Adaptive Bayesian localization of motor representation areas

Transcranial magnetic stimulation (TMS) enables non-invasive localization of cortical motor representations, with important clinical applications in presurgical planning. Existing methods either disregard spatial information about the TMS-induced electric field (E-field) or use acquisition schemes that do not leverage previously elicited motor responses to guide subsequent stimulation. We present an adaptive Bayesian localization method that combines real-time E-field optimization with per-trial probabilistic inference. The cortical origin of the motor responses is represented as a spatial probability distribution that is updated after each stimulus. Subsequent stimuli are then optimized to maximize the expected localization improvement given previous responses. We validated the method experimentally, using multi-locus TMS for adaptive localization and single-coil TMS as a non-adaptive reference with randomized coil placements. Across eight subjects, the adaptive protocol at least halved the number of stimuli required for stable localization compared to the randomized protocol, converging in 60 stimuli on average, with 95% highest-density regions often below 10 mm2 by 150 stimuli.