Functional magnetic resonance imaging - MRC …

1 Functional magnetic resonance imagingof the human object-vision system methodological and empirical contributionsProefschriftter verkrijging van de graad van doctor aan de Universiteit Maastricht,op gezag van de Rector Magnificus, Mols,volgens het besluit van het College van Decanen, in het openbaar te verdedigenop donderdag 14 oktober 2004 om uurdoorNikolaus R. GoebelCopromotorDr. E. G. Kok, voorzitterDr. L. BlomertDr. B. JansmaProf. Dr. W. Hussy,Universit at zu K oln(K oln, Duitsland)Prof. Dr. W. Singer,Max-Planck-Institut f ur Hirnforschung(Frankfurt, Duitsland)ISBN 90-9018729-4 ContentsIntroductionvChapter 1 Topologically correct cortex segmentation1 Chapter 2 Cortical object recognition from a visual motion flowfield29 Chapter 3 Information-based Functional brain mapping57 Chapter 4 Single-object-image response patterns in inferotemporal cortex89 References107 AppendixSummary117 Samenvatting123 Acknowledgments129 Curriculum vitae130 Contact information130iiiIntroductionFunctional magnetic resonance imaging (fMRI), the empirical technique at thecenter of this thesis, is a noninvasive method of measuring brain activity inhumans and animals.

2 The whole brain or large portions of it can be measuredsimultaneously with tens of thousands of measurement channels arranged in aregular 3D grid. Each channel represents a cuboid volume element (voxel) of awidth of 1-5 high spatial resolution of fMRI represents an advantage over the other brainimaging techniques available for human studies. Alternative techniques includepositron emission tomography (PET) and electro- and magnetoencephalogra-phy (EEG, MEG). In addition to its high spatial resolution, fMRI combinesthe advantages of depth-pervasive 3D measurement (which it shares with PET)and noninvasiveness (which it shares with EEG and MEG). In contrast to PET,which involves injection of radioactive isotopes, fMRI is not known to be asso-ciated with any health risk to the activity is measured indirectly in fMRI via the vascular localized increase of neuronal activity entails a localized vascular response,which channels more oxygenated blood into the activated region.

3 The fMRIsignal reflects this increase in blood-oxygen level and is therefore temporallydelayed (peaking after about 6 s) and blurred with respect to the underlyingneuronal activity. This limits the spatial and temporal detail, at which brainactivity can be of its combination of depth-pervasive whole-brain coverage, goodspatial resolution, and noninvasiveness, fMRI has revolutionized the study ofhuman brain function. Compared to the classical technique of invasive elec-trophysiology, which is mainly used in animal research, fMRI has an obviousdisadvantage: its much lower resolution in both space and time. However, itsadvantage consists in the spatial coverage (up to whole-brain coverage) and thetremendous number of channels that can be recorded simultaneously. FMRIis therefore becoming increasingly important also in studying the monkey the black boxThe contrast mechanism of blood-oxygen-level-dependent fMRI (the dominanttechnique, which is also used in these studies) has only been discovered in the short period of its existence, fMRI has played a central role in the forma-tion of the transdisciplinary field of cognitive neuroscience.

4 Because it providesa means to study human brain activity at the systems level without any knownvviINTRODUCTION health risks, it is naturally suited to illuminate the black box, whose contentshave been described in the abstract through decades of research in cognitivepsychology. An obvious application, then, is the localization of brain structuressubserving the cognitive component functions postulated by cognitive psychol-ogy on the basis of purely behavioral field of human brain mapping attempts to characterize the global ar-chitecture of the human brain by assigning Functional labels to macroscopicbrain regions. This research paradigm goes back to phrenology, but it hasempirically proven itself. Brain imaging has demonstrated that the globalpatterns of activity associated with a wide range of cognitive processes areconsistent across individual brains.

5 Like the organs of the body, different brainregions serve specific functions. The gross Functional parcellation is largelyuncontroversial for perceptual and motor challengesIt has yet to be established empirically to what extent different parts of cortexconform to the attractive model of Functional modules. Is all of cortex likeearly visual cortex subdivided into crisply demarcated macroscopic areas thatexist in every individual brain and serve clearly defined functions that can becharacterized by one-word Functional labels? For frontal-lobe higher cognitivefunctions, for example, the spatial layout has been much less precisely definedthan for the occipital-lobe visual functions. Despite vigorous research, it remainsunclear to what level of spatial precision the Functional layout of the frontal lobesis consistent across brain function will take more than a global Functional fMRI research goes beyond mere Functional labeling of macroscopicbrain regions based on group-averaged data.

