Abstract Details

The human hippocampus is well known for its memory and learning capabilities, with growing evidence supporting its extensive involvement in movement execution and motor processing. With advancing research on neuroprosthetics and their ability to restore bodily function, research on interpreting neural oscillations from brain structures is gaining more precedence.

Ten patients (5 female, ages 21-46, mean age 33.7) diagnosed with refractory epilepsy were implanted with intracranial electrodes in hippocampal structures for seizure localization. Patients were asked to perform a direct reach Go/No-go arm movement task, with successful trials ran through a multitaper spectral analysis. Power magnitude values within the beta frequency band (13-30 Hz) were run through a two-dimensional principal component analysis (PCA) to identify high variance components for classification of data. A cross-validated discriminant analysis classifier was conducted for each patient using five different models (linear, diag-linear, pseudo-linear, diag-quadratic, and pseudo-quadratic). Error rates were calculated for each model to discern the most accurate classifier for the PCA data.

PCA results revealed possible classification of data by experimental conditions (go versus no-go). In 9 out of the 10 patients, the diag-quadratic model had the lowest error rate (19.1 ± 15.6%) of the five classification models tested.

These findings indicate that PCA combined with the diag-quadratic model can be utilized to decode neural modulations and determine intention to execute or inhibit motor function. This form of analysis and classification can be considered for brain-computer interface systems to enable individuals with motor deficits to control assistive devices.