21. Effect of region coordinates shifting on dynamic causal modelling results

Abstract

Dynamic Causal Modelling (DCM), as one of methods for effective brain connectivity analysis allows us making inferences about neural processes that underlie measured functional magnetic resonance imaging (fMRI) data. The main goal is to estimate parameters of the neuronal system model, whose outputs correspond most precisely to observed blood oxygenation level dependent (BOLD) response. As DCM is not exploratory technique, we have to define a hypothesis, which contains information about inputs, connections and brain regions. This contribution deals with the effect of region coordinates selection on the DCM results. For exact evaluation of the shifting effect we designed and implemented a data simulator based on the DCM model. Simulated BOLD signals from different regions of interest are generated with selectable fine time step and then down-sampled to the specific repetition time (TR). The idea of the shifting simulation on hemodynamic level is that decreasing the signal to noise ratio corresponds to a greater shift, which means that the signal extracted from the region of correct coordinates consists of 100% useful BOLD signal, while the signal from a region, not included in the analysis, is formed only by noise. The noise means low frequency fluctuations, realistic physiological phenomena, signal closely correlated to the useful BOLD signal or their combinations. Thereafter the connectivity is estimated by the same model, which is used for the simulation but with sparser sampling. We are interested in the amount of correctly estimated connections between selected regions of the specific model. Monte Carlo simulations are used for investigation of the model behavior. Based on the simulation results, we can say that the shifting affects the number of correctly estimated connections. However, DCM shows some uncertainty because, also in case of no shift, there exist results with one falsely estimated connection.