sign in sign up
<<
Home , Hippocampus, Hippocampus proper, Parahippocampal gyrus, Spatial view cells

THE ANATOMICAL RECORD (NEW ANAT.) 265:111–120, 2001

TUTORIAL

Unfolding the With High Resolution Structural and Functional MRI

MICHAEL M. ZEINEH, STEPHEN A. ENGEL, PAUL M. THOMPSON, AND SUSAN Y. BOOKHEIMER*

The hippocampus is a region of the that is crucial to function. Functional allows for the noninvasive investigation of the of human memory by observing changes in blood flow in the brain. We have developed a technique that employs high-resolution functional magnetic resonance imaging (fMRI) in combination with cortical unfolding to provide activation maps of the hippocampal region that surpass in anatomic and functional detail other methods of in vivo mapping of the medial . We explain the principles behind this method and illustrate its application to a novelty- paradigm. Anat Rec (New Anat) 265: 111–120, 2001. © 2001 Wiley-Liss, Inc.

KEY WORDS: biomedical imaging; brain imaging; imaging; magnetic resonance imaging; MRI; functional MRI; fMRI; medial temporal lobe; hippocampus; memory; cortical unfolding; flat maps

One of the most influential findings in temporal lobe is the hippocampus, a Buzsa´ki, 1996; Rolls, 1999). Finally, at the history of involves a site of neuroanatomic convergence in the level of cellular populations, neu- patient known as HM, who suffered the brain. At the time, little was roimaging researchers measure the from intractable . To treat his known about its importance, and the metabolic consequences of changes in epilepsy, the neurosurgeon Scoville procedure reduced the severity of neuronal firing in the awake human performed a bilateral resection of HM’s . However, HM was left during memory tasks. The question is HM’s medial temporal lobes (Scoville with significantly impaired memory the same at all levels: how does the et al., 1957). At the heart of the medial function. Specifically, he had severe hippocampus engage in the anterograde (an inability to and retrieval of information? learn new information) and mild ret- Mr. Zeineh is in the MD-PhD training rograde amnesia (an inability to program at UCLA School of Medicine. FUNCTIONAL BRAIN IMAGING: past events). From the data of HM and He is currently completing his disserta- HOW IT WORKS tion work on functional imaging of the patients like him, it has become in- human hippocampus at the Ahmanson- Lovelace Brain Mapping Center. Dr. creasingly clear that the medial tem- Functional magnetic resonance imag- Bookheimer is an Associate Professor poral lobe is a structure vital for mem- ing (fMRI) and positron emission to- at the Brain Mapping Center, UCLA ory formation and retrieval (Squire, mography (PET) both rely on an epi- School of Medicine. Her research pro- gram focuses on functional magnetic 1992). phenomenon of neural activity: when resonance imaging of language, mem- This finding has sparked an entire many in a region of the brain ory, and their disorders. Dr. Thompson field of research, spanning the gamut increase their firing rate, local blood is an Assistant Professor in the Depart- ment of at the UCLA School from molecular to systems neuro- flow increases to that region of the of Medicine. His research team devel- science, that explores the role the hip- brain (Roland, 1993). Both PET and ops new techniques for brain image analysis, and applies them to study pa- pocampus plays in memory. At the fMRI attempt to measure this change tient populations with Alzheimer dis- molecular level, scientists are study- in blood flow. With PET, radioactive ease, , brain tumors, and ing the behavior of ion-channels that H O15 is injected intravenously in pediatric brain disorders. Dr. Engel is an 2 Assistant Professor of at facilitate potentiation of neuronal re- subjects, and by tomographic recon- UCLA. His research focuses on behav- sponses, a phenomenon that is struction of radioactive decay events, ioral and neural measurements of hu- thought to be the molecular signature a map of blood flow in the brain man vision and memory. *Correspondence to: Susan Y. Bookhei- of memory formation (Milner et al., emerges. With arterial blood sam- mer, UCLA Department of Psychiatry 1998; Bennett, 2000). At the level of pling, one may obtain quantitative and Biobehavioral Sciences, Brain Map- ping Division, Ahmanson-Lovelace individual cells, researchers record maps with units of blood flow in mil- Brain Mapping Center, 660 Charles E. from rats and monkeys to identify cel- liliters per minute per gram. Statisti- Young Dr., Rm. 205, Los Angeles, CA lular and network correlates of object cal comparisons between maps of 90095; Fax: 310-794-7406. E-mail: [email protected] and place memory (O’Keefe and blood flow during resting states and Nadel, 1978; Rolls and O’Mara, 1995; maps of blood flow during memory

