Neuron, Vol. 38, 347–358, April 24, 2003, Copyright 2003 by Cell PressCortical Analysis of Visual Contextfrom our findings, that parahippocampal and retrosplen-Moshe Bar* and Elissa AminoffNMR Center at Massachusetts General Hospital ial regions mediate contextual associations, provides aframework that bridges those seemingly unrelated inter-Harvard Medical SchoolCharlestown, Massachusetts 02129 pretations.Context-Specific Cortical ProcessesSummaryThe goal of this first experiment was to compare corticalprocessing of highly contextual objects with the corticalObjects in our environment tend to be grouped in typi-processing of objects that have only a weak contextualcal contexts. How does the human brain analyze suchassociation with other objects. A preliminary survey ofassociations between visual objects and their specific35 subjects was conducted to create a list of objectscontext? We addressed this question in four functionalcomprised of strong contextual association (Strong CAneuroimaging experiments and revealed the corticalobjects). Each of the Strong CA objects was rated tomechanisms that are uniquely activated when peoplebe the most typical object of a specific context (e.g., arecognize highly contextual objects (e.g., a trafficsupermarket cart for “supermarket,” a microscope forlight). Our findings indicate that a region in the parahip-“lab”). A second list contained weak contextual associa-pocampal cortex and a region in the retrosplenial cor-tion (Weak CA) objects, which were defined in a separatetex together comprise a system that mediates bothsurvey with another group of 18 subjects as not beingspatial and nonspatial contextual processing. Interest-associated with any unique context in particular (e.g.,ingly, each of these regions has been identified in thea rope, a camera, a basket). Ideally, objects in this con-past with two functions: the processing of spatial infor-trol list would not be associated with any context at all.mation and episodic memory. Attributing contextualHowever, objects in our environment do not appear inanalysis to these two areas, instead, provides a frame-isolation but rather in multiobject settings, and it there-work for bridging between previous reports.fore seems impossible to generate a list of objects thatcannot be associated with any context. Consequently,Introductionwe selected objects that are very weakly associatedwith many possible contexts (e.g., a person, a genericVisual objects in our environment tend to appear in spe-container). The working assumption was that the recog-cific and typical contexts. For example, a blender isnition of a Strong CA object would immediately activateexpected to be found in a kitchen or on a shelf in anthe information associated with its corresponding con-appliance store. Seeing a blender anywhere else will betext, whereas Weak CA objects would not automaticallysurprising (Biederman et al., 1982). Such clustering ofelicit such contextual activation. This assumption is sup-objects into groups that tend to appear together mayported by established results of contextual primingexplain why recognition of an object that is highly asso-(Biederman, 1981; Palmer, 1975) as well as by some ofciated with a certain context facilitates the recognitionthe results we report here. The stimuli in the subsequentof other objects that share the same context (Bar andfMRI experiment were photographs of objects fromUllman, 1996; Biederman, 1981; Palmer, 1975). Is thisthose Strong CA and Weak CA lists (Figure 1). Weclustering reflected in the cortical processing of contex-scanned six subjects in this first experiment. Their tasktual associations?was to press a response key as soon as they recognizedTo address this question, we used functional mag-what each object was, without having to name it.netic resonance imaging (fMRI) and examined the corti-cal events taking place during the analysis of visualResults and Discussioncontext. Specifically, we compared brain activity, as re-flected by the fMRI signal, elicited during the perceptionMean reaction time for recognizing the objects in eachof visual objects that are highly associated with a certainof the three conditions was statistically comparablecontext (e.g., a hardhat) with the activity elicited by ob-[F(2,105) ⫽ 0.272; p ⫽ 0.76] (average reaction time forjects that are not associated with any unique context ineach condition: Strong CAB⫽ 674 ms; Strong CAI⫽ 702particular (e.g., a fly).ms; Weak CA ⫽ 684 ms). Comparing the fMRI signalThis set of experiments further allowed us to answerelicited by the recognition of highly contextual objectsa more general question about the role and processing(Strong CA) with the signal elicited by the recognitionof associations. Several regions in the brain, most oftenof Weak CA objects resulted in a map of bilateral corticalthe hippocampus, the parahippocampal cortex, and theactivation that concentrated in two main sites (Figureretrosplenial cortex, have been implicated as mediating2). The first and largest focus was in the posterior parttwo different functions: episodic memory (e.g., Ranga-of the parahippocampal cortex (PHC), straddling thenath and D’Esposito, 2001; Valenstein et al., 1987) andcollateral sulcus and the parahippocampal gyrus. Theprocessing of place-related information (e.g., Aguirre etaverage Talairach coordinates of this PHC focus wereal., 1996; Maguire, 2001). The explanation that emerges(⫺24, ⫺41, ⫺4) in the left hemisphere, and it occupied685 mm2of cortical surface where all voxels were differ-entially active at p ⬍ 10⫺8. (For the sake of simplicity*Correspondence: [email protected] of these foci were bilateral and consistent acrossall subjects (Figure 3). An additional site of significantdifferential activation in the comparison between Strongand Weak CA objects was found in the lateral occipitalcortex (LO: ⫺49, ⫺72, 13). This focus, however, waspronounced in the condition where the contextual ob-jects appeared with background (Strong CAB) and maytherefore reflect the differences in physical appearanceand amount of visual information between Strong CABFigure 1. Examples of Stimuli in the Different Conditions (Experi-ment 1)and Weak CA, rather than a context-related differenceObjects in the Weak CA condition were not associated stronglyin processing.with any specific context. We used Strong CA objects “floating” inFigure 4 illustrates the corresponding change in