A common neural architecture for encoding finger movements
Human behavior evokes cortical responses that are highly variable across individuals, challenging the existence of a shared informational architecture, previously reported in animal studies. Here, using 7-Tesla fMRI and Procrustean transformation, we show that high-dimensional alignment of neural information from finger-tapping sequences reveals a latent neural architecture for sequential finger movements that generalizes across brains. This shared representation is instantiated in both sides of the sensorimotor cortex and does not reflect trial-wise variability of motor behavior, as regional decoding differences persisted after controlling for movement-time fluctuations. This conserved sensorimotor encoding structure provides a neuroscientific foundation for scalable, calibration-free brain-computer interfaces and cross-subject models of motor rehabilitation.