A neurofeedback-guided EEG and BCI framework for personalized attention rehabilitation in ADHD

The integration of game-based cognitive training with electroencephalography (EEG)-based brain-computer interaction (BCI) has demonstrated potential for enhancing attention among individuals with attention-deficit hyperactivity disorder (ADHD). However, existing systems often lack adaptive difficulty regulation and rely solely on single-modal assessments, thereby limiting personalization and sustained engagement. This study developed and assessed an adaptive, multi-task EEG-BCI training system that combines real-time neurofeedback with machine learning-driven customization to bolster attentional capabilities. Fifty participants (25 with ADHD and 25 controls) completed attention-enhancement sessions utilizing SkiSport, a Unity-based skiing game that adjusts difficulty levels according to EEG-derived attention metrics obtained from the NeuroSky TGAM sensor. Support Vector Regression, XGBoost, and Multi-Layer Perceptron models were trained on behavioral and EEG data to predict optimal difficulty parameters. Attention and behavioural metrics were compared before and after personalisation. The findings indicated that EEG attention scores increased by an average of 15% (7.85% in controls, 21.5% in ADHD participants). The adaptive multi-task games yielded an additional 10% increase following personalization. Behavioral indices on reaction accuracy, game score, and completion time showed an overall improvement of 19%. XGBoost achieved the highest predictive accuracy on a held-out test set (R² value of 0.9826, RMSE of 0.8560, and MAE of 0.6417) for within-subject, window-level attention prediction. The proposed EEG-BCI game facilitated short-term enhancements in attention-related metrics among individuals with ADHD. The incorporation of machine learning-driven personalization into serious games offers a scalable, non-pharmacological strategy for short-term cognitive training and attentional modulation.