Dissociable brain regions process object meaning and object structure during picture naming

Abstract

Although imaging studies have indicated that the fusiform gyrus is important in tasks of picture naming, whether this region encodes an object's structure or its meaning is not clear. We used positron emission tomography to examine cerebral blood flow (CBF) changes in response to a picture naming task that varied on two dimensions: familiarity (or difficulty: hard vs easy) and category (tools vs animals). Results show that although familiarity effects are present in the frontal and left lateral posterior temporal cortex, they are absent from the fusiform gyrus. This provides strong evidence that the processing carried out in the fusiform gyrus relates to an object's structure, not to its meaning, and that the left posterior middle temporal gyrus instantiates in part the semantic network that represents the object's meaning.

Introduction

The ability to identify and name visually presented objects involves multiple levels of progressively complex processing. Studies of individuals with focal brain lesions suggest that (in right-handers), focal left frontal lesions produce an impairment in object and picture naming via disruption in the ability to access the sound form of the words [7]. A similar naming deficit has also been induced by electrical stimulation of the basal temporal area in epileptic patients [6], [30], [31]. Brain-damaged individuals with lesions of the left occipito-temporal or posterior left temporal regions, however, develop a naming deficit for quite different reasons; they have either deficits in deriving the complete visual perceptual representation of the object's structure, or are impaired in their ability to access the stored semantic representation of the object's meaning [7], [14], [26].

Brain imaging studies have helped to localise in some measure the brain structures that are most involved in these identification processes. The posterior brain structure most consistently found to be activated in picture naming tasks is the fusiform gyrus of both hemispheres [5], [25], [27], [28], [35], [41], [49], located on the ventral surface of the occipital and temporal lobes. There is conflicting data, however, regarding the level of processing instantiated in this brain region. Since the fusiform is activated during both silent naming of real objects [35] as well as passive viewing of nonsense objects, it has been assigned a role in the perception of object form (similar in function to the occipital cortex) or as the storage site of perceptual information concerning known forms. Kanwisher, Halgren, Chao and others [8], [19], [24], however, have found that the specific areas of the ventral surface of the brain can be selectively activated by different categories of objects. For example, faces have been found to selectively activate the lateral posterior fusiform, whereas artifacts produce greater activation in the medial fusiform. That distinctions based on the semantic category of the stimuli are reflected in this region seems to suggest that these regions store aspects of the object's meaning. Indeed, other studies have produced results suggesting that semantic information as opposed to structural information is stored in the fusiform area. For example, when subjects are asked to read or make semantic (i.e. category membership) decisions on words referring to objects, there is also activation of the fusiform region [33], [37], [41], [45]. Furthermore this area has been found to be activated during tasks demanding conceptual [45], [52] object colour [33] and real world location knowledge [11]. These results could also be taken to indicate that semantic features are stored in the fusiform region bilaterally. We would argue that these conflicting results have not resolved the localisation question. An alternative interpretation would implicate a top-down activation of regions involved in perceptual and structural processing by higher order ‘semantic’ areas, an explanation invoked to explain category-specific activation of occipital cortex by animal pictures [35] or by visual imagery [29].

How might the neural basis of these different levels of representation be disentangled? In positron emission tomography (PET) studies of other cognitive domains, processing of more difficult or less familiar stimuli resulted in much greater focal changes in cerebral blood flow, reflecting the necessity of recruiting larger regions of relevant cortex for the task [23], [47], [55]. We hypothesised that the naming of visually presented items with weaker semantic representations (i.e. less familiar items) would impose a greater processing load on exactly those regions of cortex instantiating a semantic rather than a purely visual perceptual or structural level of representation. We therefore have used positron emission tomography to determine which areas of the brain were activated in picture naming while orthogonally controlling two factors: familiarity and category. The two categories contrasted were animals (which included birds and marine life) and manipulable tools. For both of these semantic categories the subjects were presented with pictures of highly familiar items (easy blocks) and of known but less familiar items (hard blocks). Visual complexity was similar for all four blocks. Subjects were asked to name the picture aloud, while undergoing PET scanning with the bolus H₂¹⁵O methodology. Standard subtraction images were subsequently created by comparing CBF during picture naming with two baseline conditions (presentation of plus signs and presentation of abstract designs). Subtraction images were also created by comparing the different picture naming scans with each other. In addition, regression maps were created in which each individual's behavioural performance (accuracy of picture naming) was regressed against cerebral blood flow for that particular block of pictures. This novel approach, pioneered by Paus et al. for sensory stimuli [44] supplied a powerful corollary approach to data analysis, since the regression map made no assumptions regarding the validity of any subtraction condition, and strictly evaluated which areas were more active during more effortful naming.