© 2001 Wiley-Liss, Inc. 112 THE ANATOMICAL RECORD (NEW ANAT.) TUTORIAL

Figure 1. Anatomical diagram, with an inferior view of the whole brain at the left, an inferior view of the temporal lobe at the center, and coronal sections from the two indicated slices at the right. A: Anterior slice through . B: Posterior slice through parahip- pocampal cortex. ERC, entorhinal cortex; PRC, ; PHC, parahippocampal cortex; FG, fusiform ; Sub, ; CA 1, cornu ammonis 1; CA 2,3, CA fields 2 and 3; DG, ; CoS, collateral . tasks can reveal changes in blood flow in neural firing, there is an increase in and Bookheimer, 1994). This type of accompanying processes incoming oxyhemoglobin. In fact, the functional contrast is termed blood (Cherry and Phelps, 1996). increase in oxyhemoglobin delivery is oxygenation-level dependent (BOLD) The power of fMRI is that it accom- greater than the increase in oxygen MRI. Hence, images are taken during plishes similar goals noninvasively. In consumption (Fox and Raichle, 1986). a resting condition and during a mem- fMRI, continuous MRI scanning oc- The net effect is an increase in venous ory condition, and statistics are per- curs while a subject is engaged in a oxyhemoglobin compared with de- formed to identify which pixels have task, with the acquisition of a com- oxyhemoglobin several seconds after an increase in intensity during the plete volume of the brain every few the neural event. Deoxyhemoglobin memory condition, that is, an in- seconds over a period of minutes (this causes more perturbation, that is, crease in BOLD contrast, resulting stands in contrast to structural MRI, from a change in blood flow. in which only one volume is ac- quired). As reviewed by Cohen (1996) How does the ANATOMY OF THE and by Mori and Barker (1999), MR HIPPOCAMPAL REGION images are produced by placing a sub- hippocampus engage in ject in a strong, uniform magnetic the storage and retrieval The hippocampal region is the center- field and applying radio frequency piece of the medial temporal lobe, and (RF) pulses of electromagnetic energy of information? it is a compact structure with several to synchronously alter the precession components (Fig. 1). It can be roughly of protons in H2O. After an RF energy divided as having two constituents: pulse, the protons are precessing in nearby cortical areas, and the hip- more deviation from uniformity in the phase, and this precession can be pocampus proper. The nearby cortical magnetic field in comparison to oxy- measured with a receiver coil. The areas project to the hippocampus . Thus, more deoxyhemo- precession dephases with time result- proper, and the globin results in more signal decay ing in signal decay. In areas of the has an internal system of projections brain where the magnetic field is less and lower signal intensity; conversely, that sends output back to the nearby uniform, the dephasing occurs faster; more oxyhemoglobin results in less cortical areas. The entorhinal cortex consequently, signal decay is greater, signal decay and higher signal inten- (ERC), which lies in the anterior por- resulting in darker pixels. Images sity (i.e., brighter pixels). With the tion of the whose contrast is based on measuring greater relative venous oxyhemoglo- and borders the anterior hippocam- such differences in dephasing are bin concentration that accompanies pus, is the main gateway of input to termed T2-weighted images. activation, less signal decay occurs in the hippocampus proper (Fig. 2) (Su- When blood flow increases to a part those activated pixels, and those pix- zuki and Amaral, 1994b). It receives of the brain in response to an increase els are brighter in an image (Cohen most of its projections from two adja- TUTORIAL THE ANATOMICAL RECORD (NEW ANAT.) 113

