Enhancing Speed and Accuracy of Motor Evoked Potential Recruitment Curve Analysis Using Hierarchical Bayesian Modeling

SUMMARY Recording motor evoked potentials (MEPs) in response to electrical and electromagnetic stimulation of the sensorimotor system is an important technique to probe the nervous system. It allows for the assessment of state, measuring the extent of injury, tracking recovery, and measuring the efficacy of interventions, including neuromodulation. The relationship between stimulation intensity and MEP size is captured by the recruitment curve (RC), defined by several parameters, notably the threshold intensity that triggers an MEP. Current RC characterization techniques, which are either offline or online, have distinct limitations. Offline methods, applied post-data acquisition, rely on numerical optimization to fit RCs but often lack precision in parameter estimation. Online methods, while optimizing sampling during experiments and efficiently calculating threshold parameters in an individual muscle, have a limited scope and miss critical estimates, such as off-target changes in excitability, which could be crucial in neuromodulation studies. To address these challenges, we have developed a Python library, hbMEP, which integrates a novel RC within a hierarchical Bayesian framework for offline use, post data acquisition. Our results indicate that hbMEP provides more accurate parameter estimates and increases statistical power over conventional methods. To further capitalize on the advantages of our approach, in Aim 1, we plan to extend hbMEP with an online Bayesian recruitment curve estimation algorithm. While maintaining the robustness and accuracy of our offline approach, this enhancement leverages hbMEP's generative capabilities to optimize sample selection, targeting multiple muscles, parameters or the entire curve—capabilities beyond currently used algorithms. The second aim involves a comprehensive validation of the extended hbMEP's capabilities through physical experiments. We will validate both the technical and scientific merits of our methods by comparing them against currently dominant algorithms such as ML-PEST. Concurrently, we will further extend hbMEP with a user-interface. This will simplify advanced analyses, making them usable by a broader range of users and minimizing the necessity for extensive pre-processing. We will validate usability by aggregating study level metrics such as the System Usability Scale (SUS) and task level metrics under the guidance of Columbia’s Usability Center. Successful completion of this proposal will impact how neurophysiology experiments are conducted, offering more accurate insights into a stimulated nervous system’s responses and a more generalizable approach to test hypotheses regarding the neurophysiological effects of an intervention. This hierarchical Bayesian approach will significantly lower experimental burden by reducing the number of participants and stimuli required while maintaining accuracy and simultaneously increasing the number of muscles across which these insights are obtained. Moreover, our approach sets the stage for future closed-loop experimental designs and novel state-dependent analyses, thereby contributing substantially to the field.