M Pathway and Areas 44 and 45 Are Involved in Stereoscopic Recognition Based on Binocular Disparity

Japanese Journal of Physiology, 52, 191–198, 2002

Tsuneo NEGAWA, Shinji MIZUNO*, Tomoya HAHASHI, Hiromi KUWATA†, Mihoko TOMIDA, Hiroaki HOSHI*, Seiichi ERA, and Kazuo KUWATA

Departments of Physiology, * Radiology, and † Nursing Course, Gifu University School of Medicine, Gifu, 500–8705 Japan

Abstract: We characterized the visual path- was reported that these regions were inactive ways involved in the stereoscopic recognition of during the monocular stereopsis. To separate the the random dot stereogram based on the binocu- specific responses directly caused by the stereo- lar disparity employing a functional magnetic res- scopic recognition process from the nonspecific onance imaging (fMRI). The V2, V3, V4, V5, in- ones caused by the memory load or the inten- traparietal sulcus (IPS) and the superior temporal tion, we designed a novel frequency labeled sulcus (STS) were significantly activated during tasks (FLT) sequence. The functional MRI using the binocular stereopsis, but the inferotemporal the FLT indicated that the activation of areas 44 gyrus (ITG) was not activated. Thus a human M and 45 is correlated with the stereoscopic recog- pathway may be part of a network involved in the nition based on the binocular disparity but not stereoscopic processing based on the binocular with the intention artifacts, suggesting that areas disparity. It is intriguing that areas 44 (Broca’s 44 and 45 play an essential role in the binocular area) and 45 in the left hemisphere were also ac- disparity. [Japanese Journal of Physiology, 52, tive during the binocular stereopsis. However, it 191Ð198, 2002]

Key words: stereoscopic recognition, binocular disparity, functional MRI, M pathway, frequency labeled task sequence.

Environmental information is generally sent through Lesion analyses in humans [9–11] indicate that the several pathways that are parallel to the physically parietal cortex plays a major role with a suggestion of separated areas of the cerebral cortex, communicating right-hemisphere dominance for stereoscopic recogni- with one another to produce the integration. A stereo- tion. For example, patients with anterior temporal- scopic recognition is a typical example of the integra- lobe lesions demonstrate impaired global stereopsis tion of two-dimensional (2D) images of a three-di- but intact local stereoacuity [9]. Patients with right- mensional (3D) object projected in slightly different sided temporal lobe lesions perform signiﬁcantly planes into a single-image having a 3D effect. Stereo- worse than those with left-sided lesions in perceiving scopic vision has been studied intensively at both the depth when it is cued by ambiguous disparities [10]. theoretical and computational levels [1–3]. Since Right hemisphere superiority has been suggested from stereoscopic recognition is known to exhibit a diverse works with normal human subjects [11]. Binocularly and moderately complicated phenomena, which are driven V3 complex cells project to the parieto-occipi- well suited for study at the neurophysiological level, tal area of the superior parietal lobule and lateral and human brain mapping involved in stereoscopic recog- posterior intraparietal areas in the lateral bank and nition have been of growing interest [4–8]. fundus of the intraparietal sulcus in the cat [12] and

Received on June 28, 2001; accepted on February 6, 2002 Correspondence should be addressed to: Kazuo Kuwata, Department of Physiology, Gifu University School of Medicine, 40 Tsukasa- machi, Gifu, 500–8705 Japan. Tel: ϩ81–58–267–2227, Fax: ϩ81–58–267–2962, E-mail: [email protected]