jacent areas. Thus, correlation be- tween structure and function has largely been only at the gross level of the “hippocampal region,” encom- passing all of the aforementioned structures and their putative diverse functions. Furthermore, this has likely reduced the power of previous neuroimaging studies because signal is averaged from different neural structures that may have different or Figure 2. Simplified circuit diagram of the major connections in the medial temporal lobe. See Figure 1 caption for abbreviations. even opposite responses. To properly address the detailed anatomy of the cent areas: the perirhinal cortex (PRC) gion, selective lesions to the perirhinal region, high-resolution scanning and and the parahippocampal cortex cortex result in the most profound im- visualization are necessary. Recent (PHC) (Amaral and Insausti, 1990). pairment in this task (Aggleton and higher field MR technologies offer in- The perirhinal cortex is situated lat- Brown, 1999; Murray Bussey, 2000). creased signal-to-noise and make eral to the entorhinal cortex, lining In contrast, lesions to the (cut- scanning at high resolution possible. the collateral sulcus anteriorly. The ting off a significant portion of hip- Cortical unfolding techniques offer a parahippocampal cortex lies on the pocampal output) impair spatial dis- superior method for visualizing acti- parahippocampal gyrus posterior to crimination abilities in monkeys and vation along the convoluted cortical both the entorhinal and perirhinal in rats. This has led surface and facilitate the comparison cortices. Both perirhinal and parahip- some researchers to suggest the and averaging of data across subjects pocampal cortices receive widespread perirhinal cortex is important for (Fischl et al., 1999b; Zeineh et al., projections from many cerebral areas recognition, whereas the 2000a). (Suzuki and Amaral, 1994a; Murray hippocampus proper is more impor- and Bussey, 2000). tant in encoding and recalling the syn- The essential circuit diagram of the thesis of stimulus information with Recent higher field MR hippocampus proper is a loop (Fig. 2). context (e.g., spatial context), or epi- technologies offer The main circuit starts with the ento- sodic memory (Aggleton and Brown, increased signal-to- projecting to the dentate 1999). With regard to human lesion gyrus (the , labeled as data, focal parahippocampal lesions noise and make such because fibers perforate the sub- cause difficulty in spatial memory and scanning at high iculum on the way to the dentate gy- navigation (Bohbot et al., 1998). In rus). From there, neurons project suc- the hippocampus proper, vic- resolution possible. cessively to fields CA 3 (Cornu tims with lesions largely limited to cell Ammonis 3), CA 1, the subiculum, and field CA 1 exhibit measurable memory back to the entorhinal cortex, where impairments in recall and recognition CORTICAL UNFOLDING projections return to the parahip- tests (Zola-Morgan et al., 1986), and TECHNIQUES pocampal and perirhinal cortices. A epileptics exhibit a correlation be- detour is also taken from CA 3 via the tween cellular loss in left CA 1 and The human is highly fornix, a tract that at- verbal memory impairments (Rausch folded, presumably to increase the taches to the inferior border of the and Babb, 1993). Thus, human and amount of cortical surface area in the septum pellucidum and projects to monkey data suggests that the hip- limited volume of the human skull. the mammillary nuclei within the hy- pocampal region may consist of het- This gyrification has posed multiple pothalamus as well as the septal nu- erogeneous components that exhibit difficulties for localizing and display- clei. differential susceptibility to memory ing blood flow changes in the brain. Data from monkeys and tasks. First, individuals vary in their pattern suggest that these regions may offer Although recently many imaging of gyri and sulci. Hence, typical tech- different contributions to the medial studies have successfully localized ac- niques of volumetric aligning and pro- temporal lobe memory system. A rec- tivation to the region (Cohen et al., cessing data across subjects often blur ognition test commonly employed in 1999), parsing out the activation together gyri and sulci because they monkeys is the delayed nonmatching- among the anatomical substructures do not take into account individual to-sample task. In this memory para- has proven a formidable challenge. To differences. Second, visualization of digm, the monkey sees a stimulus and improve the signal-to-noise ratio, activation is difficult because signal after a delay sees another stimulus, many prior imaging studies have per- changes may occur either on the gyrus and the monkey responds if the sec- formed imaging with relatively low or inside the sulcus. ond stimulus is not identical to the resolution, averaged across subjects, Cortical unfolding techniques can first (i.e., if the test stimulus does not and extensively smoothed the data; successfully address these two diffi- match the sample stimulus). Of all these procedures do not accurately culties: they account for across-sub- components in the hippocampal re- preserve the boundaries between ad- ject variability by individual segmen- 114 THE ANATOMICAL RECORD (NEW ANAT.) TUTORIAL tation, and they flatten the data to et al., 2000a). This allows us to per- goal in segmenting white matter is to produce simple two-dimensional form powerful group statistics while provide a continuous strip of pixels maps (Van Essen et al., 1998). To per- maintaining the high resolution nec- upon which the gray matter in the form cortical unfolding, high resolu- essary for localizing activation to the hippocampus and adjacent cortex tion structural and functional images different anatomic substructures (Fis- rests. Toward this end, as Figure 3B are taken of individual subjects. chl et al., 1999b; Zeineh et al., 