Dynamic resting-state fMRI analysis

Abstract

Brain works dynamically, forming a complex network in which different brain regions are segregated and integrated over time. A complex brain network is observed divided into several intrinsic connectivity networks (ICNs), such as default mode network, executive control network, salience network, visuospatial attention network, depending on functional characteristics in the resting-state brain. Functional connectivity (FC) network associated with the intrinsic functional architecture of the brain has greatly aided in the study of neurological and psychiatric disease.

Originally, static FC analysis considering full-time signal patterns was mainly used to observe time-varying brain activity patterns on resting-state functional magnetic resonance imaging (fMRI) usually acquired in 5-15 minutes. However, in recent years, dynamic FC analysis considering short- time (30-60 seconds) signal patterns instead of static FC analysis has been attracted attention in discovering dynamic characteristics in brain function across groups and individuals.

In this dissertation, two novel methods were proposed for analyzing dynamic resting-state fMRI. Firstly, ICN efficiency, a new graph theory- based network parameter, was suggested to evaluate the frequency- specific contribution of ICN to brain network integration, and dynamic ICN was observed in terms of the frequency. The frequency-specific contribution of each ICN to the brain network integration was confirmed by statistically comparing the ICN efficiency in the two frequency bands. By observing the dynamic connectivity of the ICN in terms of frequency, it has been confirmed once again that the resting-state brain operates dynamically over time. Secondly, assuming that dynamic information during one minute of fMRI is unique and repetitive for each individual, the Siamese long short- term memory (LSTM) network was implemented for individual identification based on individual BOLD signal. Using Siamese LSTM, it is possible to identify the same gallery data as the specific probe data without further learning about the new data set. These results demonstrated the individual uniqueness of dynamic resting-state fMRI. Normal group and normal individual analyses for dynamic resting-state fMRI in this dissertation could contribute to the understanding and utilization of resting-state brain function.