Japanese Journal of Physiology Vol. 52, No. 2, 2002 191 T. NEGAWA et al. the macaque monkey [13]. These cortical areas are tem equipped with a superconductive magnet operat- thought to be involved in the coding of spatial posi- ing at 1.5 T was used to perform MRI studies. Thir- tion or stereopsis, or both. However, the data on the teen consenting normal male volunteers participated. localization of this function in humans are still less All subjects were able to interconvert between 3D and specific, and the relationships with the established vi- 2D within 1 s. They were right-handed adult native sual pathways [14] have not yet been reported. Here, Japanese males 30–45 years old. Their dextral ability we first reported the human brain regions activated was confirmed by the Edinburgh inventory [19]. during stereopsy of the random dot stereogram [15] The paradigm used involved one 3D image (depth) and characterized the dominant visual pathways [14], and one 2D image (form), using two categories of using statistical parametric mapping (SPM) (Well- stimuli. The subjects wore a blue filter over the left come Department of Cognitive Neurology, Institute of eye and a red one over the right during the experi- Neurology, UCL, UK) of functional magnetic reso- ment. In the 3D condition, a colored random dot stere- nance imaging (fMRI). ogram (Fig. 1a) depicting a square at 100% binocular One of the important methodological difficulties is correlation was presented on the screen through a the separation of the specific responses from the unex- computer-controlled projector, MX-PJ1 (Minolta, pected nonspecific ones triggered by the tasks, possi- Osaka, Japan), using a software Visual Basic (Mi- bly unconsciously, such as those caused by the mem- crosoft, Washington, USA). The subjects could see ory load [16] or the intention itself [17]. To overcome only monochrome random dots even for a colored ran- this, we firstly designed the task sequence termed fre- dom dot stereogram because of filters. On the other quency labeled tasks (FLT) (see MATERIALS AND hand, in 2D condition, monochrome random dots with METHODS); thereby we can separate the task-specific the square frame (Fig. 1b) were presented. The sub- responses from the nonspecific ones. The main idea is jects viewed a screen on which the 3D and 2D images the labeling of the task-specific brain responses by were presented alternatively. Each session consisted of task-specific frequencies, which are incorporated into multiple 30-s epochs in a boxcar configuration. the task sequence, in analogy with the multidimen- A gradient echo planar imaging (EPI) sequence was sional heteronuclear NMR pulse sequence [18], where used on a Horizon 1.5 MRI system with the following each signal is labeled by the nuclear-specific reso- acquisition parameters: flip angle, 90°; field of view, nance frequencies. 320 mmϫ320 mm; matrix, 128ϫ128 pixels with 14 In FLT, task is labeled by the specific frequency, slices; slice thickness, 8 mm; interslice gap, 3 mm; and brain responses caused by the task are detected by repetition time, 3 s. The spatial resolution was approx- using the correlation with the same frequency. On the imately 2.5 mmϫ2.5 mmϫ11 mm. Four hundred and other hand, nonspecific responses are not specifically twenty images were acquired during 90 s. The ses- correlated with any task-specific frequency. This is sions that showed brain motion exceeding 0.6 mm one of the novel aspects of this report. FLT may es- were performed again to avoid so-called fictitious ac- sentially distinguish brain regions activated by the tivation resulting from pixel misalignment. The fMRI stereoscopic recognition and those activated by the ar- time series data consisting of consecutive echo-planar tifacts. images for each slice were analyzed by the use of the SPM96 method on software MEDx Ver. 3.3 (Sensor MATERIALS AND METHODS Systems Inc., Virginia, USA). To increase S/N ratio, the data were smoothed with A Signa (GE Medical System, Horizon) imaging sys- a 5ϫ5ϫ11 mm full-width-at-half-maximum Gaussian

Fig. 1. Random dot stereogram utilized for the fMRI experiment. The subjects wore a blue ﬁlter over the left eye and a red one over the right eye during the experiment. The images were presented on the screen through a computer controlled projector, MX-PJ1 (Minolta), by the use of a software Visual Basic (Mi- crosoft). (a) Colored random dot stereogram. (b) Monochrome random dots with the square frame.

