Fig. 3: Imaging analysis results and decoding accuracies.

a Brain areas that were significantly activated when subjects were predicting an upcoming scene: bilateral superior parietal lobule (SPL), bilateral inferior parietal lobule (IPL), left dorsal premotor cortex (PMd), and left anterior prefrontal cortex (aPFC). The voxel activity patterns in these four areas were used to construct decoders for each region of interest (ROI) for scene prediction and confidence level. Visualization was performed using xjView toolbox (https://www.alivelearn.net/xjview). b, c Decoding accuracies for scene prediction (b, six types of scenes) and its confidence level (c, high or low) within each ROI evaluated using leave-one-session-out (LOSO) cross-validation (CV). Each box extends from the lower to upper quartiles, with a horizontal line at the median. The whiskers represent 1.5 × IQR, and cross markers indicate the outliers. Significance was tested using a one-sided Wilcoxon signed-rank test compared to chance (dashed line) (**: p < 0.01, ***: p < 0.001). These figures represent the results of the scene prediction and confidence decoders using the 6th decoding period as representative data because of its increasing decodability in our time-series analysis; the overall results of the time-series decoding analysis are shown in Supplementary Fig. 4b–e. d, e Time-series scene prediction decoding results within each ROI when the data were categorized binarily according to the prediction correctness (d, correct vs. incorrect trials) and the confidence level (e, high-confidence vs. low-confidence trials) of the prediction trial. For example, ‘correct’ indicates the accuracy of the scene prediction decoder trained and tested with only the trials in which subjects’ upcoming scene selections were correct (correct-only decoder). The decoding accuracies were evaluated using the leave-one-game-out (LOGO) CV. The solid lines reflect the median, the shaded areas indicate the range between the upper and lower quartiles, and the dotted lines indicate the range of 1.5×IQR. The cross-markers indicate outliers. Significance was tested using a one-sided Wilcoxon signed-rank test (unfilled circle: p < 0.05, unfilled square: p < 0.01, unfilled diamond: p < 0.001) compared to the chance level (dashed line). The color of the horizontal line below the plots reflects a significant difference between the two categories of trials in each decoding period (one-sided Wilcoxon rank-sum test, light gray: p < 0.05, dim gray: p < 0.01, black: p < 0.001).