicators: receiver operating characteristic curve (AUROC), sensitivity, specificity, and accuracy.
AUROC, meaning ‘area under the ROC curve’, is a statistical technique that indicates the diagnostic accuracy of a specific test tool for diagnosing a certain disease, and is mainly used as a performance evaluation index for AI models.
Typically, the closer it is to 1, the better the performance, and if it is 0.8 or higher, it is evaluated as a high-performance model.
As a result of the analysis, the AI model for screening autism spectrum disorder showed high prediction results with AUROC of 1 (100%), sensitivity of 1 (100%), specificity of 1 (100%), and accuracy of 1 (100%).
The severity prediction AI model showed performance of AUROC of 0.74 (74%), sensitivity of 0.58 (58%), specificity of 0.74 (74%), and accuracy of 0.66 (66%).
In particular, it was revealed that the optic disc area is the most important in screening for autism spectrum disorder.
Professor Cheon Geun-ah, the head of the research, said, “Through this study, we confirmed the possibility that fundus examination retinal photographs can be used as one of the biomarkers to predict the presence and severity of autism spectrum disorder.” He added, “The fundus examination takes less than 5 minutes. It is a useful test in terms of ease and speed. “We are especially grateful to the autism spectrum disorder patients and their parents who participated in the study, and we expect that the results of this study will be helpful in establishing a diagnosis and prognosis prediction system for autism spectrum disorder.”
Meanwhile, this study was conducted with support from the Korea Intelligence and Information Society Promotion Agency.
1703056393
#Developing #model #predict #autism #spectrum #disorder #fundus #examination #photos
