Classifying the valence of autobiographical memories from fMRI data

Abstract

We show that fMRI analysis using machine learning tools are sufficient to distinguish valence (i.e., positive or negative) of freely retrieved autobiographical memories in a cross-participant setting. Our methodology uses feature selection (ReliefF) in combination with boosting methods, both applied directly to data represented in voxel space. In previous work using the same data set, Nawa and Ando showed that whole-brain based classification could achieve above-chance classification accuracy only when both training and testing data came from the same individual. In a cross-participant setting, classification results were not statistically significant. Additionally, on average the classification accuracy obtained when using ReliefF is substantially higher than previous results -81% for the within-participant classification, and 62% for the cross-participant classification. Furthermore, since features are defined in voxel space, it is possible to show brain maps indicating the regions of that are most relevant in determining the results of the classification. Interestingly, the voxels that were selected using the proposed computational pipeline seem to be consistent with current neurophysiological theories regarding the brain regions actively involved in autobiographical memory processes.

Electroencephalography functional connectivity—A biomarker for painful polyneuropathy

“…To overcome such uncertainty, a histogram methodology was used. The analysis (Figure 1 , blue rectangle) was re‐run many times and the frequency (or a recurrence) of each specific functional connection being selected as a discriminating feature was counted, as proposed in [ 30 ]. An example of such a run can be seen in Figure 1 (explanatory analysis path) and Figure 2 , where the x ‐axis indicates selected functional connections (i.e., features) and the y ‐axis indicates frequency (i.e., the repeatability of a specific feature through different selections).…”

A neural signature of social support mitigates negative emotion

“…Since each timepoint of each voxel can theoretically represent a feature, the number of possible features easily exceeds the amount of data points (e.g., the total possible number of features in one trial is 235,375 voxels x 20 TR = 4,707,500 features). As such, dimension reduction and feature selection are likely to be crucial to improve classification results (45). Here, we adopted a two step supversied machine learning approach to reduce dimensionality.…”

Decoding the emotional valence of future thoughts

“…Above chance classifier accuracy in the amygdala and PCC for negative simulations is also interesting in comparison to autobiographical memories. While anterior medial regions have been reported as representing autobiographical memory valence (Frid et al, 2020), the amygdala and PCC have not, raising the possibility that these regions differentially represent phenomenological content when thinking about past versus future. This possibility merits further study because MPVA can highlight neural differences between memory and future thinking missed by univariate analyses (Kirwan et al, 2014).…”

“…Recent attempts to classify the emotion of autobiographical memories have been successful in regions typically involved in episodic memory retrieval (Frid et al, 2020;Nawa & Ando, 2014). Given the similarity in neural activation when remembering the past and imagining the future (Addis et al, 2007), we may expect neural patterning in similar regions to carry information about emotion when simulating future events.…”