Amodal completion across the brain: The impact of structure and knowledge

Abstract

This study investigates the phenomenon of amodal completion within the context of naturalistic objects, employing a repetition suppression paradigm to disentangle the influence of structure and knowledge cues on how objects are completed. The research focuses on early visual cortex (EVC) and lateral occipital complex (LOC), shedding light on how these brain regions respond to different completion scenarios. In LOC, we observed suppressed responses to structure and knowledge-compatible stimuli, providing evidence that both cues influence neural processing in higher-level visual areas. However, in EVC, we did not find evidence for differential responses to completions compatible or incompatible with either structural or knowledge-based expectations. Together, our findings suggest that the interplay between structure and knowledge cues in amodal completion predominantly impacts higher-level visual processing, with less pronounced effects on the early visual cortex. This study contributes to our understanding of the complex mechanisms underlying visual perception and highlights the distinct roles played by different brain regions in amodal completion.

Figure 1.

Convergent and divergent stimuli. (A) In this study, there were two types of partially occluded stimuli. For convergent stimuli (A1, e.g., here a banana), compatibility with structural cues and knowledge align to yield the same inferred shape. Conversely, for divergent stimuli (A2, e.g., here an apple), compatibility with structural cues and knowledge lead to different interpretations, resulting in distinct inferred shapes. (B) Specifically, completions may exhibit compatibility (S+), as shown in the left column, or incompatibility (S−) with the underlying structural cues, depicted in the right column. Furthermore, these completions may either align with knowledge (K+), as seen in the upper row, or clash with it (K−), as illustrated in the lower row. Adapted from Hazenberg and van Lier (2016).

Figure 2.

Fruit and vegetable stimuli. Ten different fruits and vegetables were used as stimuli, of which five were convergent (A) and five were divergent (B). Convergent stimuli were a banana, carrot, zucchini, cucumber, and leek and were those of which the occlusion stimulus (A1) could be completed either compatible with both structure as well as knowledge (A2) or both incompatible with structure and knowledge (A3). Divergent stimuli were an apple, kiwi, lemon, orange, and tomato and were those of which the occlusion (B1) could be completed either compatible with structure but incompatible with knowledge (B2) or incompatible with structure but compatible with knowledge (B3). Adapted from Hazenberg and van Lier (2016).

Figure 3.

Experimental conditions and timeline. In the main task of the experiment, the trials consisted of pairs of stimuli featuring a single kind of fruit or vegetable. Trials presented a pair of convergent (A) or divergent (B) stimuli that involved an (1) occlusion, (2) repetition, or (3) alternation. This resulted in a total of 12 conditions. Each of these conditions was presented in one of four configurations, accounting for all possible horizontally and vertically mirrored versions. Finally, for each of two stimulus types (convergent or divergent), there were five different fruits or vegetables; see Figure 2. Importantly, the occlusion trials feature the four main conditions of the experiment, including all combinations of compatibility (+) or incompatibility (−) with the structure (S) and knowledge (K) cues. All trials presented an initial stimulus for 500 ms, following an interstimulus interval (ISI) of 200 ms, a second stimulus for 500 ms, and finally an intertrial interval (ITI) of on average 3,000 ms.

Figure 4.

Functional ROI masks. The functional region of interest (ROI) masks for early visual cortex (EVC) (A) and lateral occipital complex (LOC) (B) as summed over participants, indicating for how many participants a voxel was contained in the mask. This visualization demonstrates that overall, there was an agreement of the location of the ROIs across participants, the EVC mask contains more voxels than the LOC mask, and the LOC mask seems to show a left-hemispheric dominance.

Figure 5.

ROI analysis repetition suppression. The graph presents the grand average parameter estimates and their corresponding standard deviations for both stimulus types, convergent (blue) and divergent (orange), and both trial types, repetition and alternation. The analysis was performed in two regions of interest (ROIs): the early visual cortex (EVC, left) and the lateral occipital complex (LOC, right). In both the EVC and the LOC, a main effect of trial type was found, with responses in repetition trials to be significantly suppressed compared to those in alternation trials. This suggests that presenting the same stimulus consecutively led to a reduced neural response compared to when stimuli were presented in an alternating manner, the repetition suppression effect. Furthermore, across both ROIs, a main effect of stimulus type was found, with divergent stimuli eliciting significantly smaller neural responses than convergent stimuli, regardless of trial type. ***p < 0.001.

Figure 6.

ROI analysis occlusion. The graph shows the grand average parameter estimates and their corresponding standard deviations for occlusion trials categorized as structure compatible (S+, blue) and structure incompatible (S−, orange), as well as knowledge compatible (K+) and knowledge incompatible (K−) completions. Two regions of interest (ROIs) were analyzed: the early visual cortex (EVC, left) and the lateral occipital complex (LOC, right). In EVC, none of the observed differences between the conditions were found to be statistically significant. In LOC, responses were significantly suppressed in response to knowledge compatible completions as compared to knowledge-incompatible completions. Additionally, in LOC, completions that were structure compatible were suppressed more than those incompatible with structure. n.s. = not significant. *p < 0.05. **p < 0.01.

Figure 7.

Whole-brain analysis. The whole-brain exploratory cluster analysis for (A) the repetition suppression effect where the contrast is defined as alternation trials minus repetition trials (red), (B) the structure effect where the contrast is defined as occlusion trials incompatible with structure (S−) minus structure compatible (S+) (green), and (C) occlusion trials incompatible with knowledge (K−) minus knowledge compatible (K+) (blue). Note that for the reversed effects (repetition minus alternation, structure compatible minus incompatible, and knowledge compatible minus incompatible), no significant clusters were found. Overall, these results show a clear repetition suppression effect and modulation of structure and knowledge in the higher-order visual ventral stream.

Figure 8.

Task performance. Depicted are the relationships between the task performance of individual participants and their three main contrasts (alternation minus repetition, structure incompatible minus compatible, knowledge incompatible minus compatible; see Figure 7), for which the average percent of signal change (PSC) of the neural response for each of the regions of interest (ROIs) is shown: early visual cortex (EVC) and lateral occipital complex (LOC). The orange lines show an ordinary least squares regression. According to the Pearson’s correlation coefficient, none of these linear relationships were significant (p > 0.05), meaning that none of the neural contrasts could be explained by participants’ task performance.