White-Matter Pathways for Statistical Learning of Temporal Structures

Abstract

Extracting the statistics of event streams in natural environments is critical for interpreting current events and predicting future ones. The brain is known to rapidly find structure and meaning in unfamiliar streams of sensory experience, often by mere exposure to the environment (i.e., without explicit feedback). Yet, we know little about the brain pathways that support this type of statistical learning. Here, we test whether changes in white-matter (WM) connectivity due to training relate to our ability to extract temporal regularities. By combining behavioral training and diffusion tensor imaging (DTI), we demonstrate that humans adapt to the environment's statistics as they change over time from simple repetition to probabilistic combinations. In particular, we show that learning relates to the decision strategy that individuals adopt when extracting temporal statistics. We next test for learning-dependent changes in WM connectivity and ask whether they relate to individual variability in decision strategy. Our DTI results provide evidence for dissociable WM pathways that relate to individual strategy: extracting the exact sequence statistics (i.e., matching) relates to connectivity changes between caudate and hippocampus, while selecting the most probable outcomes in a given context (i.e., maximizing) relates to connectivity changes between prefrontal, cingulate and basal ganglia (caudate, putamen) regions. Thus, our findings provide evidence for distinct cortico-striatal circuits that show learning-dependent changes of WM connectivity and support individual ability to learn behaviorally-relevant statistics.

Keywords: brain imaging; brain plasticity; diffusion tensor imaging; statistical learning; vision.

Figures

Figure 1.

Trial and sequence design. A, The trial design: 8–14 symbols were presented sequentially followed by a cue and the test display. B, Sequence design: Markov models of the three context-length levels. For the zero-order model (level-0): different states (A, B, C, D) are assigned to four symbols with different probabilities. For first-order (level-1) and second-order (level-2) models, diagrams indicate states (circles) and conditional probabilities (solid arrow: high probability; dashed arrow: low probability). Transitional probabilities are shown in a four-by-four (level-1) or four-by-six (level-2) conditional probability matrix, where rows indicate the temporal context and columns the corresponding target. C, Timeline of the imaging and behavioral sessions included in the study. Training involved three to five sessions for each level. DTI scans and behavioral test sessions were completed on a single day.

Figure 2.

Behavioral performance. Mean normalized PI across participants per level during the first test session (gray bars) and second test session (black bars) for (A) the training group and (B) the no-training control group. Error bars indicate SEM across participants. C, Strategy index boxplots for level-0, level-1, and level-2 indicate individual variability for the training group. The upper and lower error bars display the minimum and maximum data values, and the central boxes represent the interquartile range (25th to 75th percentiles). The thick line in the central boxes represents the median. Open circles denote outliers.

Figure 3.

Seed regions and connection probability maps. A, Seed regions for probabilistic tractography overlaid on the MNI template (z = -8). B, Connection probability maps for each seed region (vmPFC, putamen, caudate). Maps (radiologic convention: left is right) are thresholded at 0.1% of total tracts per seed and are averaged for pre- and post-training sessions and across participants. Results are displayed in MNI for I, vmPFC (x = -12, y = 40, z = -8); II, putamen (x = -32, y = 2, z = 2); and III, caudate (x = -22, y = -2, z = 4) as seeds. Whole-brain tractography was computed separately for the left and right hemisphere and for pre- and post-training sessions, and the maps were combined for visualization purposes.

Figure 4.

DTI regression with strategy. Clusters showing significantly positive (red clusters) or negative (blue clusters) regressions of connection probability change (post- minus pre-training) with strategy index calculated across all trials during training per level. A, For learning frequency statistics, clusters comprise left vmPFC to caudate, right vmPFC to caudate and right putamen to IFG. B, For learning context-based statistics, clusters comprise left vmPFC to caudate and left caudate to hippocampus. Results are displayed in radiologic convention (left is right) and are overlaid on the MNI template. An enlarged view of each significant cluster is displayed for better visibility. Scatterplots of connection probability change (post- minus pre-training) with strategy index for the peak voxel of each cluster are shown on the right panel.

Figure 5.

Group comparison (training versus no-training control). A, Whole-brain tractography: clusters showing significantly higher correlation coefficient of connection probability change (post- minus pre-training) with strategy index for the training versus the no-training control group. Results are displayed in radiologic convention (left is right) and are overlaid on the MNI template. An enlarged view of each significant cluster is displayed for better visibility. Scatterplots of connection probability change with strategy index for the peak voxel of each cluster are shown on the right panel. B, Seed-to-target tractography: correlations of connection probability change with strategy index. For learning frequency statistics, correlations were significantly different between groups for WM connectivity between right putamen and IFG (training: r = 0.65, CI = [0.44,0.90]; no-training control: r = 0.09, CI = [-0.45,0.53]) and between left vmPFC and caudate (training: r = 0.47, CI = [0.16,0.74]; no-training control: r = -0.16, CI = [-0.62,0.37]). For learning context-based statistics, correlations were significantly different between groups for WM connectivity between left caudate and hippocampus (training: r = -0.69, CI = [-0.86,-0.37]; no-training control: r = 0.16, CI = [-0.25,0.57]). Individual participant data for the training group are indicated by black circles; data for the no-training control group are indicated by gray triangles.