Everyday choices necessitate reclassifying items according to goals, such as grouping carrots with lettuce for meals or with tangerines for artistic arrangements. Dynamic categorization requires the brain to interpret sensory input and integrate current goals. Early sensory brain areas were once viewed as passive processors of visual data for interpretation.
This study evaluates visual cortex adaptation representations for dynamic behavioural support. Shape representations should be more distinguishable across decision boundaries relevant to the task. This was demonstrated in a recent study published in the Nature Communications journal.
Animal studies suggest that lower-level brain areas are more adaptable and actively involved in decision making. Margaret Henderson and colleagues used functional magnetic resonance imaging (fMRI) to investigate brain activity during a shape categorization task in humans.
Ten participants were trained to categorize abstract shapes using various rules. The experiment involved a two-dimensional shape space with stimuli. Participants applied linear decision boundaries or a complex nonlinear rule in different runs. Three task conditions corresponded to different binary categorization rules. Participants constantly switched categorization strategies for the same visual inputs.
Each task run comprised 48 trials over 261 seconds [327 Repetition Times (TRs)], with 32 trials using shapes sampled from a 16-point main grid; each was repeated twice. Participants completed 12 runs per scanning session, totaling 36 runs across three sessions, except for one participant (S06) missing 3 runs due to technical issues. Researchers used the repeated-measure Analysis of Variance (ANOVA) test in Python 3.6 to compare classifier confidence and template correlation coefficients across Regions of Interest (ROIs).
The two-tailed p-value was calculated by comparing iterations where the shuffled t-statistic exceeded or equalled the real t-statistic and vice versa. Multivariate pattern analysis of fMRI data revealed a dynamic visual representation of shape in specific retinotopic brain regions. The effect was pronounced near category boundaries where discrimination was cognitively demanding.
This modulation was strongest in the early visual areas (V1–V3), which are traditionally thought to perform simple feature extraction. Researchers trained a 16-way classifier to identify specific shapes for finer representation explorations.
The LO1 region of the visual cortex aligns more with decision boundaries, indicating that higher visual areas may categorize shape representation differently than early areas, with subtle tuning. The team analyzed classifier confidence between correct and incorrect trials to assess behavioural consequences of neural changes.
In easy trials, a standard set of 16 shapes in a 4×4 grid was used, while hard trials featured shapes sampled from challenging areas near the active boundary of the shape space. This study challenges the traditional view of the visual cortex, revealing its active role in decision-making and real-time adaptation of sensory representations during tasks. Early visual areas V1 and V2 exhibited task-related changes while LO1 adapted by re-coding stimuli into task-aligned categories.
The study increases our understanding of sensory processing by showing how decisions dynamically modulate the visual cortex. In conclusion, these findings support future research on attention-leading and adaptive systems based on humans.
Reference: Henderson MM, Serences JT, Rungratsameetaweemana N. Dynamic categorization rules alter representations in human visual cortex. Nat Commun. 2025;16:3459. doi:10.1038/s41467-025-58707-4