2000a). illustrates, we classify pixels as white Structural images have only one time We describe the unfolding proce- matter in the parahippocampal and point and have high spatial resolution, dure in two parts: structural unfolding fusiform gyri and extend the white while functional images have multiple and functional mapping. To accom- matter all the way around to the top of time points and have high temporal plish the structural unfolding, high the hippocampus. The goal in seg- resolution that facilitates the mea- spatial resolution structural scanning menting CSF is to limit the extent to surement of BOLD contrast. The takes place, followed by segmenta- which gray matter is grown (i.e., so structural images are segmented; that tion, identification of boundaries be- that it does not cross a sulcal bound- is, pixels in the image are labeled as tween the medial temporal subre- ary or extend into the CSF). We clas- gray matter, white matter, and CSF gions, and computational flattening. sified the collateral sulcus as CSF, fa- (Teo et al., 1998; Van Essen et al., For the functional mapping, we ac- cilitating the separation of adjacent 1998). The gray matter surface is then quire coplanar functional scans and cortex on different banks of the sulcus extracted and stretched until it is a perform statistical processing to gen- (inferior no. 5), as well as the hip- two-dimensional flat surface. Statis- erate activation maps. Overlaying the pocampal fissure (superior no. 5), fa- tics are computed on the functional activation maps on the flattened hip- cilitating the separation of fields CA2, images to generate activation maps pocampi from the structural unfold- 3, and the dentate gyrus from field CA (Cohen, 1997), and these activation ing delivers the high-resolution func- 1 and the subiculum. Classifying the maps are then superimposed on the tional activation maps. boundary of the segmentation as CSF flat maps. This technique was first allows for a limited volume of gray used with fMRI in early visual areas to Structural Unfolding of the matter to be grown that encompasses provide maps of retinotopy, the map- only hippocampal and adjacent para- Hippocampus ping of the visual field from the retina hippocampal gray matter. to (Engel et al., 1994; The goal of the structural unfolding At the end of the segmentation, we Sereno et al., 1995). Primary visual procedure is to create a gray matter have a three dimensional topological cortex lies in the along volume (a three-dimensional set of volume of gray matter pixels. By topo- the , a particularly pixels that are classified as gray mat- logical, we mean that we keep track of winding sulcus, and visualizing reti- ter) and stretch and unfold this vol- the spatial relationships between each notopy is difficult in three dimen- ume until it is flat. We acquire high- gray matter pixel and its neighboring sions. Segmentation and unfolding resolution structural images (0.4 ϫ gray matter pixels. The first layer of ϫ converts the complex sulcal pattern to 0.4 3 mm slices) with T2 contrast so gray matter pixels forms a thin mani- a simpler flat representation where ar- we can discern the anatomic detail of fold, and subsequent layers on top of eas V1, V2, and V3 can be easily visu- the region (Fig. 3B). In a T2 weighted this manifold form a volume. The un- alized. image, white matter is dark (low T2 folding procedure works by taking the decay time), gray matter is brighter first layer manifold, and stretching it (medium T ), and CSF is brightest so that it is flat while maintaining to- UNFOLDING THE HIPPOCAMPUS 2 (high T2). As illustrated by the slice pology and minimizing distance er- The hippocampus is a compact struc- prescription in Figure 3A, we acquire rors (Wandell et al., 1996; Engel et al., ture, and the region contains poten- these images perpendicular to the 1997; Zeineh et al., 2000b). Distance tially heterogeneous areas as alluded long axis of the hippocampus to min- errors are the difference in distance to previously. Unfolding techniques imize slice-to-slice variation in anat- between two pixels along the manifold are ideally suited for reducing the an- omy. before unfolding and along the flat atomic complexity of the region to a Using this image set, we proceed to map after unfolding; such errors are more tractable two-dimensional prob- segment and create a gray matter vol- inevitable in the distortion that results lem. By applying unfolding tech- ume, that is, an organized array of from flattening. If you recall flat maps niques to the hippocampal region, we pixels covering all of the medial tem- of the earth, the Northern and South- can localize activity with greater pre- poral structures of interest that we ern poles are heavily distorted to cision than has ever been possible to will subsequently unfold. In our un- maintain continuity. This distortion the different subregions of the medial folding technique, we manually label can be eliminated if cuts are made on temporal lobe, namely, the CA fields, pixels as white matter and CSF. A re- latitude lines. Similarly, when the subiculum, and parahippocampal ar- gion expansion algorithm grows out whole brain is flattened, researchers eas (Zeineh et al., 2000b). Further- layers of gray matter, with the first often make cuts to reduce overall dis- more, we have created a hippocampal layer starting at the white matter sur- tortion (Drury et al., 1996; Van Essen reference template based on the sub- face, the second layer starting on top and Drury, 1997; Fischl et al., 1999a). jects in our studies that facilitates the of the first, and the last layer ending Fortunately, we unfold only the me- registration of subjects in the same adjacent to the CSF boundary (Fig. dial temporal lobe, and minimal dis- space (Thompson et al., 2000; Zeineh 3B, bottom) (Teo et al., 1998). The tortion is incurred in the unfolding TUTORIAL THE ANATOMICAL RECORD (NEW ANAT.) 115

