Identifying ADHD boys by very-low frequency prefrontal fNIRS fluctuations during a rhythmic mental arithmetic task

Computer-aided diagnosis of attention-deficit/hyperactivity disorder (ADHD) aims to provide useful adjunctive indicators to support more accurate and cost-effective clinical decisions. Deep- and machine-learning (ML) techniques are increasingly used to identify neuroimaging-based features for objective assessment of ADHD. Despite promising results in diagnostic prediction, substantial barriers still hamper the translation of the research into daily clinic. Few studies have focused on functional near-infrared spectroscopy (fNIRS) data to discriminate ADHD condition at the individual level. This work aims to develop an fNIRS-based methodological approach for effective identification of ADHD boys via technically feasible and explainable methods. fNIRS signals recorded from superficial and deep tissue layers of the forehead were collected from 15 clinically referred ADHD boys (average age 11.9 years) and 15 non-ADHD controls during the execution of a rhythmic mental arithmetic task. Synchronization measures in the time-frequency plane were computed to find frequency-specific oscillatory patterns maximally representative of the ADHD or control group. Time series distance-based features were fed into four popular ML linear models (support vector machine, logistic regression (LR), discriminant analysis and naïve Bayes) for binary classification. A 'sequential forward floating selection' wrapper algorithm was adapted to pick out the most discriminative features. Classifiers performance was evaluated through five-fold and leave-one-out cross-validation (CV) and statistical significance by non-parametric resampling procedures. LR and linear discriminant analysis achieved accuracy, sensitivity and specificity scores of near 100% (