Fig. 3: Axis-based feature coding in the VTC.

a, b Example channels exhibiting axis coding. a FG. b ITG. (left) Pearson correlation between z-scored HGP and the first PLS component of the feature map (uncorrected). Each dot represents an object image and color coding denotes the object category. The gray line represents the linear fit. (right) Visualization of the encoded feature axis. Note that both channels showed a significant relationship with the feature map (PLS regression, permutation P < 0.001). Images were obtained from Deng J, Dong W, Socher R, Li L-J, Li K, Fei-Fei L (2009). IEEE Conference on Computer Vision and Pattern Recognition 2009: 248–55. IEEE. c Percentage of channels exhibiting axis coding. Dark colors represent an above-chance number of selected channels in the corresponding ROIs (binomial test: P < 0.05; Bonferroni correction), while light colors indicate chance-level selection. d Distribution of axis-coding channels in the VTC and MTL. Color coding shows the strength of axis coding (see Methods). e Strength of axis coding averaged across all visually responsive channels in each ROI (FG: n = 201; ITG: n = 141; PH: n = 34; AH: n = 47; amygdala: n = 78). Dark colors represent significant ROIs (right-tailed one-sample t-test against 0: P < 0.05). A two-tailed two-sample t-test was performed to compare the strength of axis coding between the left and right hemispheres. f Temporal profile of axis coding among the significant ROIs from the left hemisphere. Shaded area denotes ±SEM across axis-coding channels. Top bars indicate time points showing significant axis coding (two-tailed one-sample t-test against the mean pre-trial baseline: P < 0.05, corrected by FDR⁹¹) for each ROI. g Pearson correlation between the response onset latency and MNI y-coordinates of axis-coding channels in the FG (n = 85). Each dot represents a channel, and the gray line represents the linear fit. To control for individual differences in response latency, the onset latency for each channel was normalized by subtracting the latency of the channel with the highest axis-coding strength in each session. h Representational similarity between axis-coding channels in the VTC and category-selective channels in the MTL. Statistical difference was determined using a permutation test (see Methods). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Source data are provided as a Source Data file.