Clustering of fNIRS-Based Cortical Activation Patterns During Digital Upper Limb Motor Tasks in Individuals With Stroke

The present study aimed to characterize cortical activation and connectivity patterns in individuals post-stroke during digital upper limb motor tasks using functional near-infrared spectroscopy (fNIRS). We enrolled 10 individuals with chronic impairment subsequent to stroke (seven men; mean age, $64.3~\pm ~9.2$ years; mean time since stroke, $108.2~\pm ~60.5$ months). All participants had a unilateral lesion and moderate-to-mild upper limb dysfunction. The fNIRS data were recorded using a 16-source and 16-detector system, with 51 channels sampled at 5.1 Hz. The participants performed four motor tasks. Each task session followed a block design consisting of four 90-s block cycles (60 s of task execution followed by 30 s of rest). From these recordings, 200 activation and connectivity maps were extracted across the task blocks. K-means clustering was applied to identify distinct cortical activation patterns. The following three patterns were identified: Cluster 1, widespread activation and strong connectivity, higher Fugl-Meyer Assessment Upper Extremity (FMA-UE) scores, and better task accuracy; Cluster 2, moderate activation and connectivity, suggesting balanced task engagement; Cluster 3, limited activation and weak connectivity, linked to lower motor function and greater task difficulty. Multinomial logistic regression showed that higher FMA-UE scores increased the likelihood of being classified into Cluster 1. These findings suggest that clustering of cortical patterns reflects motor capacity and task performance for individuals post-stroke. With further validation, this approach may serve as a biomarker for real-time task adaptation and personalized rehabilitation strategies.