Abstract
OBJECTIVES: Continuous Performance Tests (CPTs) are widely utilized as objective measures in the assessment of Attention-Deficit/Hyperactivity Disorder (ADHD). The integration of sensor data in smartphones has become increasingly common as a way of monitoring several behavioural indicators of mental health. Machine learning has started being utilized in the field of ADHD to improve diagnosis. This investigation explores (i) the feasibility of using smartphone devices to administer a CPT for ADHD assessment and (ii) whether data from built-in sensors in smartphone devices is useful for predicting a diagnosis. METHODOLOGY: The study uses data from a control group of neurotypical individuals and an ADHD cohort of unmedicated patients. The dataset is divided into a training and test set, and a machine learning model is developed using the training set. The model is trained by dividing features into four groups, Demographic, CPT, Face, and Motion, which are then sequentially added and evaluated on their ability to predict ADHD. RESULTS: A total of 952 neurotypical individuals and 292 unmedicated ADHD patients were part of the study. The best performing model combines all feature groups by a sensitivity of 0.808, specificity of blue and area under the precision-recall curve (PR-AUC) of 0.799, with a considerable performance increase due to the phone sensor features addition. Results did not differ significantly by age group (6-11 and 12-60 years old) and sex. CONCLUSION: The study provides a robust machine-learning model that is based on a large control group together with an ADHD cohort. The experiments demonstrated that ADHD can be assessed with high accuracy using CPTs on smartphones. Integrating face-tracking and motion sensor data with CPT features further enhanced performance, indicating that data from a smartphone device can surpass the accuracy of traditional computer-based ADHD assessments.