Article Figures & Data
Figures
- Figure 1.
Experimental paradigm. Each trial started with one of the three visual cues (200 ms), two of them instructing subjects to covertly attend either left (A) or right (B) hemifields, indicated by the two white dots. A third cue, the choice cue (C), prompted the participants to choose either left or right visual field to attend. Following a variable cue-to-target delay (2,000–8,000 ms), a target in the form of a grating patch appeared briefly for 100 ms with equal probability in the left or the right hemifield. Participants were asked to discriminate the spatial frequency of the grating displayed in the cued or chosen location and ignore the uncued or unchosen side. After a second variable ISI (2,000–8,000 ms), participants were presented with a “?SIDE?” screen and were asked to report the hemifield they paid attention to in that trial. An ITI, varied randomly from 2,000 to 8,000 ms, elapsed before the start of the next trial.
- Figure 2.
Behavioral results from instructed and choice trials. A, C, Comparison of accuracy (ratio of correct trials and total number of valid trials) between different attention conditions for UF (A) and UCD (C) datasets. B, D, Comparison of RT between different attention conditions for UF dataset (B) and UCD dataset (D).
- Figure 3.
Whole-brain GLM analysis of cue-evoked BOLD activity. Attention cues (left and right combined) evoked significant activation in DAN (FEF, IPS/SPL) and the extrastriate visual cortex (p < 0.05 FDR corrected) for both UF (A and B) and UCD (C and D) datasets. However, contrasting attention-directing cues (cue-left vs cue-right) revealed no activation when thresholded at p < 0.05 with FDR correction (data not shown). High-order visual regions only began to appear after lowering the threshold to p < 0.01 uncorrected: cue-left > cue-right (E and F) and cue-right > cue-left (G and H).
- Figure 4.
ROI-level analysis of cue-evoked BOLD activity. BOLD activity difference between attention to contralateral and ipsilateral visual fields were computed and averaged across the hemispheres for instructional (A–D) and choice trials (E–H) for both datasets: UF (A, E) and UCD (B, F). C, G, Activation differences from UF and UCD datasets were averaged from instructed (C) and choice trials (G) and the respective p values obtained via meta-analysis. D, H, correlation between behavior and activation differences across ROIs from instructed (D) and choice trials (H). Here the correlation coefficients are averaged across the datasets, and the p values from both the datasets were combined using meta-analysis. The error bars denote the standard error of the mean (SEM). *p < 0.05 FDR.
- Figure 5.
MVPA of cue-left versus cue-right: instructed trials. A, B, Decoding accuracies in different visual ROIs for UF dataset (A) and UCD dataset (B). C, D, Posterior view of color-coded decoding accuracies for UF dataset (C) and UCD dataset (D). E, Scatterplot comparing UCD versus UF decoding accuracies where each dot represents a visual ROI. The error bars denote the SEM. *p < 0.05 FDR.
- Figure 6.
Decoding accuracy versus behavior: instructed trials. A, B, The percentage of variance explained by the principal components of decoding accuracies across ROIs for (A) UF and (B) UCD datasets. C, D, Decoding accuracy represented by score on first PCA component versus behavioral efficiency (z score) for (C) UF and (D) UCD datasets. E, Correlation between decoding accuracy and behavior across ROIs. Here the correlation coefficients are averaged across the two datasets, and the p values from both the datasets were combined using meta-analysis; the correlation is significant for all ROIs at p < 0.05 FDR.
- Figure 7.
Decoding accuracy for attend-left versus attend-right for choice trials. A, Decoding accuracies averaged from UF and UCD datasets for different visual ROIs; p values were determined via meta-analysis. B, Scatterplot comparing UCD versus UF decoding accuracies where each dot represents a visual ROI. C, D, Decoding accuracy versus behavioral efficiency for UF dataset (C) and for UCD dataset (D). E, Correlation between decoding accuracy and behavior across visual-topic ROIs. The error bars denote the SEM. *p < 0.05 FDR.
- Figure 8.
Comparison between choice trials and instructed trials (UF and UCD datasets combined). A, Choice-left versus choice-right decoding accuracies plotted against cue-left versus cue-right decoding accuracies; each dot represents a visual ROI. B, Cross-decoding accuracy comparison across ROIs where classifiers were trained on instructed trials and tested on choice trials. C, Difference between cross-decoding accuracy (classifiers trained on instructed trials and tested on choice trials) and self-decoding accuracy (classifiers trained on choice trials and tested on choice trials) for different ROIs. The error bars denote the SEM. *p < 0.05 FDR.
In this issue
