Ultra-Rapid serial visual presentation reveals dynamics of feedforward and feedback processes in the ventral visual pathway

  1. Yalda Mohsenzadeh
  2. Sheng Qin
  3. Radoslaw M Cichy
  4. Dimitrios Pantazis  Is a corresponding author
  1. Massachusetts Institute of Technology, United States
  2. Freie Universität Berlin, Germany
6 figures, 2 tables and 2 additional files

Figures

Rapid serial visual presentation (RSVP) task.

(a) Experimental procedure. The stimulus set comprised 12 face targets, 12 object targets, and 45 masks of various objects. Participants viewed a RSVP sequence of 11 images, with the middle image …

https://doi.org/10.7554/eLife.36329.003
Decoding of target images from MEG signals.

(a) Multivariate pattern analysis of MEG signals. A support vector machine (SVM) classifier learned to discriminate pairs of target images using MEG data at time point t. The decoding accuracies …

https://doi.org/10.7554/eLife.36329.004
Figure 3 with 1 supplement
Categorical information encoded in MEG signals.

(a) Time course of categorical division depending on presentation rate. For each condition, the MEG decoding matrix was divided into 3 segments for pairs of within-face, within-object, and …

https://doi.org/10.7554/eLife.36329.006
Figure 3—Figure supplement 1
Linear decoding of faces vs. objects category.

For each condition and at each time point a SVM classifier was trained to decode object versus face trials with a leave-one-out procedure. The shape and peaks of the time series largely match the …

https://doi.org/10.7554/eLife.36329.007
Temporal generalization of target image decoding for the 500, 34, and 17 ms per picture conditions.

The SVM classifier was trained with MEG data from a given time point t (training time) and tested on all other time points (testing time). The temporal generalization decoding matrix was averaged …

https://doi.org/10.7554/eLife.36329.008
Representational similarity of MEG to fMRI signals at EVC and IT.

(a) For every time point t, the EVC-specific MEG RDM was compared (Spearman’s rho) with the EVC-specific fMRI RDM, yielding a time series of MEG-fMRI representational similarity at EVC. (b) Same as …

https://doi.org/10.7554/eLife.36329.010
Author response image 1
Decoding seen versus unseen faces in the 17ms per picture condition.
https://doi.org/10.7554/eLife.36329.014

Tables

Table 1
Peak and onset latency of the time series for single image decoding (Figure 2) and categorical division decoding (Figure 3), with 95% confidence intervals in brackets.
https://doi.org/10.7554/eLife.36329.005
Presentation ratePeak latency (ms)Onset latency (ms)
Grand total500 ms per picture121 (102–126)28 (9–53)
34 ms per picture100 (96–107)64 (58–69)
17 ms per picture96 (93–99)70 (63–76)
Within-faces500 ms per picture113 (104–119)59 (30–73)
34 ms per picture104 (96–109)74 (62–81)
17 ms per picture98 (96–104)86 (78–117)
Within-objects500 ms per picture102 (93–102)48 (10–55)
34 ms per picture97 (90–97)60 (27–67)
17 ms per picture94 (87–95)70 (64–74)
Between minus
within
500 ms per picture136 (130–139)46 (15–51)
34 ms per picture169 (165–177)73 (67–78)
17 ms per picture197 (175–218)139 (67–155)
Table 2
Peak and onset latency of the time series for MEG-fMRI fusion at EVC and IT, with 95% confidence intervals in brackets.
https://doi.org/10.7554/eLife.36329.009
Peak latency (ms)Onset latency (ms)
500 ms per pictureIT131 (127–135)63 (50–75)
EVC104 (93–120)53 (50–57)
34 ms per pictureIT166 (162–173)70 (50–88)
EVC87 (83–97) and
169 (164-176)*
71 (61–81)
17 ms per pictureIT195 (170–203)162 (55–183)
EVC80 (75–100) and
202 (190-219)*
71 (50–76)
  1. *Time series had two early peaks.

Additional files

Source data 1

Figures 2, 3 and 5 source data and code.

Decoding target images; resolving categorical information; and computing MEG-fMRI representational similarities.

https://doi.org/10.7554/eLife.36329.011
Transparent reporting form
https://doi.org/10.7554/eLife.36329.012

Download links