A Machine Learning Approach for Better Understanding the Neuromodulation of Locomotion

Abstract: Motor intent and control rely on complex high-level and spinal networks. Untilnow, little is known about this system’s organization and mechanisms. Whilecognitive abilities play an essential role in planning movements, learning andmemorizing, their involvement during stereotyped tasks execution, aslocomotion, is still controversial. Recently, the relationship between cognitivefunctions and gait has received increasing attention.Here, a machine learning approach is used to investigate the engagement ofdi↵erent cortical areas during motor activity. In particular, data coming fromthree subjects with implanted electrodes have been analyzed in the frequencydomain to predict their tasks’ state. The choice of intracortical data hasallowed to elude motion artifacts’ presence and exploitation concern. Goodand satisfactory results have been achieved in the case of not highlystereotyped activity. During ambulation, an evidence of an engagement of thebrain has been shown even if with lower classification performances. Moreover,the cortical areas that have emerged in this analysis seem to be in line withthe relative functionality hypothesized in literature.