Identifying neuroimaging biomarkers of attention-deficit hyperactivity disorder (ADHD) from cortical hemodynamic responses during Go/NoGo task using machine learning approaches.

BACKGROUND

Robust and reliable biomarkers for the objective diagnosis of attention deficit hyperactivity disorder (ADHD) are lacking. Here we aimed to detect the cortical hemodynamic properties of ADHD children during Go/NoGo task based on functional near-infrared spectroscopy (fNIRS) technology to identify neurobiological features for objective diagnosis.

METHODS

Ninety-seven children with ADHD and eighty-four children with healthy controls (HC) were recruited in this study. We calculated the difference between the peak oxyhemoglobin (oxy-Hb) concentration in Go/NoGo block and the baseline (Go block) of different regions of interest (ROI), the ROI hemodynamic responses in Go/NoGo block under general linear model, and functional connectivity (FC) between ROI. Then, we extracted important fNIRS features for distinguish ADHD and HC, and used four machine learning models combined with cross-validation to detect the effect of fNIRS on the classification and diagnosis of ADHD.

RESULTS

Children with ADHD exhibited abnormal activation in the right inferior frontal gyrus (IFG) and left precentral gyrus, along with altered connectivity patterns in frontoparietal regions (p < 0.05). Among the classifiers, the Random Forest model achieved the best performance, with an average AUC of 0. 91, accuracy of 0.892, sensitivity of 0.95, and specificity of 0.824. The NoGo error rates, FC between the right superior frontal gyrus (SFG) and left SFG contributed the most in the model.

CONCLUSION

The abnormal frontoparietal hemodynamic response in children with ADHD may serve as objective, quantifiable biomarkers for assessing executive dysfunction and facilitating early screening of ADHD.