Precision Neuromodulation of Sensorimotor Systems

Abstract

Therapeutic neuromodulation has an established history for clinical indications, such as deep brain stimulation for movement disorders and spinal cord stimulation for pain, despite an incomplete understanding of its mechanism of action. Novel neuroprosthetics have the potential to enable wholly new therapies, including sensory restoration and treatment of affective disorders. In order to fully realize the potential of these interventions, precise parameterization of stimulation, informed by better understanding of underlying processes, is required. This dissertation explores the temporal and spatial determinants of outcomes for stimulation within the context of clinical and experimental sensorimotor neuromodulation. The first study of the dissertation defines a new functional target for subthalamic deep brain stimulation for Parkinson disease. We use logistic LASSO to identify features of wideband neural recordings associated with therapeutic stimulation regions derived from patient-specific anisotropic tissue activation models. The study identifies several electrophysiological markers of optimal stimulation regions, which were used in a support vector machine classifier to predict therapeutic activation regions with 64% sensitivity and 82% specificity. By predicting entire regions of therapeutic activation, this algorithm provides a tool for efficient optimization of stimulation programming. The second study investigates the use of empirical mode decomposition, a relatively novel signal analysis tool notable for its ability to extract time-variant and non-sinusoidal signal components, for neuroprosthetic control. We directly compare the performance of empirical mode decomposition against Fourier bandpass filtering, wavelet analysis, and principal spectral component analysis within the context of electrocorticography-based finger movement decoders. Using a Naïve Bayes classifier to detect thumb movement and decode finger flexion, our results indicate that it does not outperform conventional tools despite significantly higher computational cost. The third study presents a novel form of lead localization utilizing impedance. The study presents a scalable and computationally efficient whole-brain impedance atlas derived from individual patient diffusion tensor images. The study then shows that in vivo impedance measurements generally match patterns observed in electrostatic simulations of impedance in patient-specific anisotropic brain conductance models. However, we find that monopolar impedances measured using a clinical macroelectrode provide spatial information at the resolution of 2-5 millimeters, requiring additional refinement to achieve precision necessary for clinical use. The fourth study evaluates stimulation of a novel subthalamic target for modulation of pain. Rodent studies show that stimulation of zona incerta can provide analgesic effect, and clinical evidence suggests that stimulation of a nearby nucleus, nominally used to treat Parkinson disease, often also results in improvement of pain symptoms. We directly test the analgesic effect of zona incerta stimulation in humans and show that stimulation at 20 Hz, the physiological frequency of zona incerta, selectively reduces perceived heat pain. Stimulation at 60 and 130 Hz does not modulate sensation. Likewise, stimulation of zona incerta does not modulate sensation of non-painful heat or pressure pain threshold. The final two studies briefly examine temporal dynamics of evoked sensory activity—within a single unit and across multiple channels. The first study demonstrates, in humans undergoing thalamic deep brain stimulation surgery for essential tremor, that microstimulation of a small region in sensory thalamus with a pulse pattern modeled after its own evoked activity can reproduce the original sensation. The second study shows that natural tactile stimulation evokes highly asynchronous activity in sensory cortex of a primate model.