Revisiting the effective connectivity within the distributed cortical network for face perception

Abstract

The classical core system of face perception consists of the occipital face area (OFA), fusiform face area (FFA), and posterior superior temporal sulcus (STS). The functional interaction within this network, more specifically the effective connectivity, was first described by Fairhall and Ishai (2007) using functional magnetic resonance imaging and dynamic causal modeling. They proposed that the core system is hierarchically organized; information is processed in a parallel and predominantly feed-forward fashion from the OFA to downstream regions such as the FFA and STS, with no lateral connectivity, i.e., no connectivity between the two downstream regions (FFA and STS). Over a decade later, we conducted a conceptual replication of their model using four different functional magnetic resonance imaging data sets. The effective connectivity within the core system was assessed with contemporary versions of dynamic causal modeling. The resulting model of the core system of face perception was densely interconnected. Using hierarchical linear modeling, we identified several significant forward, backward, and lateral connections in the core system of face perception across the data sets. Face perception increased the forward connectivity from the OFA to the FFA and OFA to the STS and increased the inhibitory backward connectivity from the FFA to the OFA, as well as the lateral connectivity between the FFA and STS. Emotion perception increased forward connectivity between the OFA and STS and decreased the lateral connectivity between the FFA and STS. Face familiarity did not significantly alter these connections. Our results revise the 2007 model of the core system of face perception. We discuss the potential meaning of the resulting model parameters and propose that our revised model is a suitable working model for further studies assessing the functional interaction within the core system of face perception. Our work further emphasizes the general importance of conceptual replications.

Keywords: Conceptual replication; Dynamic causal modeling; Emotion processing; Face perception; fMRI.

Fig. 1

Dynamic causal model of the interactions within the core system of face perception by Fairhall and Ishai (2007). Driving input (faces) enters the OFA, which propagates the information in a parallel manner toward the FFA and STS. Assumptions about the effect of faces were drawn from the A-matrix. Assumptions about the effects of emotion and fame were drawn from separate B-matrices (see Material and Methods for further information on the terminology of DCM, see Discussion for further information on the modeling strategy).

Fig. 2

Experimental paradigm for study A. In paradigm A, pictures of either neutral, happy, angry, or fearful faces (Langner et al., 2010) were shown in the experimental condition, and houses were shown in the control condition. Single stimuli and blocks were intervened by a gray screen. Participants were instructed to maintain the fixation of their gaze throughout the entire experiment. They were further instructed to press a button if a stimulus was presented twice in a row (one to two times per block). The total experiment lasted about 30 min.

Fig. 3

Model space. Models of the core system of face perception tested with Bayesian model selection (BMS). Connectivity was investigated by modifying the forward, lateral, and feedback connections between the three investigated regions, namely the OFA (blue), FFA (green), and STS (purple). Driving input by faces was set on the OFA (C-matrix, short arrow). All context-independent connections (A-matrix) are displayed with arrows, except the inhibitory self-connections. All interregional connections were modulated (B-matrix) by ‘faces’ (studies A-D), ‘emotion’ (studies A and B), and ‘fame’ (study D). The winning model of the original study FI (#2) and the winning model of our revised model comparisons (#24, see Results section) are marked with dashed rectangles. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4

Bayesian model selection results. Left panel: The posterior model probabilities are displayed. We see that model #24 has the highest relative probability with 0.248 (study C1) to 0.417 (study B). Right panel: The model exceedance probabilities are displayed. In all the data sets, model #24 exhibited a high exceedance probability (>0.9).

Fig. 5

The average connectivity within each study. Studies A, B (upper panels), and C (lower panels) were divided into two scanning sessions. The connectivity between the following three regions is illustrated: the OFA (blue), FFA (green), and STS (purple). In the left panel, the A-matrix (context-independent coupling) is shown. In the middle panel, the driving input (‘faces,’ C-matrix) and B-matrix (‘faces’) are displayed, and in the right panel, the B-matrix (‘emotions’) is shown. Black arrows indicate significant connections (i.e., significant within-study). Gray arrows indicate non-significant connections. The number alongside each arrow indicates the average connection strength. Self-connections (A-matrix) were omitted in the figures but distributed around −0.5 (see Fig. S2). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 6

A revisited model for the core system of face perception. Driving input (‘faces’) enters the OFA. Significant connections, as revealed by HLM, are displayed with black arrows and depicted with numbers. Non-significant (determined by HLM) but present (determined by BMS) connections are illustrated by gray arrows without numbers. The context-independent connections and modulatory effects of ‘faces’ and ‘emotion’ are displayed separately. ‘Fame’ did not significantly modulate any present connection and is therefore not shown. The final model of the original study is depicted in Fig. 1 for comparison.