The bottom-up and top-down processing of faces in the human occipitotemporal cortex
Figures

Face-selective areas identified by fMRI localizer and face-evoked MEG source activation displayed on an inflated right hemisphere of a typical subject.
(A) Face-selective statistical map (faces>objects) showing four face-selective regions (rOFA, rpFFA, raFFA and rpSTS). (B) Face-evoked MEG source activation patterns represented as LCMV value maps at different time points (120-160 ms) after the stimulus onset. LCMV values represent signal power normalized by noise.

Face-selective areas identified by fMRI localizer and face-evoked MEG source activation displayed on an inflated right hemisphere of typical subject one.
(Left) Face-selective statistical map (faces>objects) showing face-selective regions (OFA, pFFA, aFFA and pSTS). (Right) Face-evoked MEG source activation patterns represented as LCMV value maps at different time points (120-160 ms) after the stimulus onset. LCMV values represent signal power normalized by noise. Results of these four subjects were shown here because of their relatively clear fMRI defined face-selective areas.

Face-selective areas identified by fMRI localizer and face-evoked MEG source activation displayed on an inflated right hemisphere of typical subject two.
(Left) Face-selective statistical map (faces>objects) showing face-selective regions (OFA, pFFA, aFFA and pSTS). (Right) Face-evoked MEG source activation patterns represented as LCMV value maps at different time points (120-160 ms) after the stimulus onset. LCMV values represent signal power normalized by noise. Results of these four subjects were shown here because of their relatively clear fMRI defined face-selective areas.

Face-selective areas identified by fMRI localizer and face-evoked MEG source activation displayed on an inflated right hemisphere of typical subject three.
(Left) Face-selective statistical map (faces>objects) showing face-selective regions (OFA, pFFA, aFFA and pSTS). (Right) Face-evoked MEG source activation patterns represented as LCMV value maps at different time points (120-160 ms) after the stimulus onset. LCMV values represent signal power normalized by noise. Results of these four subjects were shown here because of their relatively clear fMRI defined face-selective areas.

Face-selective areas identified by fMRI localizer and face-evoked MEG source activation displayed on an inflated right hemisphere of typical subject four.
(Left) Face-selective statistical map (faces>objects) showing face-selective regions (OFA, pFFA, aFFA and pSTS). (Right) Face-evoked MEG source activation patterns represented as LCMV value maps at different time points (120-160 ms) after the stimulus onset. LCMV values represent signal power normalized by noise. Results of these four subjects were shown here because of their relatively clear fMRI defined face-selective areas.

Temporal response characteristics of face-selective ROIs.
(A) The time courses of face (solid line) and object (dotted line) induced responses averaged across subjects, for the seven face-selective ROIs. Shaded area means SEM. The green bar indicates significant difference between face and object. Significance was assessed by cluster-based permutation test (cluster-defining threshold p<0.05, significance level p<0.05) for each ROI. (B) The peak latency averaged across subjects for each ROI (mean ± SEM). The peak latency of raFFA is significantly later than rpFFA (t11 = 3.21, p=0.025, Bonferroni corrected) (C) The mean peak latencies for the face-selective ROIs were shown on inflated cortical surfaces of both hemispheres at corresponding locations.

Temporal response characteristics of face-selective ROIs for unfamiliar faces.
(A) The time courses of unfamiliar face (solid line) and object (dotted line) induced responses averaged across subjects, for the seven face-selective ROIs. Shaded area means SEM. (B) The peak latency averaged across subjects for each ROI (mean± SEM).

Temporal response characteristics and granger causality analysis for face-selective ROIs during perception of Mooney and normal faces.
(A) Normal and Mooney face images. (B) The peak latency averaged across subjects for each face-selective ROI (mean ± SEM). Mooney faces elicited a response with significantly longer latency in rOFA than normal faces (paired t test, t23 = 4.009, p=0.001). (C) Time courses averaged across subjects for bilateral OFA and pFFA. Gray line is OFA and red line is pFFA. Shaded areas denote SEM. The circles above time courses represent peak latencies of individual subjects. rOFA was engaged significantly later than rpFFA when processing Mooney faces (Paired permutation test p=0.02. Bonferroni corrected). (D) Granger causality analysis performed within a series of 50 ms time windows. Arrows represent statistically significant causal effects (p<0.05, FDR corrected, F test. See Materials and methods for details).

Temporal response characteristics for face-selective ROIs in response to distorted face.
(A) Example stimuli and averaged time courses for each face-selective ROI. The green horizontal bar indicates significant difference between distorted face and object (cluster-defining threshold p<0.01, corrected significance level p<0.05). (B) Peak latency averaged across subjects for each ROI. The peak latency of raFFA is significant later than rpFFA (paired t test, p=0.019, t8 = 2.92).

Temporal response characteristics for face-selective ROIs in response to contextual cues.
(A) Example stimuli. (B) Time courses averaged across subjects for each condition. For each ROI, Blue horizontal bars indicate significant difference between degraded faces with relevant body cues and objects, and red horizontal bars indicate significant difference between degraded faces with irrelevant body cues and objects (cluster-defining threshold p < 0.05, corrected significance level p < 0.05). (C) The peak latency averaged across subjects for each face-selective ROI (mean± SEM).
Videos
MEG activation of a typical subject.
Tables
Number of subjects showing fMRI defined face-selective areas.
Experiment 1 | Experiment 2 | Experiment 3 | Experiment 4 | ||
---|---|---|---|---|---|
famous face | Unfamiliar face | ||||
IOFA | 13/13 | 9/9 | 25/26 | 9/9 | 13/14 |
IpFFA | 13/13 | 9/9 | 26/26 | 9/9 | 14/14 |
IpSTS | 13/13 | 9/9 | 18/26 | 9/9 | 11/14 |
rOFA | 13/13 | 9/9 | 26/26 | 9/9 | 14/14 |
rpFFA | 13/13 | 9/9 | 26/26 | 9/9 | 14/14 |
raFFA | 12/13 | 9/9 | 18/26 | 9/9 | 12/14 |
rpSTS | 13/13 | 9/9 | 23/26 | 9/9 | 14/14 |
Number of subjects used in peak latency analysis.
Experiment 1 | Experiment 2 | Experiment 3 | Experiment 4 | |||
---|---|---|---|---|---|---|
famous face | unfamiliar face | normal face | Mooney face | distorted face | contaxtual cues defined face | |
IOFA | 13/13 | 9/9 | 24/26 | 24/26 | 9/9 | 13/14 |
IpFFA | 12/13 | 9/9 | 25/26 | 25/26 | 9/9 | 12/14 |
IpSTS | 11/13 | 8/9 | - | - | - | |
rOFA | 13/13 | 9/9 | 24/26 | 26/26 | 9/9 | 13/14 |
rpFFA | 13/13 | 9/9 | 24/26 | 25/26 | 9/9 | 13/14 |
raFFA | 12/13 | 8/9 | 18/26 | 15/26 | 9/9 | 10/14 |
rpSTS | 12/13 | 7/9 | - | - | - | - |
Additional files
-
Source code 1
Preprocessing.
- https://cdn.elifesciences.org/articles/48764/elife-48764-code1-v2.py
-
Source code 2
Source localization.
- https://cdn.elifesciences.org/articles/48764/elife-48764-code2-v2.py
-
Source code 3
Extract timecourse.
- https://cdn.elifesciences.org/articles/48764/elife-48764-code3-v2.py
-
Transparent reporting form
- https://cdn.elifesciences.org/articles/48764/elife-48764-transrepform-v2.docx