A robust strategy for decoding movements from deep brain local field potentials to facilitate brain machine interfaces

A major thrust in brain machine interface (BMI) is to establish a robust, bi-directional direct link between the central nervous system (CNS) and artificial devices (e.g. medical implants, artificial organs, neural stimulators, robotic hands, etc.) for cybernetic interface and treatment of a range of neurodegenerative conditions. Significant effort has centered on support of motor control through external devices and direct stimulation through implanted electrodes in the brain, ideally supporting paralyzed or neurally damaged patients by bypassing damaged regions of the brain. The majority of neural decoding studies have focused on cortical areas for BMIs; however deep brain structures have also been involved in motor control. The subthalamic nucleus (STN) in the basal ganglia, for example, is involved in the preparation, execution and imagining of movements, and represents an unexplored alternative source with great potential for driving BMIs. The goal of this study is to establish this potential through decoding of deep brain local field potentials (LFPs) related to movement execution and laterality of visually cued movements. LFPs were recorded bilaterally from the STN through deep brain stimulation electrodes surgically implanted in patients with Parkinson´s disease. The frequency dependent components of the LFPs were extracted using the wavelet packet transform. In each frequency component, signal features were extracted as the instantaneous power computed using the Hilbert transform. Based on these extracted features, a new feature selection strategy was developed to efficiently select the optimal feature subset. Two classifiers, the Bayesian and support vector machine (SVM) were implemented alongside this novel feature selection strategy, and evaluated using a cross-validation procedure. With optimised feature subset, average correct decoding of movement achieved 99.6±0.2% and 99.8±0.2% and subsequent laterality (left or right) classificatio- reached 77.9±2.7% and 82.7±2.8% using the Bayesian and SVM classifier respectively. The work suggests that the neural activity in the basal ganglia can be used for controlling BMIs and holds great promise for a future generation of interfaces based in the STN.