Figure 3. A: Sagittal T2-weighted scout image with the superimposed slice prescription for the 16 coronal high-resolution structural images covering the medial temporal lobe. Red slices indicate the 11 functional images taken, and the blue box indicates the anatomic location of the medial temporal lobe. B: Segmentation of the coronal high-resolution structural image set: beige ϭ gray matter; yellow ϭ CSF; green ϭ gray matter. (1) White matter in the parahippocampal gyrus (2) White matter in the occipitotemporal gyrus (3) CSF in the lateral ventricle (4) The fornix (5) CSF in the hippocampal fissure (top) and collateral sulcus (bottom) (6) The posterior cerebral artery and basal vein (7) Boundaries of the segmentation C: Demarcation of the subregions on the coronal image set. PRC, perirhinal cortex; ERC entorhinal cortex; FG, ; MFV, medial fusiform vertex; PHC, parahippocampal cortex; Sub, subiculum; CA, cornu ammonis 1; CA23DG, CA 2 and 3 and dentate gyrus; CADG, CA 1, 2, 3 and dentate gyrus; CoS, collateral sulcus; AntC, anterior calcarine sulcus. D: Unfolded hippocampi with superimposed demarcations. The units are in millimeters. Dark gray pixels correspond to anterior slices, while bright gray pixels correspond to posterior slices.

