Mind the drift - improving sensitivity to fMRI pattern information by accounting for temporal pattern drift

Abstract

Analyzing functional magnetic resonance imaging (fMRI) pattern similarity is becoming increasingly popular because it allows one to relate distributed patterns of voxel activity to continuous perceptual and cognitive states of the human brain. Here we show that fMRI pattern similarity estimates are severely affected by temporal pattern drifts in fMRI data -even after voxelwise detrending. For this particular dataset, the drift effect obscures orientation information as measured by fMRI pattern dissimilarities. We demonstrate that orientation information can be recovered using three different methods: 1. Regressing out the drift component through linear modeling; 2. Computing representational distances between conditions measured in independent imaging runs; 3. Crossvalidation of pattern distance estimates. One possible source of temporal pattern drift could be random walk like fluctuations -physiological or scanner related -occurring within single voxel timecourses. This explanation is consistent with voxel-wise detrending not alleviating pattern drift effects. In addition, this would explain why crossvalidated pattern distances are robust to temporal drift because a random walk process is expected to give rise to non-replicable drift directions. Given these findings, we recommend that future fMRI studies take pattern drift into account when analyzing pattern similarity as this can greatly enhance the sensitivity to experimental effects of interest.

Human Orbitofrontal Cortex Represents a Cognitive Map of State Space

“…Correlating patterns estimated from the same run can introduce biases in the correlation matrix through the effect of temporal contingencies on the estimated correlation (Alink et al 2015; Diedrichsen et al 2011). To prevent this, we calculated pattern similarity by cross-correlating patterns estimated from separate runs, ensuring that the events modeled by each pair of regressors are fully temporally separated and therefore the regressors are orthogonal.…”

“…To prevent this, we calculated pattern similarity by cross-correlating patterns estimated from separate runs, ensuring that the events modeled by each pair of regressors are fully temporally separated and therefore the regressors are orthogonal. Because the noise from two different runs is unlikely to be correlated, this method does not introduce biases into the estimated correlations of patterns (Alink et al 2015). Specifically, for each run, each state map (masked by the anatomical OFC) was correlated with all state maps from all other runs.…”

Sequential and efficient neural-population coding of complex task information

“…An even more stringent test uses a conservative cross-validation paradigm [107][108][109] , where two sets of decoding directions were computed separately using two disjoint subsets of the data. In this case, random neural fluctuations selected by (say) decoding directions computed using the first half of trials should have no relationship to those computed using the second half of trials, and we should thus find no noise-induced structure in angles between decoding directions computed in two different halves of the data.…”

Computational approaches to fMRI analysis

“…One caveat for this general method is that all voxels are weighted equally, unlike in classifier-based MVPA, and thus there is a risk of contamination from uninformative or noisy features. Another caveat is that pattern similarity can be easily confounded, including by univariate activity 40 and temporal proximity 41 : in such cases, effects on similarity might be interpreted as neural patterns converging or diverging in representational space, when in fact the underlying structure of the neural patterns has not changed.…”