Erratum: High-resolution imaging reveals highly selective nonface clusters in the fusiform face area

Nature Neuroscience 9, 1177–1185 (2006); Published online 6 August 2006

Baker et al.¹ have drawn to our attention flaws in our analysis of voxel selectivity arising because noise can generate similar results and because this analysis excluded voxels whose amplitudes were negatively correlated across runs, which the Methods failed to state. In addition, Simmons et al.² note that our selectivity index incorrectly assigns high values to voxels with low or negative activations. A re-analysis of our data to address these concerns shows that our assertion that many voxels in the fusiform face area (FFA) show high selectivity to nonface objects was incorrect (Fig. 1). In the re-analysis, we used the standard general linear model³ to identify active voxels and cross-validated the results (Supplementary Methods online). Owing to these errors, Figures 4 and 8 in the original paper are not valid. The other conclusion remains correct: the fusiform face area is heterogeneous, in that regions of high selectivity for faces are intermingled with regions of lower selectivity (Fig. 1b). These flaws do not affect reproducibility analyses that included all FFA voxels (Fig. 3 and Supplementary Fig. 2), principal component analysis (PCA; Fig. 6), pattern analyses (Supplementary Fig. 5) or comparison of high-resolution and standard-resolution fMRI data (Fig. 7), which remain valid. We sincerely regret these errors.

Figure 1: Re-analysis of FFA selectivity.

Re-analyses were performed on 1-mm³ voxels in the FFA that were activated significantly more to objects than to the scrambled baseline (defined by a voxel-by-voxel general linear model (GLM), intact > scrambled, P < 0.001, no spatial smoothing). Re-analysis of average category responses with cross validation. (a) Voxels were sorted by the category that elicited the maximal response in odd numbered scans, and the average response amplitudes are plotted for even numbered scans. The rightmost column reflects the average response across all activated 1-mm³ FFA voxels. Percentage signal is measured relative to scrambled baseline. Error bars indicate s.e.m. across five subjects from published sessions. All amplitudes were significantly different from the scrambled baseline (t-test across subjects, P < 0.05), except for the bar marked by the asterisk. Stimuli: F, faces; A, animals; C, cars; S, sculptures. (b) Re-analysis of individual voxel selectivity using the signal detection theory measure d′, with cross validation.

where μ and σ indicate the mean and standard deviation of responses. The preferred category for each voxel was determined as the category that yielded maximal response during odd numbered scans. d′ was calculated relative to this category using data from even numbered scans.

Note: Supplementary information is available on the Nature Neuroscience website.