Recent years have witnessed an explosion of interest in neuroimaging the human brain and in using it to predict brain-based disease. However, the field generally conceives of neuroimaging as revealing disease-specific activation areas or networks, rather than explicitly considering the forces that govern the brain networks’ development, and treating these nodes and connection as a self-interacting dynamical system, evolving over time. Such an approach is critical for improving our understanding, and therefore predictions, regarding trajectories for phenomena such as neurodegenerative disease, addiction, as well as recovery. These trajectories are likely to be nonlinear (e.g., involving thresholds, saturation, and self-reinforcement) as well a highly specific to each individual. The conceptual transition—from thinking of neuroimaging as providing a static biomarker to thinking of its ability to capture a dynamic process—is critical for probing how a disease first develops in the brain, as well as for asking why two individuals—with identical clinical diagnoses—might show markedly different prognoses over time.
Most physiological diseases are dysregulatory, which means that the negative feedback loops that maintain homeostasis break down in various ways. While the term ‘dysregulated’ tends to be used qualitatively within the human/clinical neuroscience literature, regulation has a precise and testable meaning in control systems engineering, which cannot be assessed by standard neuroimaging methods. To date, functional magnetic resonance imaging (fMRI) is generally used for human neuroimaging in essentially two ways: to infer brain activation maps (areas of differential hemodynamic response) and to infer brain connectivity between dyads (regions that are co-activated). Activation maps are inferred by statistical comparisons between experimental conditions or populations, revealing task-activated neuroanatomical areas. Newer connectivity-based techniques rely upon time-course cross-correlations between two voxels or regions to infer connection strength . This is true of resting-state study designs, which remove the subtraction element of fMRI analysis, but maintain dependence on identifying regions of interest, as well as graph theoretic measures that quantify global connectivity features via correlation matrices.
Two lesser-known fMRI analysis methods that conceptually differ from these two methods are dynamic causal modeling (DCM) and nonlinear complexity methods. DCM uses ordinary or stochastic differential equations to calculate transfer functions between nodes. However, DCM was designed not to describe self-interacting control systems with dynamics that evolve over time, but rather to create directed graphs that can be compared using Bayesian model-selection techniques. As such, DCM actually strips away dynamic features, as its outputs consist of coefficients that represent connection strengths between dyads. In the last few years, our group and others have started to use nonlinear complexity methods (such as power spectrum scale invariance, Lyaponov exponents, and Shannon entropy), in conjunction with fMRI, to probe dysregulated negative feedback loops in the brain. These methods, first applied to physiology in the context of the autonomic nervous system, exploit the fact that negative feedback loops provide unique dynamic signatures, which are disrupted when the system deviates from homeostatic regulation. Complexity methods describe system-wide phenomena, and excel as a diagnostic tool because results provide information about the locus and type of dysregulation. Their primary disadvantage is that they yield minimal information about a system’s structure, and cannot provide simulations that predict future trajectories.
The analytical methods that LCNeuro currently develops are designed to integrate neuroimaging more fully within computational neuroscience, which uses systems of coupled differential equations to describe dynamic behavior of a nonlinear system as it evolves over time. This is an approach that has not been used previously, but which we believe has the potential to offer a transformative impact on neuroimaging as an investigative tool in quantifying brain circuit regulation.