101. Mapping event-related fMRI time series into graphs

Adrian Onicas – Computer Vision Data Science Team, Sano Centre for Computational Science, Krakow, PL


Accurate identification of signal changes in time series is central for decoding task-evoked brain activity in fMRI. However, traditional approaches, such as linear modeling by applying a convolution with a hemodynamic response function assume uniformity in the shape of the functional brain response over time, disregarding its variability. To overcome this limitation, visibility graphs (VG) offer a novel method of transforming time series into graphs, enabling the application of complex systems tools for the analysis of signal changes over time. This transformation is robust to between-scan differences in signal amplitude and linear trends and has been previously shown to preserve the properties of the original time series. Yet, it remains unclear whether the temporal network topology captures stimulus-evoked changes in functional brain signals. In this talk, we’ll address the concept of temporal networks and the extent to which visibility graph topology derived from fMRI time series can recover known events in synthetic and real functional MRI data, with applications for non-linear analysis.

About the author

Adrian’s background is in clinical psychology (MSc) and computational neuroscience (Ph.D.). His research interests cover the replicability of neuroimaging methods and applications of advanced neuroimaging techniques (MRI, fMRI, and DTI) in neurological disorders. He has experience with the analysis of large neuroimaging datasets and has taken part in some significant projects in the field, including the Neuroimaging Analysis Replication and Prediction Study (NARPS) and the Advancing Concussion Assessment in Pediatrics (A-CAP) study. Previous work has focused on non-linear signal processing methods, neuroimaging analysis pipelines replicability, multisite data harmonization, and multimodal brain alterations following traumatic brain injury.