Abstract
The repertoire of human visual recognition skills is amazingly broad and ranges from rapid “gist”-based scene categorization to the fine scrutiny of minute object details. While ventral occipital cortex is implied in all these abilities, the computational organisation that enables this is still poorly understood. Recent eye-tracking studies have shown that eye-movement characteristics can be used to distinguish between two different modes of perception, one associated more with global processing, and the other with more detailed visual analysis. We reasoned that if these perceptual modes reflect genuinely different cortical processing, we should be able to use eye-movements to tease apart the underlying neural correlate. In our functional MRI experiment, participants freely viewed images of visual indoor scenes while their brains were scanned and their eye-movements tracked. We define two classes of eye-movement events to approximate the different viewing modes. Brief fixations followed by large saccades were defined as “scanning” events, whereas long fixations followed by short saccades represent “inspection” events. These events were subsequently used in the analysis of the fMRI data. Independent component analysis indicated the existence of two clusters in ventral occipital cortex. The cluster of activity in ventro-medial occipital cortex was preferentially associated with scanning events while inspection events were preferentially associated with activity in the ventrolateral cluster. Hence, this shows that fMRI signals recorded from ventral cortex can be segregated based on eye-movements. Information processing during scanning events is suggested to be of statistical nature, given their brevity and the peripheral location of the saccade target. The longer inspection events presumably enabled additional scrutiny of features as well as computation of spatial relationships. Hence, our work suggests that the human ventral stream subdivides into two vision systems that enable perception based on distinct visual information and neural computations.
This study was supported by European Commission grant 043261 (Percept) to Frans W. Cornelissen.