From Oscillations to Brain States: Real-Time EEG-TMS for Adaptive Neuromodulation.

Transcranial magnetic stimulation (TMS) enables non-invasive, focal modulation of cortical circuits by inducing electric currents in the brain through electromagnetic induction, thereby influencing neuronal excitability and synaptic plasticity. High inter- and intra-individual variability has led, however, to moderate efficacy and reproducibility of stimulation and treatment protocols, motivating a shift toward brain-state-dependent stimulation. Over the past decade, real-time phase-triggered EEG-TMS has established the oscillatory phase-particularly focusing on the sensorimotor mu rhythm-as a key determinant of cortical excitability and plasticity modulation. The field, however, remains largely confined to univariate, sensor-space analyses of local mu-rhythm phase, missing large-scale network dynamics. Recent advances in online EEG source reconstruction and multivariate machine and deep learning (ML/DL) approaches have begun to move beyond local phase toward whole-brain, network-level state estimation, achieving encouraging preliminary accuracies in predicting trial-by-trial cortical excitability, with promising applications in network-dysregulation conditions such as chronic pain. Integrating source-space reconstruction and individual biological variability, and adaptive ML/DL pipelines into closed-loop frameworks promises to move beyond generic stimulation protocols toward selective, network-targeted neuromodulation tailored to the individual's dynamic brain state. Against this background, this review provides a critical overview of current achievements and limitations, while highlighting emerging methodological directions toward fully brain-state-adaptive and network-targeted EEG-TMS. We further present an illustrative use case of adaptive EEG-TMS for pain modulation, where treatment responses remain heterogeneous and the relevant dynamics are distributed across networks, and which therefore stands to gain most from individualized, network-targeted protocols.