Modeling Simultanagnosia

Simultanagnosia is a visual cognitive disorder following a bilateral lesion in parieto-occipital brain areas. It affects the patients integrative perception so that scenes or hierarchically organized objects cannot be perceived as a whole but just in a piecemeal fashion. Qualitative explanations consider impairments of attentional selection mechanisms, feature binding, or global shape processing. Until now, however, no computational model has been used to quantitatively reproduce the performance patients suffering from simultanagnosia. We focus on modeling the impairment of recognizing hierarchical stimuli (Navon letters), which have been used in several studies with patients and healthy subjects. In particular, we apply the established HMAX model of object recognition, specifically trained on letter recognition, to investigate the role of low-level, bottom-up processing, salience, and the selection of a window of attention. We also assess to which extent a top-down modulatory mechanism is necessary to quantitatively reproduce the global letter recognition performance in patients. Our results indicate that as long as a bottom-up segmentation of the global shape from local elements is possible, global shape recognition succeeds. However, when topdown form completion appears necessary to identify the global shape, the current, bottom-up processing model fails. Moreover, we replicate training effects in the global task, which are comparable with patients performance in similar tasks. The present results suggest several promising future research directions to extend the model for modeling the mechanisms underlying global shape recognition in healthy subjects as well as the impairments in patients suffering from simultanagnosia.