Validation of BGaze method supporting ADHD diagnosis

Abstract

Introduction

Eye movement recordings can provide information about higher-level processing of visual information. Recent evidence shows a novel role for eye vergence in orienting attention (Solé Puig et al., 2013). Based on such eye tracking data, the BGaze method (Braingaze; Spain) detects visual attention. The outcomes of the BGaze method have been applied to classify ADHD patients from healthy controls.

Aim

In this study, we validated the BGaze method.

Objective

We therefore recorded eye movements in children while performing a visual detection task.

Methods

We evaluated the BGaze method using 4 types of supervised machine learning algorithms. In total, 138 different trained models were tested. Nineteen ADHD diagnosed patients (children 7–14 years of age) and 19 healthy age matched controls were used to build the 138 models. We performed 30 times repeated random sub-sampling validation. In each repeated random split, training set consisted of 80% of the data and test set of the remaining 20%. Finally, all the 138 models were tested with a validation set consisting of 232 children, including 22 ADHD patients.

Results

Across all the 138 models, BGaze method showed an average accuracy of 90.84% (minimum 86.21%; maximum, 95.26%) and an average AUC of 0.95 (minimum 0.90; maximum, 0.97). Best models gave accuracies of 92%, AUCs of 0.96 and FN and FP rates of 4.3% and 7.5%, respectively. Mean scores during the training-testing phase averaged 99.63%.

Conclusion

The BGaze method is robust, accurate, and can provide an objective tool supporting the clinical diagnosis of ADHD.

Disclosure of interest

The authors have not supplied their declaration of competing interest.