BaMS3: bayesian motor mapping with structured inference for anatomical precision

Objective.Transcranial magnetic stimulation (TMS)-based motor mapping is a key method for non-invasively localizing cortical muscle representations from motor evoked potentials (MEPs). However, conventional vertex-wise-mapping approaches treat cortical locations independently and do not account for spatial interactions, non-linear recruitment dynamics, or variable trial-to-trial noise, limiting spatial precision particularly in anatomically complex and folded cortex. We present BaMS3 (Bayesian motor-mapping with 3-stage structured fitting), a probabilistic framework that jointly models spatial sensitivity, nonlinear recruitment dynamics, and variable noise in TMS-based motor mapping.Approach. BaMS3 incorporates anatomical and physiological determinants of MEP generation by modeling spatial dependencies across cortical locations, nonlinear input-output relationships, and location-specific variability. The framework was evaluated using subject-specific synthetic simulations and empirical datasets from eight healthy participants and compared with conventional-based mapping in terms of hotspot localization accuracy and motor map focality.Main results.BaMS3 consistently outperforms conventional-based mapping, yielding more anatomically precise and spatially focal motor maps while preserving canonical hotspot localization in empirical datasets. The largest improvements were observed in anatomically complex cortical regions, including sulcal walls and cortical folds.Significance.By modeling TMS motor mapping as a probabilistic inference problem, BaMS3 provides a principled approach for individualized functional brain mapping and may improve spatial target definition for causal brain mapping, cognitive neuroscience, and preoperative functional localization.