Home
This is the wiki documentation for BrainTrak, the brain parameter estimation and tracking system. The corresponding research publication is link.
The purpose of BrainTrak is to fit the parameters of forward models to EEG data. Conceptually, BrainTrak consists of two parts
- The theory of expressing the goodness-of-fit as a probability distribution (e.g., Denison,Punskaya), sampling this distribution by using Markov Chain Monte Carlo methods (e.g., Metropolis,Haario), and Bayesian belief updating to track the parameters over time (developed independently by Abeysuriya and Cooray)
- The implementation of this system in the software package, which is called BrainTrak
The theory is applicable to general curve fitting and tracking, and can be applied to a wide range of nonlinear optimization problems for time-evolving systems. The implementation in BrainTrak is specific to fitting neural field models to EEG spectra. The role of BrainTrak is to produce fits like the one below.
Probably the most significant development of BrainTrak compared to previous fitting work (e.g., Rowe) is the treatment of uncertainties. In addition to the best-fit parameters, BrainTrak produces a joint probability distribution quantifying the goodness-of-fit for different parameter combinations. This can then be displayed as marginal distributions in various projections of the parameter space. For example, in XYZ the fit can be displayed as a dot corresponding to the XYZ quantities for the best fit parameters, and the uncertainties can be displayed as a cloud where the density corresponds to the likelihood of the parameters.
We can also compute the marginal distributions for each parameter individually, also shown above.
To get started, it is highly recommended that you work through each of these pages in the order below.