Methods for Correcting Artifacts in FMRI Time Series

Mark Reimers (Biostatistics, VCU)

A major problem in the initial stages of neural signal analysis is the identification and compensating artifacts, many of which reflect physiological processes. In functional MRI data the major sources of artifacts are head motion, changes in signal strength with breathing, and pulse. We would like to estimate the size of these effects across all the measures, but we usually don’t have accurate independent measures of these effects; furthermore these effects generally do not closely track external measures of breathing or pulse.

I introduce a method of constructing synthetic controls to provide a first step to identify artifacts in fMRI or other neural time series data. Synthetic controls are differences of little biological significance, which however differ in their relation to anticipated (but unmeasured) artifacts. For a typical MRI scan, with alternating planes of excitation, differences between grey matter voxels in adjacent planes may play this role. The large number of such differences turns out to have very strong systematic patterns summarized by a very few principal components, which are almost orthogonal to the predictors derived from the experimental design. These factors in turn predict typically 50% of variance in voxels not used for the construction of the factors. These factors may be further improved by an iterative constrained fitting procedure. This procedure seems to improve the S/N ratio of the data by a factor of two.

 

Preferred presentation format: Poster
Topic: Neuroimaging

Document Actions