Reconstructing Perceived and Retrieved Faces from Activity Patterns in Lateral Parietal Cortex.

Area of Science:

• Neuroscience
• Cognitive Science
• Computer Vision

Background:

• The angular gyrus (ANG), a posterior parietal cortex region, is implicated in memory.
• Previous studies offer limited insight into the nature of ANG content representations.

Purpose of the Study:

• To investigate if the angular gyrus (ANG) actively represents complex, multidimensional stimuli like faces.
• To reconstruct faces from fMRI activity patterns in the ANG.

Main Methods:

• Applied principal component analysis to generate "eigenfaces" from a large face image set.
• Modeled relationships between eigenface values and fMRI activity patterns.
• Used machine learning to predict face components and reconstruct faces from brain activity.

Main Results:

• Successfully reconstructed perceived faces from fMRI activity in the occipitotemporal cortex and lateral parietal regions, including the ANG.
• Reconstructed specific facial information like affect and skin color from ANG activity.
• Demonstrated that a model trained on ANG activity during face perception could reconstruct independently held faces from memory.

Conclusions:

• The angular gyrus (ANG) forms complex, stimulus-specific representations.
• ANG activity patterns reflect representations during both perception and remembering.
• Provides compelling evidence for the ANG's role in actively representing remembered content.