Using Low-Frequency EEG Signals to Classify Movement Stages in Grab-and-Lift Tasks

Nowadays, Brain-Computer Interface (BCI) systems are considered a tool with enormous potential to establish communication alternatives, restore functions, and provide rehabilitation processes to patients with neuromotor impairment. A wide variety of invasive and non-invasive methods has been studied to control BCI systems, especially with electroencephalography (EEG) signals. However, despite numerous studies in this field, much work remains to be done to understand the underlying neural mechanisms and to develop versatile and reliable BCI systems. Typically, BCI systems oriented to motion decoding are based on information extracted from sensorimotor rhythms, which correspond to the EEG signal in the mu (8–12 Hz) and beta (18–30 Hz) bands. In this work, we focus on the search for information in low-frequency bands (0.1–7 Hz). To accomplish this goal, we work on the classification of six stages of gripping and lifting movements of an object. The features of the signals were extracted applying Discrete Wavelet Transform (DWT) and Empirical Mode Decomposition (EMD). Our results suggest that, for this case, the most significant amount of discriminant information is within the (0–4 Hz) band (maximum accuracy of 89.22 ± 0.81%). Another remarkable result is the high similarity observed between the waveforms belonging to the same stage between different subjects. This result is especially motivating since numerous studies have demonstrated that the EEG signals present a high inter-subject and inter-session variability.