192 Japanese Journal of Physiology Vol. 52, No. 2, 2002 Functional MRI of Stereopsy kernel, and a high-pass filter and proportional global normalization (mean: 100) were applied. A statistical analysis was performed by using a delayed (6 s) boxcar model function in the context of the general linear model used in SPM96. Specific effects were tested by applying appropriate linear contrasts to the parameter estimates for each condition, resulting in a t-statistic for each voxel. Here the t-statistics, which were trans- formed to Z scores, constitute an activation map. These images were interpreted by referring to the probabilistic behavior of a Gaussian field. Functional MRI results were presented with contrast between two sets of images described earlier and showed activated areas that conformed to the statistical criteria of sig- nificance (pϽ0.05, corrected by multiple comparison). The subtractive nature of the process effectively eliminated the activation areas common to the two types of visual stimuli, such as the primary visual cortex. Reproducibility of the analyzed data was confirmed individually for each subject. The anatomic identification of activated areas was performed individually by mapping these areas onto the subject’s own anatomic Fig. 2. General block diagrams of the frequency-la- T1-weighted images obtained with identical coordi- beled task sequences and correlation analyses. (a) Dia- nates. After identification of the neuroanatomic site of gram of TASK A (stereoscopic recognition based on binoc- the areas of activation performed individually for each ular disparity). (b) Diagram of TASK B (left-finger opposition). (c) Correlation I for the detection of activated regions subject, the activated areas from all eight subjects by TASK A. (d) Correlation II for the detection of activated were calculated by a random effect model and mapped regions by TASK B. (e) Correlation III for the detection of in 3D coordinates onto the best fitted gyrus/sulcus nonspecifically activated regions by TASKS A and B. patterns of normalized images with standard coordinates [20]. Table 1. Summary of single subject analyses. The task sequences designed for the analysis of the Brodmann’s No. of stereoscopic recognition according to the FLT strategy Brain region area subjects are shown in Fig. 2a and b. The functional MRI acquisition parameters were the same as the usual fMRI ex- Inferior frontal gyrus (R) 45 10 periment described above. Task A has a frequency of Inferior frontal gyrus (L) 45 12 0.017 Hz, and B has 0.033 Hz. Here we employed the Supramarginal gyrus (R) 40 (IPS) 6 stereoscopic recognition based on the binocular dis- Supramarginal gyrus (L) 40 (IPS) 8 parity and the left finger opposition for tasks A and B, Cuneus (R) 19 (V3, V4) 8 Cuneus (L) 19 (V3, V4) 8 respectively. The subjects were asked to conduct both Medial occipitotemporal 18 (V2) 6 tasks A and B simultaneously, according to the guid- gyrus (R) ance displayed on the screen. For the analysis, we used three kinds of correlation schemes shown in Fig. 2c–e. Correlation schemes I, of tasks become large, because different frequencies II, and III were employed for discovering the responsi- are independent in terms of Fourier transformation. ble brain areas for the stereoscopic recognition, the The purpose of FLT is to introduce the task-frequency finger-tapping motion, and the nonspecific responses, axis in the fMRI experiment to discriminate the multi- respectively. ple responses originated from different brain func- It should be noted that these correlations are not tions, which are labeled by different task frequencies. mutually complementary or orthogonal, because the effect of intention may not be orthogonal to that of RESULTS any specific task. However, these will tend to be independent when the number of images and the number The number of subjects showing the significant acti-

Japanese Journal of Physiology Vol. 52, No. 2, 2002 193 T. NEGAWA et al.

(a) (b)

356 7 11 7

6 11 12 2-d R 9 10 Fig. 3. Activated brain re- 1 8 gions. Significantly activated brain regions (SPM(Z), p cor- 5 Ͻ 2-v 812 3 1 rected 0.05) mapped onto 4910 4 2-d 2-v the glass brain (Talairach- (c) Tournoux), showing the main 5101112 effect of the binocular stereop- 4 sis versus the baseline obtained in the multiple-subject analysis. 1, V1; 2-d, V2d (dor- 3 sal); 2-v, V2v (ventral); 3, V3; 4, V4; 5, V5 (MT, middle tem- 2-d poral); 6, IPS (intraparietal sulcus); 7, FEF (frontal eye field); 8, ITG (inferotemporal gyrus); 1 9, STS (superior temporal sulcus); 10, area 22; 11, area 44; 12, area 45. (a) Sagittal view. (b) Coronal view. (c) Axial 2-v 9 6 7R 8 view. vation in the corresponding Brodmann’s area (pϽ Table 2. Standard coordinates of the activated brain 0.05, corrected) obtained by the fMRI single-subject areas during the binocular stereopsis. analyses are summarized in Table 1. This analysis was conducted for a comparison with the PET results [4]. Standard coordinate We analyzed the activation patterns of all subjects, RL using a random effect model. The primary advantage of this analysis is that it allows inferences to be made V2d 34.0 , Ϫ62.0, 12.0 Ϫ42.0, Ϫ62.0, 8.0 about humans in general [21]. Composite images of (dorsal) (3.13)* (2.84) Ϫ Ϫ Ϫ the activated areas are displayed on a 3D glass brain V3 34.0 , 64.0, 12.0 24.0, 76.0, 8.0 (2.92) (2.99) in Fig. 3. Standardized coordinates of the maximum Z V4 24.0, Ϫ66.0, Ϫ12.0 Ϫ36.0, Ϫ68.0, Ϫ8.0 score in each activated area were summarized in Table (3.37) (3.24) 2. Significant activation was mainly observed at dorsal V5/MT 36.0, Ϫ56.0, Ϫ12.0 Ϫ40.0, Ϫ68.0, Ϫ8.0 V2 (V2d) and V3, V4 and V5/middle temporal (MT) (3.80) (4.07) [22], superior temporal sulcus (STS) [23], area 44 and STS 36.0, Ϫ62.0, 16.0 Ϫ22.0, Ϫ78.0, 8.0 45 [20]. (3.06) (3.01) IPS 28.0, Ϫ58.0, 36.0 EPI intensity changes (%) were shown as a function (3.09) of time (s) in Fig. 4. Intensities at V3 (᭺), those at Area 44 Ϫ58.0, 18.0, 28.0 Broca’s area (᭹) in Fig. 4a, and those at V5 (MT) (᭺) (3.31) and STS (᭹) in Fig. 4b altered quite coherently as a Area 45 Ϫ60.0, 18.0, 24.0 function of time. However, those at inferotemporal (3.14) ᭡ gyrus (ITG) ( ) in Fig. 4b exhibited no significant * Values in parentheses indicate the Z scores in SPM change. analysis. Figure 5a, b, and c show the statistical images (SPM(Z), pϽ0.05, corrected) obtained by the use of the FLT scheme with correlations I, II, and III, respec-

