Screening children for autism using retinal images analysis


  • Heather Mason
  • Univadis Medical News
Der Zugang zum gesamten Inhalt dieser Seite ist nur Angehörigen medizinischer Fachkreise vorbehalten. Der Zugang zum gesamten Inhalt dieser Seite ist nur Angehörigen medizinischer Fachkreise vorbehalten.

A study reporting on the use of a machine learning approach to analyse retinal images in children revealed that this could be a useful risk assessment tool for autism spectrum disorder (ASD) screening, according to an article published in  Lancet'EClinicalMedicine.

The analysis included 46 ASD participants and 24 non-ASD controls. Among them, 23 age-gender matched ASD and non-ASD participant-pairs were constructed for primary analysis. Automatic retinal image analysis (ARIA) methodology applying machine-learning technology was used to optimise information from the retina to develop a classification model for ASD. To obtain validation of the model, the researchers used a 10-fold cross-validation approach to assess its validity.

The findings show significant retinal differences between children with ASD and their age- and gender-matched control. In particular, ASD subjects have significantly larger optic disc diameter and larger optic cup diameter. By employing machine learning and complex analysis methods, ASD could be detected with much higher accuracy than simply relying on a set of retinal characteristics.

The authors say that this method could be used as a community-based risk assessment tool instead of a diagnostic tool to offer parents the opportunity to provide early intervention to children with a high risk of ASD.