Differential destinations, dynamics, and functions of high- and low-order features in the feedback signal during object processing
- State Key Laboratory of Cognitive Science and Mental Health, Institute of Biophysics, Chinese Academy of Sciences, China
- University of Chinese Academy of Sciences, China
- Institute of AI, Hefei Comprehensive National Science Center, China
Figures
Figure 1 with 3 supplements
Feedback information to foveal V1 in the peripheral object task (n=18).
(A) Four types of objects (two orientations×two object categories) were used in the experiment. (B) Two objects from the same type were presented in the periphery, and participants had to discriminate whether they were identical. (C) Strong fMRI responses were found in peripheral V1 corresponding to the stimulus locations and high-level regions, but not in foveal V1. (D) Schematic representation of feedforward and feedback connections in different cortical layers of V1. (E) High-order category information from feedback signals could be decoded in foveal, but not peripheral, V1. (F) The category information could be decoded in superficial and deep layers of foveal V1, but absent in all layers of peripheral V1. (G–H) Low-order orientation information from feedback signals could be decoded only in the deep layer of foveal V1 and across layers in peripheral V1. Error bars reflect ±1 SEM. * indicates paired t-test with significance of p<0.05. ** indicates paired t-test with significance of p<0.01. † indicates marginal significance. The p-values shown in this figure are uncorrected for multiple comparisons. The dashed line represents the chance level of decoding performance. See also Figure 1—figure supplement 3.
Figure 1—figure supplement 1
Activation maps of the foveal stimulus and the sizes of the foveal V1 region of interest (ROI) in two example participants.
(A) Voxels showing stronger fMRI responses to the foveal checkerboard than to the peripheral checkerboard. (B) Voxels showing stronger fMRI responses to foveal objects than to the baseline. Blue solid lines indicate the spatial range of the foveal V1 ROI for each participant.
Figure 1—figure supplement 2
Example registration of functional images to anatomical space in four participants.
The green and blue lines represent the white matter and pial surfaces, respectively, both derived from the anatomical image. Transparent colored overlays mark the regions of interest (ROIs).
Figure 1—figure supplement 3
Fine-scale laminar profiles of orientation and category information in foveal V1 in peripheral object task.
Significant category information was found in the deep and superficial layers (ts>1.94, ps<0.035), and significant orientation information was found in the deep layers (ts>2.26, ps<0.019), but no significant difference was observed between two types of information (ts<0.61, ps>0.55). Error bars reflect ±1 SEM. * indicates paired t-test with significance of p<0.05. The dashed line represents the chance level of decoding performance.
Figure 2
Object representation in foveal V1 during a fixation task (n=18).
(A) The object stimulus was presented at the fovea as a task-irrelevant stimulus and participants performed a one-back task to the fixation color. (B–C) Category information could not be decoded in foveal V1. (D–E) Orientation information could be decoded in all layers of foveal V1. Error bars reflect ±1 SEM. * indicates paired t-test with significance of p<0.05. ** indicates paired t-test with significance of p<0.01. The p-values shown in this figure are uncorrected. The dashed line represents the chance level of decoding performance.
Figure 3 with 1 supplement
Temporal dynamics of object representation in different cortical regions (n=15).
(A) In the MEG experiment, two objects of the same type were presented in the periphery. The locations of the objects changed across trials. (B–C) Using the source localization algorithm, the neural responses of three cortical regions were extracted. In all regions, both category and orientation information could be decoded either within the same location set (B) or across different location sets (C). (D) Same-location and cross-location decoding performances were normalized and plotted together to facilitate their comparison. The dynamics of decoding performance were significantly delayed for cross-location decoding in early visual cortex (solid red and blue lines). The dashed black line represents the chance level of decoding performance. The gray bar on the time axis indicates the presentation time of the object stimuli. Colored shaded areas reflect ±1 SEM. The colored bars below the time courses indicate the significance (cluster-forming threshold p<0.01, corrected significance threshold p<0.01) of the decoding accuracy in a cluster permutation test. See also Figure 3—figure supplement 1.
Figure 3—figure supplement 1
Latency for category (A) and orientation (B) information in three cortical regions during peripheral object recognition task.
Time courses of decoding accuracy were smoothed with a Gaussian kernel with a half-width of 150 ms, and latency was estimated as the time from stimulus onset to 75% of peak decoding performance. When calculating the group mean of latency, data exceeding 2 SDs were excluded. The latency of cross-location decoding for category information was significantly longer in early visual cortex than in occipitotemporal cortex (t(11)=2.86, p=0.03), and the effect was marginally significant for orientation information (t(12)=2.52, p=0.06), indicating the decoding performances were driven by feedback signals. In same-location decoding, the latency of category information was similar between early visual cortex and occipitotemporal cortex, and the latency of orientation information is much shorter in early visual cortex (t(13)=4.01, p=0.001). * indicates paired t-test with significance of p<0.05. ** indicates paired t-test with significance of p<0.01. † indicates marginal significance (Holm-Bonferroni corrected).
Figure 4
Information transmission between cortical regions revealed by Granger causality analysis (n=15).
(A) The strength of representation was estimated by the distance from the neural response pattern to the decoding hyperplane. The dynamics of representation strength were used to estimate information transmission between cortical regions. (B–C) The Granger causality of feedforward (blue) and feedback (orange) category (B) and orientation (C) information between three cortical regions during peripheral object discrimination task. (D–E) The Granger causality of feedforward and feedback information of category (D) and orientation (E) during a fixation detection task with an object presented at the fovea. The directions and initial timings of information flow between cortical regions were also indicated by arrows and the onset times next to the arrows. The gray bar on the time axis indicates the presentation time of the object stimuli. The colored shaded areas reflect ±1 SEM. The colored bars below the time courses indicate the significance (cluster-forming threshold p<0.01, corrected significance threshold p<0.01) of Granger causality in a cluster permutation test.
Figure 5
Behavioral relevance of feedback information in the visual cortex (n=15).
(A) The strength of information representation at each time point of each trial was estimated by cross-location decoding, its correlation with reaction time across trials was calculated to estimate its behavioral relevance. (B) Significant behavioral relevance was observed for category information in early visual cortex between 200 and 400 ms after stimulus onset, consistent with the time window of feedback signals. Colored shaded areas reflect ±1 SEM. The colored bars below the time courses indicate the significance (cluster-forming threshold p<0.05, corrected significance threshold p<0.05) of the correlation in a cluster permutation test. (C) Schematic summary depiction of information flow among key cortical regions (top) and different types of feedback information from temporal cortex to different cortical layers of V1 (bottom).
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Differential destinations, dynamics, and functions of high- and low-order features in the feedback signal during object processing
eLife 13:RP103788.
https://doi.org/10.7554/eLife.103788.3