process on a relatively small stretch of we have a transform that takes pixels tological and MRI atlases, we demar- cortex, so we do not cut any part of from 3D volumetric space to 2D flat cate the boundaries between the fol- our segmentation (Zeineh et al., space. lowing regions: fusiform gyrus (FG), 2000b). After the first layer is un- Once we have flat structural images, parahippocampal cortex (PHC), peri- folded, subsequent layers are com- we create a map of the different archi- rhinal cortex (PRC), entorhinal cortex pressed onto this layer and the corti- tectonic subregions (Fig. 3C). Using a (ERC), subiculum (Sub), CA 1, and CA cal volume has been flattened. Thus, rule-based system in concert with his- fields 2, 3, and dentate gyrus (CA23DG). 116 THE ANATOMICAL RECORD (NEW ANAT.) TUTORIAL

enables us to acquire volumes of the hippocampus with a resolution of 1.6 ϫ 1.6 ϫ 3 mm every 3.5 sec. The resulting set of images has three spa- tial dimensions and one temporal di- mension, and we term this data set a time series of volumes. The spatial resolution is adequate for registration with the high-resolution structural data, and the temporal resolution en- ables us to observe BOLD hemody- Figure 4. Blood flow response: dark lines are the boxcar waveform, and shaded lines are namic signal changes, which peak the boxcar waveform smoothed and delayed to model typical hemodynamic response about 5–7 sec after a change in neuro- latencies. Reproduced from Zeineh et al., 2000b with permission of the publisher. nal firing (Cohen, 1997). We applied the flat technique to a novelty-encoding paradigm, a mem- The last category combines several Functional Activation Maps in ory activation task that compares the histological fields that even at our Flat Space encoding of novel stimuli the subject high resolution are inseparable. We has never seen before with the same define these boundaries on the struc- For the functional imaging, we take stimulus repeated several times. Be- tural 3-D images, and using the trans- echo-planar MR images while the sub- cause an intact medial temporal lobe form from 3-D to 2-D space, project ject is performing a memory task. is required to create new , these boundaries to the flat maps, Echo-planar imaging (EPI) is a one would expect medial temporal ac- thus defining the structure of the flat method of MR imaging that optimizes tivity when subjects are instructed to maps (Fig. 3D). Curves are fit to the temporal resolution by acquiring an encode novel stimuli. In monkeys, re- boundaries for easier viewing of acti- entire two dimensional slice after a searchers have found temporal lobe vation maps. single RF pulse (Cohen, 1996), and it neurons that respond greater to novel

Figure 5. Single-subject correlation map superimposed on anatomy in (A) coronal space, anterior and posterior slices; and (B) flat space, right and left hippocampi. Blue pixel outlines indicate areas of low sensitivity. Reproduced from Zeineh et al., 2000b. with permission of the publisher. TUTORIAL THE ANATOMICAL RECORD (NEW ANAT.) 117

CA23DG. This offered concrete evi- dence that both the hippocampus and adjacent cortical areas are involved in novelty-encoding processes. The procedure was applied to seven other subjects with results similar to that of Figure 5. To better character- ize the signal changes, the time course for each region was computed (Fig. 6). That is, in parahippocampal cortex as an example, the time series was aver- aged across all pixels in the parahip- pocampal cortex over all subjects in Figure 6. Time series of ROIs averaging across subjects and hemispheres. The vertical scale both hemispheres (Zeineh et al., is in units of percent change, and the horizontal scale represents time with a shaded 2000b). The resulting time series background indicating novel blocks and a white background indicating repeat blocks. confirmed that both hippocampal and parahippocampal structures stimuli in comparison to repeated repeat blocks (Fig. 4). Before looking were active in the paradigm. Fur- stimuli (Riches et al., 1991). The nov- for such pixels, we perform motion- thermore, hippocampal structures elty-encoding task tests the idea that correction on the data by aligning all (Sub and CA23DG) exhibited a con- the medial temporal lobes encode volumes with the first volume in the siderable delay in the development novel stimuli by contrasting epochs of time series (Woods et al., 1998). This of a strong contrast between novel novel stimuli with epochs of repeated corrects for signal intensity changes and repeat stimuli, and activity was stimuli. The paradigm has been associated with movement of the significantly greater at the end com- known to elicit activation in the “hip- head. After coregistering the func- pared with the beginning of the par- pocampal region,” but has been lack- tional EPI data with the high-resolu- adigm. We propose that encoding ing a more precise localization to ei- tion structural data, we project each processes were occurring for the re- ther the hippocampus proper or the three dimensional volume onto flat peat stimuli during the first few adjacent parahippocampal regions space using the unfolding transforma- blocks, and only late in the paradigm (Tulving et al., 1994; Stern et al., 1996; tion to arrive at a two dimensional did encoding truly stop and the con- Gabrieli et al., 1997). This paradigm time series of images (Zeineh et al., trast became apparent. This can be thus served as a perfect target for our 2000b). After these processing steps, differentiated from the parahip- method. Each stimulus was a complex correlation maps are computed to pocampal and fusiform response, color image of indoor or outdoor identify pixels that have a boxcar time which is more consistent with direct scenery, and subjects were viewing series (Cohen, 1997). For each pixel, visual stimulation. Using this tech- the images with magnet-compatible we calculate the Pearson correlation nique, we have been able to not only goggles. There were 10 alternating between that pixel’s time series and a pinpoint the localization of activa- blocks of novel and repeated stimuli, smoothed boxcar function (smoothed tion within the medial temporal lobe and blocks were 40 sec long with an and delayed to model a typical hemo- but also discern differences in tem- image was presented every 2.5 sec dynamic response, see Fig. 4), and we poral characteristics. (Fig. 4). During the novel blocks, sub- threshold the correlation map and su- jects viewed color pictures of complex perimpose it on the anatomical map. WARPING DATA IN FLAT SPACE scenery they had not seen previously, A single subject’s correlation map is and during the repeat blocks, subjects illustrated on Figure 5, both in the In addition to offering a superior vi- viewed the same two pictures repeat- coronal plane and in flat space, with sualization of activation patterns, edly. To ensure subjects paid atten- red indicating positively activated pix- transforming subjects to flat space fa- tion to both the novel and repeat els, or pixels positively correlated cilitates the comparison of data across blocks, we instructed subjects to press above a 0.25 threshold with the subjects. In flat space, the relationship different mouse buttons for indoor smoothed boxcar waveform, and blue between the pattern of activation and and outdoor scenes. indicating negatively activated pixels, the anatomy is clear, and the ana- The hypothesis is that, during novel or pixels with a negative correlation tomic consistency across subjects is blocks, subjects will engage in more with the smoothed boxcar waveform. adequate to facilitate inspection by encoding than during repeat blocks, Although activation can be discerned eye. Furthermore, by using computa- and involved medial temporal areas posteriorly in the coronal images, it is tional warping techniques, we can will be more metabolically active and much easier to view the location and align data across subjects into the have an increase in BOLD signal. In extent of activation on the flat maps. same space (see also Toga and other words, pixels that contain veins On both sides, activation lined the col- Thompson, 2001, for discussion of that drain active regions will have a in the fusiform gyrus warping). Typically, researchers align “boxcar” temporal pattern of signal and parahippocampal cortex, and on subjects in 3D volumetric space and intensity with increases during the the left hemisphere, activation in- perform group statistics to increase novel blocks and baseline during the cluded the subiculum and fields their power. The alignment is imper- 118 THE ANATOMICAL RECORD (NEW ANAT.) TUTORIAL

