Assessing the Utility of Fronto-Parietal and Cingulo-Opercular Networks in Predicting the Trial Success of Brain-Machine Interfaces for Upper Extremity Stroke Rehabilitation.

Brain-machine interfaces (BMIs) have the potential to improve stroke rehabilitation by actively facilitating sensory-cognitive-motor connections to restore movement. However, individuals with cognitive impairments are often excluded from BMI-based neurorehabilitation due to concerns about impaired cognition, specifically reduced attention and executive control. We propose leveraging the trial-wise dynamics of large-scale cognitive control networks-specifically, the frontoparietal (FPN) and cingulo-opercular (CON) networks-to build neural markers of cognitive control. Using existing BMI datasets, we demonstrate that trial-wise activity within these networks predicts motor task performance, suggesting that cognitive control signals in these networks could serve as adaptive modulations for BMI-based rehabilitation. Our system is able to predict unsuccessful BMI trials at the population level about 84.2% of the time on average, with an overall mean accuracy of 72.2% in a 3-fold cross-validation. Additionally, in a leave-one-subject-out validation, our system achieved 71% specificity on average, with an overall mean accuracy of 68.3%. Notably, model performance varies across subjects, with some individuals showing up to 92% specificity and 100% sensitivity. Unlike previous studies that primarily focus on resting-state data, our findings point toward the untapped potential of incorporating cognitive network state monitoring into BMI systems to optimize online performance through trials. Specifically, we suggest that our pre-trained models can be fine-tuned with subject-specific information to design more targeted rehabilitation programs that enhance motor performance by identifying precise attention and learning tasks to improve the successful response of the network model in patients with significant cognitive impairment.