Multidimensional Motor Evoked Potentials (MultiMEP): Digging up buried information from single trials

Background: To investigate covert motor processes, transcranial magnetic stimulation (TMS) studies often use motor-evoked potentials (MEPs) as a proxy for inferring the state of motor representations. Typically, these studies test motor representations of actions that can be produced by the isolated contraction of one muscle, limiting both the number of recorded muscles and the complexity of tested actions. Furthermore, univariate analyses treat MEPs from different muscles as independent, overlooking potentially meaningful intermuscular relationships encoded in MEPs amplitude patterns at the single-trial level. Objective: We addressed these limitations by adopting a decoding approach to MEPs analogous to multivoxel pattern analysis in neuroimaging. Methods: Using our novel Multidimensional Motor Evoked Potentials (MultiMEP) approach, we tested 22 participants by applying a decoding analysis to MEPs recorded from 24 electrodes during motor imagery of three complex hand actions. Inspired by the 1m1a inference framework, we conducted an exploratory cross-classification analysis by training a classifier on one domain (MultiMEP evoked during motor imagery or action execution electromyographic patterns) and testing it on the other. Results: Imagined actions were classified, based on MultiMEP patterns, with an accuracy of 74%. The cross-classification analysis yielded above-chance accuracies of 54% (execution-to-imagery) and 71% (imagery-to-execution). Conclusions: This proof-of-principle study demonstrates that MEPs encode richer information than previously assumed both at single-subject and at single-trial levels. Our results suggest that MultiMEP decoding represents a first step toward a paradigm shift in studying motor processes with TMS, much like multivoxel pattern analysis revolutionized the way the brain-cognition relationship has been studied through neuroimaging.