No fixed target for tongue motor networks: individualized AI neuromodulation lifts target-state occupancy from 82% to 97%

The neuromodulation target shifts from fixed brain regions to each participant's own tongue sensorimotor network, a map continuously updated by real-time brain signal. Earlier real-time fMRI neurofeedback either fed back one or two region-of-interest signals defined anatomically or by atlas, or mapped group network parcellations into individual space, so the target itself did not follow individual functional organization. Scans from 30 healthy participants alternated between control and iNM with an identical task sequence in both conditions, so differences can be attributed to the modulation rather than the task schedule.

During scanning the platform reads the blood-oxygen signal every 2 seconds and uses it to adjust reinforcement of the target network while suppressing the competing swallow-related state. It does so with a general linear model that iteratively updates direction-specific beta weights to characterize the individual network, then an unsupervised autoregressive hidden Markov model that infers the target and interference states from the signal. Across 720 trials per condition, only 16 iNM trials were classified as swallow interference during the tongue-movement interval, versus 133 control trials.

Results come from real-time fMRI scans of 30 healthy participants, and effects persisted into the swallow and rest periods after modulation stopped within the same scanning session; applicability to patients remains a hypothesis to be tested rather than a result reported here.

The authors state that both clinical routes, for supranuclear and infranuclear lesions, remain hypotheses to be tested in patient cohorts, and that the present study establishes only that these networks can be selectively reconfigured in healthy individuals with effects beyond the active modulation window.