The computational properties of the human brain arise from an intricate interplay between billions of neurons connected in complex networks. However, our ability to study these networks in healthy human brain is limited by the necessity to use noninvasive technologies. This is in contrast to animal models where a rich, detailed view of cellular-level brain function with cell-type-specific molecular identity has become available due to recent advances in microscopic optical imaging and genetics. Thus, a central challenge facing neuroscience today is leveraging these mechanistic insights from animal studies to accurately draw physiological inferences from noninvasive signals in humans.
On the essential path towards this goal is the development of a detailed “bottom-up” forward model bridging neuronal activity at the level of cell-type-specific populations to noninvasive imaging signals. The general idea is that specific neuronal cell types have identifiable signatures in the way they drive changes in cerebral blood flow, cerebral metabolic rate of O2 (measurable with quantitative functional Magnetic Resonance Imaging, fMRI), and electrical currents/potentials (measurable with magneto/electroencephalography, MEG/EEG). This forward model would then provide the “ground truth” for the development of new tools for tackling the inverse problem – estimation of neuronal activity from multimodal noninvasive imaging data.
To illustrate this approach, we focus on the primary somatosensory cortex where bottom-up models can be built and calibrated taking advantage of well-studied neuronal network phenomena such as the surround and transcallosal inhibition. In animals (e.g., mice) we can utilize microscopic measurement technologies to precisely and quantitatively probe concrete microscopic neuronal, vascular, and metabolic parameters while manipulating cell-type-specific neuronal activity (blue boxes in the Figure). These microscopic data can then be used to simulate the corresponding macroscopic physiological parameters (CBF, CMRO2, and current dipole moment) and their reflection in noninvasive observables (red boxes in the Figure). Thus, in mice, we can develop a detailed forward model bridging neuronal activity at the level of cell-type-specific populations to noninvasive imaging signals. Furthermore, we can validate this model at each step against real data. For human translation, first we would have to calibrate the model parameters to account for systematic differences due to known physical scaling laws such as differences in vessel size or latency of the cortical neuronal response. Then, the remaining uncertainty in the translation of model parameters from mouse to human (as well as the measurement noise and subject-to-subject variability) would be factored into a single Bayesian estimation framework to obtain estimates of the parameters of interest (i.e., activity of cell-type-specific neuronal populations) and quantify the uncertainty of estimation.
Read more on our approach to the physiological underpinning of human noninvasive imaging in our recent publication in Philosophical Transactions of the Royal Society B.