Computational single-neuron mechanisms of visual object coding in the human temporal lobe

Abstract

Abstract Understanding how the human brain encodes visual objects involves deciphering the neural computations and circuits in the temporal lobe. Here, we recorded intracranial EEG from the human ventral temporal cortex (VTC) and medial temporal lobe (MTL), as well as single-neuron activity in the MTL, to investigate the computational mechanisms of neural object coding. The VTC exhibited axis-based feature coding, and a neural feature space could be constructed using VTC neural axes, within which visual objects clustered according to high-level categorical relationships. Importantly, MTL neurons encoded receptive fields within this VTC neural feature space, exhibiting selective responses to objects that shared perceptual and conceptual similarities. This computational framework, therefore, explains how dense, feature-based representations in the VTC are transformed into sparse, high-level representations in the MTL. We further validated our findings using an additional dataset with different stimuli. Notably, we uncovered the physiological basis of this computational framework by demonstrating VTC-MTL interactions at multiple levels. Together, our neural computational framework provides a mechanistic understanding of the neural processes underlying object recognition.