In an ever-changing and complex environment, the ability to precisely and rapidly regulate actions is crucial for effective operations. This requires an online decision-making process, which includes gathering information from multiple sensory inputs sources: vision, audition, gustation, olfaction, and somatosensation; integrating these information with prior knowledge and current goals; making rapid decisions to generate motor commands; and executing motor commands. During movement, we are also frequently required to stop the current movement, make new decisions based on new incoming information, and switch to other action plans when needed. These processes involve complex interplay of multiple brain regions across the cortical-subcortical network, in which information and goals are converted to action sequences while taking into account of uncertainty. Although both the decision-making and motor control aspects of action regulation have been extensively investigated, how different functions-e.g., action selection, stopping, switching-are interrelated both behaviorally and mechanistically remain elusive, partly because of the tendency of exploring these functions in isolation, making it difficult to develop a unified theory of action regulation. The main goal of this work is to construct a large-scale, unified neurocomputational framework that predicts how the brain selects, stops and switches actions.

Part of this thesis explores computational modeling of action regulation functions. In two related studies, we evaluated our model hypotheses by analyzing the motor behavior of both neurotypical individuals and Parkinson's disease (PD) patients during tasks involving action inhibition. This approach demonstrated the model's capacity to explain key features of action regulation behaviors in both groups. Building on these findings, in a collaborative work, we extended this model, enabling precise adaptive environment-aware target reaching in robotic manipulation.

We are also interested in the obstacle avoidance behavior in cluttered environments. Based on stochastic optimal control (SOC) theory, we constructed a computational framework that integrates value information associated with goals, obstacles and actions, and demonstrating its ability to capture key aspects of human reaching behavior in complex environments where both obstacles and targets are present.

Neurological disorders affect millions of people worldwide, with devastating health and cost burdens. Therapeutic neuromodulation has received tremendous attention due to its promise as a viable alternative for treating drug-resistant neurological conditions. While promising, clinical outcomes are sub-optimal and patient specific because the mechanisms of actions of neuromodulation remain speculative at best. Primarily because the functional architecture of the brain and spinal cord is not well understood due to intrinsic challenges associated with existing neuroimaging and monitoring modalities. Functional ultrasound imaging (fUSI) is a recently developed minimally invasive neuroimaging technique that can record blood flow dynamics at a level of sensitivity, spatial and temporal precision previously not available. Here we leverage fUSI to investigate hemodynamic responses as a proxy to neural activity modulations in the brain and spinal cord induced by drug, neurological disease, and neuromodulation. In animal preclinical studies, we find that MK-801 and ketamine NMDA antagonists cause specific spatiotemporal changes to cerebral blood volume (CBV), and subsequent MSN theta-frequency deep brain stimulation has strong effect on hippocampal CBV in a MK-801 drug induced hypoperfusion brain. In human clinical work, we predict spinal cord hemodynamic responses to epidural electrical stimulation of the spinal cord at a single trial-level, with high accuracy. Additionally, we decode bladder-pressure state – a physiological function solely from spinal cord fUSI signal acquired during filling and emptying of the bladder. Given these outcomes we extend our results to show that aberrant epileptic-seizure brain state causes distinct changes to local brain connectivity and network activity. We found a strong reduction in cross-correlation coefficients of local interictal fUSI signal acquired from neighboring healthy brain regions after surgical resection of aberrant seizure brain foci. Importantly, we demonstrate that non-invasive common peroneal nerve stimulation (PNS) disrupts local brain network-connectivity by reducing fUSI interictal signal correlation coefficients below a threshold when the PNS stimulator is turned on. Our ability to characterize and decode brain and spinal cord states in response to drug, disease, and neuromodulation in-vivo opens avenues to understand spinal cord and brain function, dysfunction, and effects of neuromodulation – ultimately vital for translational development of real-time closed-loop neurorehabilitation systems.

Purpose

Patients and methods

Results

Conclusion