NEWS & PERSPECTIVE

Early intervention for autism now possible with automatic retinal image analysis

28 Feb 2021

The prevalence of Autism Spectrum Disorder (ASD) has been increasing tremendously over the last decades.1 While the symptoms of ASD could emerge as early as 12 months of age and a solid diagnosis can be established at the age of 3, most children with ASD in Hong Kong are only referred for assessment after 6 years old which could have delayed the prime time for intervention.2 As retinal changes are suggested to be associated with autism development, researchers from the Chinese University of Hong Kong have recently developed an ASD screening tool utilizing the automatic retinal image analysis (ARIA) technology to help facilitate an earlier ASD diagnosis and intervention.3

ASD refers to a group of early-onset neurodevelopmental disorders where patients lack social, discoursing and nonverbal interactions such as facial expression, eye contact and body gestures.4 In the United States, the prevalence of ASD among children of 8 years old had increased from 16.8 per 1,000 patients in 2014 to 18.5 in 2016, representing an approximately 10% rise.5 In view of the growing prevalence, early and prompt diagnosis and intervention have become more critical for this disorder. At present, there are questionnaires for screening autism but are varied in sensitivity and predictive value.3 Moreover, Hong Kong currently lacks a standard screening program for ASD in children, and a waiting time of 2 years for diagnosis further impedes the early identification of ASD cases.3

To help address the issues encountered in the screening of ASD in Hong Kong, researchers from The Faculty of Medicine at The Chinese University of Hong Kong have developed the cloud-based ARIA technology that utilizes a machine-learning algorithm to optimize information of the retina for the development of an ASD classification model.3 Using ARIA, the risk of developing ASD can be assessed based on changes in retinal features, such as a reduction in retinal nerve fiber layer thickness which is correlated with autism pathophysiology.3,6

With reference to previous experience of using similar technologies to identify certain cerebral disorders such as stroke, the researchers have conducted a study to explore the accuracy of ARIA in screening the ASD.3 Initially, 46 ASD patients from 3 special needs schools were recruited for the study and 24 healthy individuals were recruited as controls.3 Among the study participants, 23 ASD-control pairs with matched age and gender were identified and included for primary analysis.3 The subjects then had their retinal images captured by a nonmydriatic fundus camera, which were further analyzed by ARIA to compare between the ASD and the control group.3

The results revealed that the sensitivity and specificity of this ARIA classification model were 95.7% (95% CI: 76.0%-99.8%) and 91.3% (95% CI: 70.5%-98.5%), respectively, with an area under receiver operating characteristic (ROC) curve of 0.974 (95% CI: 0.934-1.000).3 When the subjects were stratified by sex, the sensitivity and specificity for male participants could reach 97.2% (95% CI: 83.8%-99.9%) and 100% (95% CI: 77.1%-100%), respectively, while female participants could only achieve a sensitivity of 90.0% (95% CI: 54.1%-99.5%) and a significantly lower specificity of 71.4% (95% CI: 30.2%-94.9%), suggesting a higher chance of missing female cases.3 Although the relationship between retinal changes and autism was not a primary focus of this study, the results also showed that ASD patients did have more prominent retinal characteristics including larger optic disc diameter and larger optic cup diameter.3

This study highlights the applicability of utilizing the machine-learning algorithm, ARIA, to objectively classify the risk of ASD based on retinal images.3 “The use of retinal image analysis is non-invasive, fully automatic and relatively convenient. This technique provides an objective screening method that can be implemented in a community setting and provide an efficient tool to assess the risk before clinical and behavioral assessment.” summarized Prof. Benny Chung Ying Zee, the primary investigator of the study. The researchers believed that the employment of this highly sensitive and specific screening tool for ASD can save the need for awaiting time-consuming routine referrals and assessments to help achieve early intervention, thereby making a positive impact for the ASD sufferers in the long run.3

Get access to our exclusive articles.