Decoding causal mechanisms in brain-muscle activation through AI-driven signal processing

Movement disorders, particularly dystonia, present persistent challenges in diagnosis and treatment due to complex, non-linear interactions between central neural circuits and peripheral motor systems. These bidirectional interactions can become pathologically altered following neurophysiological disturbances. Traditional connectivity analyses often fail to capture the high-dimensional dynamics of such systems, motivating the development of advanced computational frameworks to uncover underlying causal mechanisms. This thesis introduces several advancements that improve our understanding of causal mechanisms in neurophysiology. It presents a novel framework for modeling physiological signals using wavelet sparsity and Gaussian assumptions to distinguish physiological noise from measurement artifacts. The method also enforces mathematical stability constraints while allowing noise separation to identify feature signatures, enhancing causal analysis robustness. The methodology begins with interpretable linear models before advancing to random forests and neural networks to better represent the neural-muscular interactions in movement disorders. A novel dynamic Bayesian network framework enables causal inference in temporal modeling, distinguishing direct and indirect causal effects via a two-layered analysis. This addresses key limitations of Granger causality, offering a clearer interpretation of relationships. The framework incorporates counterfactual analysis grounded in Bayesian principles to assess how observed changes impact the neural-muscular system. Unlike conventional methods, it embeds interventions into a dynamic causal context, inferring potential system responses. These advancements contribute to the study of neuroplasticity, motor control, and various neurological conditions. By integrating mathematically rigorous causal inference with clinically interpretable models, this research enhances the role of computational neuroscience in clinical decision-making. The methods support data-driven diagnostics and intervention planning, paving the way for precision neurology. At its core, this work presents a framework for dissecting neural–muscular communication, revealing causal pathways and breakdowns in sensorimotor signaling. The models uncover hidden dynamical structures, identify biomarkers, and inform interventions, bridging theory and practice in computational neurophysiology.