Figure 7. A: Creating a right hippocampal template by averaging the 9 boundaries across subjects. B: Warping a time series image to the template. C: Novelty activation maps produced by random-effects analyses of group data (n ϭ 8). fect, however, because of anatomic so that each anatomical curve con- space and perform either fixed-effects variability across subjects, and 3D tains an equal number of uniformly or random-effects analyses (Worsley smoothing is applied to minimize spaced points, and average the resa- and Friston, 1995; Holmes and Fris- the impact; the net effect is a loss in mpled boundaries over all of the sub- ton, 1998). resolution and localization confi- jects (Fig. 7A bottom) (Thompson et We chose to employ a random-ef- dence. Through segmentation of al., 2000). This set of averaged curves fects analysis to generate maps of ac- individual subjects, we account for thus forms the flat hippocampal tem- tivation representative of the popula- anatomical differences, and by un- plate, and it is representative of the tion: for each pixel and in each folding, we display individual sub- anatomy of our subject population. A subject, the slope of the regression jects in 2-D space. By warping data transform is computed that warps was calculated with the pixel’s time to a flat template, we can compute each subject’s anatomical image to series as the ordinate (Y) and the powerful group statistics while the template. This transform exactly smoothed-boxcar waveform as the ab- maintaining the high resolution nec- overlays each of the nine boundaries scissa (X). Computing a t-statistic that essary to distinguish medial tempo- and continuously warps the entire im- tests the significance of the slope in ral subregions (Fischl et al., 1999b; age while minimizing changes in area each pixel averaged across subjects re- Zeineh et al., 2000a). as measured by a warping energy sulted in a t-statistic activation map The first step is to create an average function (Thompson and Toya, 1998). (Holmes and Firston, 1998). As ran- hippocampal template based on the Application of this transform to each dom-effects analyses are very conser- anatomic boundaries of all of our sub- time series image (Fig. 7B) results in a vative, an uncorrected threshold of jects (n ϭ 8). We take each of the full time series of data in a common 0.05 was suitable for creating the curves fitted to these boundaries in space. One can then employ the gen- overlay maps depicted in Figure 7. flat space (Fig. 7A top), resample them eral linear model on each pixel in flat These maps illustrate clear activation TUTORIAL THE ANATOMICAL RECORD (NEW ANAT.) 119 of the parahippocampal gyrus and thank James Olds for his sponsorship Fischl B, Sereno MI, Dale AM. 1999a. Cor- fusiform gyrus bilaterally, as well as of our abstract at the 2000 Experi- tical surface-based analysis: II. Inflation, subicular and CA23DG activity, con- mental Biology Conference in San Di- flattening, and a surface-based coordi- nate system. NeuroImage 9:195–207. sistent with the data from the group ego that has led to the writing of this Fischl B, Sereno MI, Tootell RB, Dale AM. time courses. article. For their careful comments on 1999b. High-resolution intersubject av- this manuscript, we would like to eraging and a coordinate system for the thank Ho Lee, Richard Albistegui- cortical surface. Hum Brain Mapp CONCLUSIONS 8:272–284. DuBois, and one anonymous re- Fox PT, Raichle ME. 1986. Focal physio- Our study has shown that the contro- viewer. We would also like to ac- logical uncoupling of cerebral blood flow versial novelty-encoding paradigm knowledge Brian Wandell for the use and oxidative metabolism during so- elicits activation in both the hip- of his segmentation and unfolding matosensory stimulation in human sub- pocampus proper and parahippocam- software. jects. Proc Natl Acad Sci USA 83:1140– pal cortex (Zeineh et al., 2000b). We 1144. Gabrieli JDE, Brewer JB, Desmond JD, have further dissociated the temporal LITERATURE CITED Glover GH. 1997. Separate neural bases responses from these two regions and of two fundamental memory processes Aggleton JP, Brown MW. 1999. Episodic in the human medial temporal lobe. Sci- measured a significant response la- memory, amnesia, and the hippocam- ence 276:264–266. tency in the hippocampus proper; this pal-anterior thalamic axis. Behavioral & Holmes A, Friston K. 1998. Generalizabil- Brain Sci 22:425–489. may concord with the cessation of en- ity, random effects & population infer- Amaral DG, Insausti R. 1990. Hippocam- coding processes for repeated stimuli. ence. Neuroimage 7:S754. pal formation. In: Praxinos G, editor. Milner B, Squire LR, Kandel ER. 1998. The application of cortical unfolding The human nervous system. San Diego: Cognitive neuroscience and the study of techniques to fMRI facilitated this ac- Academic Press. p 711–755. memory. 20:445–468. complishment by delivering high- Bennett MR. 2000. The of long Mori S, Barker PB. 1999. Diffusion mag- term potentiation of transmission at syn- power group activation maps with a netic resonance imaging: its principle apses. Prog Neurobiol 60:109–137. resolution that maintains the distinc- and applications. Anat Rec (New Anat) Bohbot VD, Kalina M, Stepankova K, 257:102–109. tion among the different anatomic Spackova N, Petrides M, Nadel L. 1998. Murray EA, Bussey TJ. 2000. Perceptual- subregion. Spatial memory deficits in patients with mnemonic functions of the perirhinal lesions to the right hippocampus and to cortex. Trends in Cog Sci 3:142–151. the right parahippocampal cortex. Neu- O’Keefe J, Nadel L. 1978. The hippocam- FUTURE DIRECTIONS ropsychologia 36:1217–1238. pus as a . London: Oxford Buzsa´ki G. 1996. The hippocampo-neocor- University Press. p. 190–230. With the tools to localize activation to tical dialogue. Cereb Cortex 6:81–92. specific substructures within the hip- Cherry S, Phelps M. 1996. Imaging brain Rausch R, Babb TL. 1993. Hippocampal pocampus, we will begin to engage in function with positron emission tomog- neuron loss and memory scores before raphy. In: Toga A, Mazziotta J, editors. and after temporal lobe surgery for epi- focused hypotheses concerning hip- lepsy. Arch Neurol 50:812–817. pocampal function in memory. These Brain mapping: the methods. San Diego: Academic Press. p 191–221. Riches IP, Wilson FA, Brown MW. 1991. include differential specificity of the Cohen M. 1996. Rapid MRI and functional The effects of visual stimulation and perirhinal, entorhinal and parahip- applications. In: Toga A, Mazziotta J, ed- memory on neurons of the hippocampal pocampal cortices as well as the ele- itors. Brain mapping: the methods. San formation and the neighboring parahip- pocampal gyrus and inferior temporal ments of the hippocampus proper. As Diego: Academic Press. p 223–255. Cohen M, Bookheimer S. 1994. Localiza- cortex of the . J Neurosci MRI hardware advances, software tion of brain function using magnetic 11:1763–1779. techniques such as cortical unfolding resonance imaging. Trends Neurosci 17: Roland P. 1993. Brain activation. New will bring us closer to the kind of data 268–277. York: Wiley-Liss. p. 1–82. currently obtained only in animal Cohen MS. 1997. Parametric analysis of Rolls ET. 1999. and the representation of place in the primate studies. The convergence of human fMRI data using linear systems methods. NeuroImage 6:93–103. hippocampus. Hippocampus 9:467–480. imaging with animal electrophysiol- Cohen NJ, Ryan J, Hunt C, Romine L, Rolls ET, O’Mara SM. 1995. View-respon- ogy will propel the neuroscientific Wszalek T, Nash C. 1999. Hippocampal sive neurons in the primate hippocampal study of memory by enabling neuro- system and declarative (relational) mem- complex. Hippocampus 5:409–424. scientists to generate and test models ory: summarizing the data from func- Scoville WB, Milner B. 1957. Loss of recent tional neuroimaging studies. Hippocam- memory after bilateral hippocampal le- of memory that span the gamut of pus 9:83–98. sions. J Neurol Neurosurg Psychiatry 20: systems to . Drury HA, Van Essen DC, Anderson CH, 11–21. Lee CW, Coogan TA, Lewis JW. 1996. Sereno MI, Dale AM, Reppas JB, Kwong Computerized mappings of the cerebral KK, Belliveau JW, Brady TJ, Rosen BR, ACKNOWLEDGMENTS cortex: a multi-resolution flattening Tootell RB. 1995. Borders of multiple This work was supported by NIH Na- method and a surface-based coordinate visual areas in humans revealed by func- tional Research Service Award Grant system. J Cog Neurosci 8:1–28. tional magnetic resonance imaging. Sci- ence 268:889–893. MH12167 from the National Insti- Engel SA, Glover GH, Wandell BA. 1997. Retinotopic organization in human vi- Squire LR. 1992. Memory and the hip- tutes of Mental Health, NIH Medical sual cortex and the spatial precision of pocampus: a synthesis from findings Scientist Training Program Grant functional MRI. Cereb Cortex 7:181–192. with rats, monkeys, and humans. Psy- GM08042, the UCLA Affiliates, the Ah- Engel SA, Rumelhart DE, Wandell BA, Lee chol Rev 99:195–231. manson Foundation, the Pierson- AT, Glover GH, Chichilnisky EJ, Shadlen Stern CE, Corkin S, Gonza´lez RG, Guima- MN. 1994. fMRI of human visual cortex raes AR, Baker JR, Jennings PJ, Carr CA, Lovelace Foundation, the Jennifer- [letter] [published erratum appears in Sugiura RM, Vedantham V, Rosen V. Jones Simon Foundation, and the Nature 1994 Jul 14;370:106]. Nature 369: 1996. The par- Tamkin Foundation. We would like to 525. ticipates in novel picture encoding: Evi- 120 THE ANATOMICAL RECORD (NEW ANAT.) TUTORIAL

