Attention modulates neural representation to render reconstructions according to subjective appearance

Area of Science:

• Neuroscience
• Cognitive Science
• Computer Vision

Background:

• Visual perception is influenced by both external stimuli and internal cognitive processes, such as attention.
• Reconstructing images from brain activity has been achieved, but it remains unclear how top-down attention affects these reconstructions.

Purpose of the Study:

• To investigate the impact of attention on image reconstruction from visual cortical activity.
• To determine if and how reconstructions reflect top-down aspects of perception.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to measure brain activity while participants attended to one of two superimposed images.
• A deep neural network (DNN) based method was employed to decode fMRI activity into image features and reconstruct images.

Main Results:

• Reconstructed images predominantly resembled the attended image, not the unattended one.
• The reconstructions could be modeled by adjusting the contrast of superimposed images, mimicking the subjective experience during attention.
• Attentional modulations were observed across various hierarchical visual representations, aligning with brain-DNN correspondences.

Conclusions:

• Top-down attention actively modulates neural representations in the visual cortex.
• Attention counteracts stimulus-driven responses to align neural activity with subjective perception.
• Image reconstruction methods can capture the influence of attention on visual perception.