Automatic Recognition and Feature Extraction of Motor-Evoked Potentials Elicited by Transcranial Magnetic Stimulation

Abstract

Transcranial magnetic stimulation can be used as a tool for assessment of corticospinal tract integrity by extracting features from elicited motor evoked potentials (MEPs). However, the selection of MEPs from recordings that present them, and the computation of amplitude and latency are usually performed manually. A methodology based on computing summed power calculated by means of the Discrete Fourier Transform in 10-ms windows was used for automatization of this process and compared to manual measurements. Results showed that MEP selection coincided in 69% of cases, and amplitude and latency had average differences of 7% between methods. Therefore, automatization of MEP computations is feasible using frequency-based features.