Inconsistencies between human and macaque lesion data can be resolved with a stimulus-computable model of the ventral visual stream

Abstract

Decades of neuroscientific research has sought to understand medial temporal lobe (MTL) involvement in perception. The field has historically relied on qualitative accounts of perceptual processing (e.g. descriptions of stimuli), in order to interpret evidence across subjects, experiments, and species. Here we use stimulus computable methods to formalize MTL-dependent visual behaviors. We draw from a series of experiments (Eldridge et al., 2018) administered to monkeys with bilateral lesions that include perirhinal cortex (PRC), an MTL structure implicated in visual object perception. These stimuli were designed to maximize a qualitative perceptual property ( feature ambiguity) considered relevant to PRC function. We formalize perceptual demands imposed by these stimuli using a computational proxy for the primate ventral visual stream (VVS). When presented with the same images administered to experimental subjects, this VVS model predicts both PRC-intact and -lesioned choice behaviors; a linear readout of the VVS should be sufficient for performance on these tasks. Given the absence of PRC-related deficits on these ambiguous stimuli, we (Eldridge et al., 2018) originally concluded that PRC is not involved in perception. Here we (Bonnen & Eldridge) reevaluate this claim. By situating these data alongside computational results from multiple studies administered to humans with naturally occurring PRC lesions, this work offers the first formal, cross-species evaluation of MTL involvement in perception. In doing so, we contribute to a growing understanding of visual processing that depends on--and is independent of--the MTL.