dence from functional magnetic reso- Thompson PM, Woods RP, Mega MS, Toga Woods RP, Grafton ST, Holmes CJ, Cherry nance imaging. Proc Natl Acad Sci. USA AW. 2000. Mathematical/computational SR, Mazziotta JC. 1998. Automated im- 93:8660–8665. challenges in creating deformable and age registration: I. General methods and Suzuki WA, Amaral DG. 1994a. Perirhinal probabilistic atlases of the human brain. intrasubject, intramodality validation. and parahippocampal cortices of the Hum Brain Mapp 9:81–92. J Comput Assist Tomogr 22:139–152. monkey: cortical afferents. Toga A, Thompson PM. 2001. Maps of the Worsley KJ, Friston KJ. 1995. Analysis of J Comp Neurol 350:497–533. brain. Anat Rec (New Anat) 265:37–53. fMRI time-series revisited—again [com- Suzuki WA, Amaral DG. 1994b. Topo- Tulving E, Markowitsch HJ, Kapur S, ment]. NeuroImage 2:173–181. graphic organization of the reciprocal Habib R, Houle S. 1994. Novelty encod- connections between the monkey ento- ing networks in the human brain: Zeineh M, Thompson P, Engel S, Bookhei- rhinal cortex and the perirhinal and positron emission tomography data. mer S. 2000a. Averaging flat maps of hip- parahippocampal cortices. J Neuro- Neuroreport 5:2525–2528. pocampal activity across subjects. Neu- science 14:1856–1877. Van Essen D, Drury H, Joshi S, Miller M. roImage 11:S537. Teo PC, Sapiro G, Wandell BA. 1998. Seg- 1998. Functional and structural map- Zeineh MM, Engel SA, Bookheimer SY. menting cortical gray matter for func- ping of human cerebral cortex: solutions 2000b. Application of cortical unfolding tional MRI visualization. In: IEEE 6th are in the surfaces. Proc Natl Acad Sci techniques to functional MRI of the hu- International Conference on Computer USA 95:788–795. man hippocampal region. NeuroImage Vision, New Delhi, India, 292–297. Van Essen DC, Drury HA. 1997. Structural 11:668-683. Thompson P, Toga AW. 1998. Anatomical- and functional analyses of human cere- Zola-Morgan S, Squire LR, Amaral DG. ly-driven strategies for high-dimensional bral cortex using a surface-based atlas. 1986. Human amnesia and the medial brain image warping and pathology de- J Neurosci 17:7079–7102. temporal region: enduring memory im- tection. In: Toga AW, editor. Brain warp- Wandell BA, Engel SA, Hel-Or HZ. 1996. pairment following a bilateral lesion lim- ing. San Diego: Academic Press. p 311– Creating images of the flattened cortical ited to field CA1 of the hippocampus. 336. sheet. Invest Ophthal Vis Sci 37:S1081. J Neurosci 6:2950–2967.