Memory success effects for each task. A, In the explicit memory task, hits elicited greater activity than CRs (Memory Success Effects) in various regions, including those implicated previously in successful memory retrieval. B, In the concealed memory task, memory success effects (warm colors) were found in a similar region in the IPS, whereas the reverse comparison (CRs > hits; cool colors) revealed activity in the left AnG and bilateral hippocampus. C, Memory success effects overlapped in the two tasks in several regions, including the IPS and VTC. Activity is rendered on 3D Caret inflated brain and 2D mean across-subject brain, both in standardized MNI space; height and extent thresholds: p < 0.001, k = 16. hip, Hippocampus; mIPS/SPL, medial IPS/SPL.
Memory success effects are modulated by attempts to conceal memory. A, Memory success effects reverse between the explicit and concealed memory tasks in regions associated with memory retrieval, suggesting that participants were able to successfully deploy countermeasures. B, Memory success effects were greater in the concealed versus explicit memory task in the right IPS. Activity rendered as described in Figure 1. Graphs depict mean univariate β weights across participants (errors indicate SEM) for clusters in the left hippocampus, AnG, and IPS (B).
Memory strength in the concealed memory task predicts whether participants can modulate retrieval success effects in the left AnG. Clusterwise β values were extracted from a left AnG (L AnG) cluster identified in the univariate task × memory interaction (Fig. 2A), and d′ in the concealed memory task was regressed against β values in the concealed memory task (hits > CRs). This regression reveals that participants with superior memory performance in the concealed memory task were less likely to show inverse retrieval success effects (CRs > hits) in the AnG, suggesting that participants with stronger memories were less able to exert goal-directed control over memory-related activity in the left AnG.
Memory decoding trained on the explicit recognition memory task. A, Performance of memory classifiers (hits vs CRs) trained on standard recognition memory data and tested on held-out trials from the same condition (explicit memory trials, EX→EX; blue) or on trials when they were attempting to conceal their true memory state (concealed memory trials, EX→CM; green). Violin plots depict the mean AUC for ROC curves (for description of this metric, see text). Plots demonstrate that memory decoding was above chance during a standard recognition memory test (EX→EX, blue) but reduced to chance levels when participants attempted to conceal memory (EX→CM, green). AUC values are plotted for each participant's data using unique color identifiers, with lines connecting each participant's classification performance, and plot-width depicting participant density in that performance range. B, Classifiers trained and tested on each time point (2 s volume acquisition, TR) revealed similar results as in A, with above-chance decoding on standard memory trials (blue) and chance decoding on concealed memory trials (time point 4–6 s was slightly above chance). C, Maps indicating the diagnostic value that each voxel provides the classifier (importance maps) in biasing a hit choice (warm colors) or a CR choice (cool colors). Error bars in B indicate SEM. EX, Explicit memory task; CM, concealed memory task; EX→EX, classifier trained on explicit trials, tested on held-out explicit trials (4-fold cross-validation); EX→CM, classifier trained on explicit trials, tested on concealed trials.
Classification by confidence, memory strength, and time point. A, Classifier accuracy at different levels of confidence, ranging from all trials (top 100%) to the top 10% most confident classifier responses. Values in blue represent classifier performance from the whole-brain explicit → explicit classifier (EX→EX) for late TRs (TR3–TR5), which reveal better performance than the explicit → concealed classifier (EX→CM) for late TRs (green circles), as well as the explicit → concealed classifier for TR3 (green triangles). B, Individual differences in classification performance on TR3 for concealed trials correlated with memory strength (d′) in the concealed trials suggests that the classifier may have detected transient memory signals that emerged before participants' attempts to conceal the signals. C, Classifier trained on memory patterns from TR3 in the explicit task detect novelty patterns on TR5 and TR6 in the concealed condition (green), suggesting that feigned memory can indeed fool a classifier into reliably predicting the opposite response, at later TRs (i.e., when countermeasures have been deployed, after memory state has been determined). EX→EX, Classifier trained on explicit trials, tested on held-out explicit trials (4-fold cross-validation); EX→CM, classifier trained on explicit trials, tested on concealed trials.
Searchlight analysis of memory decoding during use of countermeasures. Maps of group-average searchlight accuracy indicated voxels that exhibited above- or below-chance accuracy (warm and cool colors, respectively) on the critical explicit→concealed classification of hits versus CRs. Searchlights were performed separately on each time point, revealing local information that demonstrated above-chance decoding in medial frontoparietal regions at early time points (0–4 s) when participants were presumably making memory decisions and below-chance decoding at later time points (6–8 s), including the left AnG (circled), when participants were presumably attempting to conceal memory state. Maps are thresholded at p < 0.05 (corrected for multiple comparisons, FWE). Clusters outlined in black additionally survive Bonferroni's correction for the six temporal comparisons. Color map represents AUC values (with 0.5 representing chance).
Strategic goal states associated with countermeasures. A, Left, Univariate contrast reveals brain regions engaged during attempts to conceal memory by shifting attention to novel perceptual aspects of the face stimuli. Right, Complementary contrast reveals brain regions involved in concealing novelty, by generating memories in response to novel faces. B, Plots of memory decoding for classifiers trained on patterns from concealed memory trials. Green plots reveal above-chance decoding when tested on concealed trials (CM→CM, green) but reduced to chance levels when tested on explicit recognition data (EX→CM, blue). AUC values are plotted for each participant's data using unique color identifiers, with lines connecting each participant's classification performance and plot-width depicting participant density in that performance range. C, Maps indicating the diagnostic value that each voxel provides the concealed memory-trained classifier (importance maps) in biasing a hit choice (warm colors) or a CR choice (cool colors). CM→CM, Classifier trained on concealed trials, tested on held-out concealed trials (4-fold cross-validation); CM→EX, classifier trained on concealed trials, tested on explicit trials.