Modelling functional connectivity networks in fMRI data

Prof. Kristoffer H. Madsen from the Danish Research Centre for Magnetic Resonance at Copenhagen University will talk on fMRI data modeling.

Spontaneous or task evoked fluctuations in the functional Magnetic Resonance Imaging (fMRI) data can be used to assess functional relations between brain regions. However, given the datasets typically recorded in high spatial resolution with a limited number of (temporal) samples available under low signal to noise conditions, it is an on-going challenge to detect stable and generalizable functional connectivity among brain regions.  For data recorded over several subjects these problems are exacerbated due to additional subject-dependent variability and other between sessions noise sources. Spatial subspace representations by component models can be used to define reduced-noise representations of functional connectivity. Using a flexible cross-validation framework assessing the ability of the models to predict voxel-wise covariance in new data, I will demonstrate that both too constrained (average) models and highly flexible (individual) models are poor predictors of whole brain
connectivity, whereas models account for subject variability within a common spatial subspace have better generalization performance.

Kristoffer H. Madsen received his PhD from the Technical University of Denmark, at the institute for mathematical modelling in 2008 concerning modelling strategies for function magnetic resonance imaging. He has since worked at the Danish Research Centre for
Magnetic Resonance at Copenhagen University Hospital Hvidovre primarily focusing on analysis of functional magnetic resonance imaging (fMRI) and studies that combine fMRI with other modalities such as EEG and transcranial brain stimulation techniques.