6 Functional regions are defined inindividual subjects. Their activity levels during different cognitive processes arecompared. The time course of the activity and Functional interactions betweenregions are investigated. Based on anatomical magnetic - resonance -imagingdata, the folded cortex of each subject is reconstructed as a two-dimensionalmanifold. Such individual cortical-surface representations provide constraintsfor advanced methods of statistical analysis of Functional data and allow visu-alization of activity patterns on flattened cortical maps. These methodologicaldevelopments allow current fMRI research to characterize more precisely thedistributed computational processes occurring in the thesisMotivated by questions about visual object recognition, this thesis combinesmethodological and empirical contributions. It presents two new methods: (1) acortex-segmentation method that takes the known topology of the cortical sheetinto account and (2) a statistical analysis method that utilizes the informationcontained in fine-grained fMRI activity patterns.

7 These methods are applied intwo empirical studies. The first study, taking the global view, addresses the inter-action between the dorsal and the ventral visual stream two subsystems oftenconceptualized as largely independent. The second study, zooming in on fine-grained activity patterns, investigates the distributed representation of objectimages in ventral temporal cortex. We apply method (2) to study how partic-viiular exemplars of an object category ( particular faces) are represented infine-grained response patterns that may be unique to each individual a whole, the thesis demonstrates the advantages of developing theory, ex-periment, and statistical analysis methods in close interaction. This interactionis in evidence in the first half (Chapters 1 and 2), but even closer in the second(Chapters 3 and 4). In the second half, a theoretical prediction (the existenceof macropopulation codes for object exemplars) motivates the developmentof a new analysis technique (information-based Functional brain mapping).

8 In conjunction with a novel experimental approach (the single-object-imageapproach), the analysis technique leads to the challenge of a widespreadassumption (that the fusiform face region represents faces at the individuallevel) and suggests the existence of a Functional region that has not previouslybeen described (the anterior inferotemporal face-exemplar region).OverviewThis thesis combines methodological developments (Chapters 1 and 3) and em-pirical investigations of human object vision (Chapters 2 and 4). The methodsdeveloped concern the analysis of anatomical (Chapter 1) and Functional (Chap-ter 3) magnetic - resonance - imaging data and are employed in the empirical in-vestigations. The chapters can be read in isolation as each is self-contained andcorresponds to a paper either published (Chapters 1 and 2), submitted (Chapter4), or in preparation (Chapter 3).

9 Chapter 1 presents a method revealing the spatial structure of the cortex inanatomical magnetic resonance volumes. An explicit representation of the corti-cal sheet of each subject allows brain activity to be visualized on a cortical mapand is prerequisite to a number of advanced methods of statistical analysis offMRI data. Our approach avoids the topological errors produced by 2 presents an investigation of object recognition based on perceptionof a visual motion flowfield. Object recognition is known to involve the ventralstream of visual cortex. The dorsal stream, in contrast, processes informationabout object motion and location. In the natural world, objects are often inmotion relative to the observer, so both systems are usually coactivated. In thisstudy, we use structure-from-motion stimuli to address how the dorsal and theventral visual subsystem interact in object 3 presents information-based Functional brain mapping, a novel statis-tical method, which finds brain regions whose fine-grained spatial response pat-tern contains information about the experimental condition.

10 The information-based approach is in contrast to conventional activation-based brain mapping,which is best suited for finding regions whose spatially averaged activity isgreater in one condition as compared to a control 4 explores the representation of visual objects in ventral temporal cor-tex. In contrast to previous studies, we do not average responses to differentimages of the same object category, but estimate the spatial response patternselicited by four particular images (two faces and two houses). This allows us toaddress how exemplars within the same category are distinguished. The studyis crucially dependent on the method of information-based brain mapping in-troduced in the previous Summary (in English and Dutch in the Appendix) provides an accessibleguided tour of the thesis. It introduces the minimal background required tomotivate each project and explains the relationships between the projects at aconceptual level.