Abstract:
|
The study of dynamic functional connectivity (dFC) has come to the forefront of research efforts in the field of neuroimaging. Several different approaches to dFC analysis have been suggested, such as sliding window correlations, change point models, dynamic connectivity detection, and hidden Markov models (HMMs). Recently, we have proposed using hidden semi-Markov models (HSMMs), which allow for non-geometrically distributed sojourn times, to find each subject's most probable series of network states, the probabilities of transitioning from one state to another, as well as the networks associated with each state. However, a comprehensive simulation study is needed to investigate the performance of HMMs, HSMMs, and several other approaches, such as the traditional sliding window analysis. We perform a simulation study to compare the estimated states, transition probabilities, and sojourn times across models for simulated data sets where we vary the true sojourn distributions, signal to noise ratios, number of states, and transition probabilities. By doing so, we inform users of the strengths and limitations of each method for a range of data scenarios.
|