Neural mechanisms underlying the hierarchical construction of perceived aesthetic value

Kiyohito Iigaya, Sanghyun Yi, Iman A. Wahle, Sandy Tanwisuth, Logan Cross, John P. O’DohertyNature Communications 2023 · DOI

Abstract

Little is known about how the brain computes the perceived aesthetic value of complex stimuli such as visual art. Here, we used computational methods in combination with functional neuroimaging to provide evidence that the aesthetic value of a visual stimulus is computed in a hierarchical manner via a weighted integration over both low and high level stimulus features contained in early and late visual cortex, extending into parietal and lateral prefrontal cortices. Feature representations in parietal and lateral prefrontal cortex may in turn be utilized to produce an overall aesthetic value in the medial prefrontal cortex. Such brain-wide computations are not only consistent with a feature-based mechanism for value construction, but also resemble computations performed by a deep convolutional neural network. Our findings thus shed light on the existence of a general neurocomputational mechanism for rapidly and flexibly producing value judgements across an array of complex novel stimuli and situations.

Introduction

While neuroscience has identified brain regions like the medial prefrontal cortex (mPFC) as key players in representing the "value" of a stimulus, the computational process of transforming high-dimensional visual inputs—like a complex painting—into a single subjective judgment remains poorly understood. Traditional neuroaesthetics has often focused on identifying active brain regions rather than the underlying algorithms. This study investigates whether the brain constructs aesthetic value through a hierarchical integration of features, a process that mirrors the layer-by-layer computations of Deep Convolutional Neural Networks (DCNNs).

Methods

To achieve high statistical reliability, the researchers employed a "deep scanning" fMRI approach. Six participants were scanned extensively over four days, completing 1,000 unique trials each. In each trial, participants viewed a unique painting from Wikiart.org (spanning genres like Cubism and Impressionism) and provided a subjective liking rating.

Experimental paradigm: Participants viewed unique paintings and provided aesthetic ratings while undergoing fMRI.

To model the brain's computations, two complementary approaches were used:

  1. Linear Feature Summation (LFS) Model: This model decomposes images into low-level features (hue, saturation, contrast) and high-level features (concreteness, dynamics), predicting value through a weighted linear combination.
  2. Deep Convolutional Neural Network (DCNN): A VGG-16 network, pre-trained on object recognition, was adapted to output aesthetic ratings based on internal layer activations.

Hierarchical Construction of Value

The study found that the brain processes art in a strictly hierarchical fashion. Early visual cortex (V1-V3) activity correlated strongly with low-level features and early DCNN layers. In contrast, higher-order visual areas aligned with abstract, high-level features and deeper DCNN layers.

Hierarchical model of aesthetic value construction, integrating low-level features (e.g., hue, contrast) and high-level features (e.g., concreteness) to predict subjective liking.

Unlike the visual cortex, the posterior parietal cortex (PPC) and lateral prefrontal cortex (lPFC) acted as "mixing bowls," representing a combination of both low- and high-level features. These regions appear to serve as integration hubs, feeding information into the mPFC, which consistently represented the final subjective value across all participants. Functional coupling between these integration hubs and the mPFC was specifically enhanced during the decision-making phase.

Conclusion

These results suggest that aesthetic appreciation is not merely a product of simple associative memory but an active, on-the-fly computation. By extracting and weighting features within a shared "feature space," the brain can rapidly evaluate novel and complex stimuli. While the small sample size limits generalizations about individual differences, the striking alignment between neural gradients and DCNN architectures provides a powerful framework for understanding how the human brain transforms pixels into beauty. Future research using recurrent neural networks may further clarify the temporal dynamics of this value construction process.

Summary updated

Abstract

Little is known about how the brain computes the perceived aesthetic value of complex stimuli such as visual art. Here, we used computational methods in combination with functional neuroimaging to provide evidence that the aesthetic value of a visual stimulus is computed in a hierarchical manner via a weighted integration over both low and high level stimulus features contained in early and late visual cortex, extending into parietal and lateral prefrontal cortices. Feature representations in parietal and lateral prefrontal cortex may in turn be utilized to produce an overall aesthetic value in the medial prefrontal cortex. Such brain-wide computations are not only consistent with a feature-based mechanism for value construction, but also resemble computations performed by a deep convolutional neural network. Our findings thus shed light on the existence of a general neurocomputational mechanism for rapidly and flexibly producing value judgements across an array of complex novel stimuli and situations.
BibTeX
@inproceedings{iigaya2023neural,
  author = {Kiyohito Iigaya and Sanghyun Yi and Iman A. Wahle and Sandy Tanwisuth and Logan Cross and John P. O’Doherty},
  title = {{Neural mechanisms underlying the hierarchical construction of perceived aesthetic value}},
  booktitle = {Nature Communications},
  year = {2023},
  doi = {10.1038/s41467-022-35654-y},
}
Figure