Retinal fundus imaging as biomarker for ADHD using machine learning for screening and visual attention stratification

Area of Science:

• Ophthalmology
• Neurodevelopmental Disorders
• Artificial Intelligence in Medicine

Background:

• Attention-deficit/hyperactivity disorder (ADHD) presents diagnostic challenges due to symptom heterogeneity.
• Current ADHD diagnosis relies on clinical assessment, lacking objective biomarkers.
• Executive function (EF) deficits are a core feature of ADHD, varying in severity.

Purpose of the Study:

• To investigate machine learning (ML) analysis of retinal fundus photographs as a noninvasive biomarker for ADHD screening.
• To evaluate the potential of retinal imaging for stratifying executive function (EF) deficits in ADHD.
• To explore the utility of specific retinal features in predicting ADHD and EF subtypes.

Main Methods:

• Retinal fundus photographs from 323 children and adolescents with ADHD and controls were analyzed using the AutoMorph pipeline.
• Four machine learning models were employed for ADHD screening and EF subdomain prediction.
• Shapely additive explanation (SHAP) method was used to interpret model predictions.

Main Results:

• ADHD screening models achieved high diagnostic accuracy, with an area under the receiver operating characteristic curve (AUROC) ranging from 95.5% to 96.9%.
• Stratification of EF deficits showed strong performance for the visual subdomain (AUROC > 85%) but weaker performance for the auditory subdomain.
• Retinal feature analysis identified potential biomarkers for ADHD and visual attention deficits.

Conclusions:

• Retinal fundus photography combined with ML offers a promising noninvasive tool for ADHD screening.
• This approach demonstrates potential for stratifying executive function deficits, particularly in the visual attention domain.
• Further research can validate retinal imaging as a clinical aid for ADHD assessment and management.