ADHD CLASSIFICATION using brainstem-based ocular biomarkers demonstrated promising diagnostic performance in children and adults, according to a new multicentre study.
Current attention deficit hyperactivity disorder (ADHD) assessment often relies on clinical history, rating scales, and performance testing, all of which can be influenced by subjective interpretation.
To address this challenge, investigators developed a machine learning framework that analysed task-evoked pupil diameter and binocular eye movement synchrony during a visual attention task.
Researchers evaluated 439 participants recruited from 14 clinical centres across Spain and the UK, including 324 children and 115 adults.
Researchers classified participants as either healthy, with ADHD, or uncertain.
The study generated two outputs, a diagnostic score and an impulsivity score, by combining information from independent models assessing pupil dynamics and eye movement synchrony.
In the paediatric cohort, selective classification achieved sensitivity and specificity values of 0.79 and 0.82 respectively for the diagnostic score.
The impulsivity score achieved sensitivity and specificity values of 0.74 and 0.70.
By deferring uncertain cases, overall diagnostic reliability improved compared with standard binary classification.
The researchers then evaluated performance in an independent adult cohort to assess cross-population generalisability.
Despite expected differences between paediatric and adult populations, the diagnostic score retained discriminative capability.
Under selective classification, the diagnostic score achieved sensitivity of 0.66 and specificity of 0.86, while the impulsivity score reached sensitivity of 0.68 and specificity of 0.92.
These findings indicated that the ocular biomarkers remained informative beyond the population used for model development.
The approach also maintained practical feasibility, with complete diagnostic predictions generated in approximately 20 seconds per participant.
Explainability analyses provided insight into the physiological features driving model performance.
Attention weighting showed that trials associated with increasing pupil diameter after cue presentation contributed most strongly to classification decisions.
Saliency analyses further identified periods immediately following cue-onset and stimulus-onset as the most influential temporal windows.
The authors projected that combining the biomarker with standard rating scales could theoretically enhance clinical pathways.
Under a screening approach, combined sensitivity was estimated at 95%, while a confirmation strategy was projected to achieve specificity of 96%.
The researchers concluded that objective ocular biomarkers may help support ADHD classification, improve diagnostic consistency, and optimise use of specialist assessment resources.
Reference
Ranathunga C et al. Real-world clinical validation of brainstem-based ocular biomarkers for ADHD classification in children and adults. Sci Rep. 2026;DOI:10.1038/s41598-026-56036-0
Featured image: Mahachoke 4289-6395 on Adobe Stock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