194 Japanese Journal of Physiology Vol. 52, No. 2, 2002 Functional MRI of Stereopsy

Fig. 4. Phase proﬁles of EPI intensities. (a) Echo planar image intensity change at V3 (᭺) (Talairach coordinate: 34, Ϫ64, 12) and Broca’s area (᭹) (Ϫ58, 18, 28) as a function of time. (b) Echo planar image intensity change at V5 (MT) (᭺) (36, Ϫ56, Ϫ12), STS (᭹) (36, Ϫ62, 16) and ITG (᭡) (Ϫ44, Ϫ48, Ϫ12) as a function of time.

(a) (c)

(b)

Fig. 5. Effects of FLT. Signiﬁcantly activated brain regions (SPM(Z), p corrected Ͻ0.05) mapped onto the glass brain (Talairach-Tournoux), during application of the FLT sequence shown in Fig. 2a and b. (a) Activated regions detected by use of correlation scheme I in Fig. 2c; (b) scheme II in Fig. 2d; and (c) scheme III in Fig. 2e.

Japanese Journal of Physiology Vol. 52, No. 2, 2002 195 T. NEGAWA et al.

Fig. 6. P and M pathways. Dia- grammatic representation of the projections of the P and M pathways to the specialized areas of the striate and prestriate visual cortex and the interconnections between them. Activated regions during the stereoscopic recognition based on the binocular disparity are colored gray. tively. The SPM image obtained by the use of correla- Table 3. Comparison of the activated regions (؉) and tion I (Fig. 5a) shows the selective activation of the the nonactivated regions (؊) between the binocular and visual areas and Broca’s area. On the other hand, the the monocular stereopsis.* obtained image by the use of correlation II (Fig. 5b) Binocular stereopsis Monocular stereopsis* shows the activation of the primary motor area typi- cally observed for the finger opposition. The statisti- RL RL cally significant regions obtained by the use of correlation III may reflex the nonspecific responses caused V2d Ϫϩ ϪϪ ϩϩ ϪϪ by tasks A and B. However, the activation of areas 44 V3 V4 ϩϩ ϪϪ and 45 was not detected by the use of either correla- V5 (MT) Ϫϩ ϩϩ tion II or correlation III (Fig. 5b and c). STS ϩϩ ϩϩ Figure 6 shows the simplified diagrammatic repre- IPS ϩϪ ϩϩ sentation of the projection of the M and P pathways to ITG ϪϪ ϪϪ the specialized areas of the striate and prestriate visual Area 44 Ϫϩ ϪϪ Ϫϩ ϪϪ cortex [5, 14, 22, 23]. Activated brain regions during Area 45 the stereoscopic recognition based on the binocular * From Inui et al. [8]. disparity were colored gray, and nonactivated regions were colored white. It is intriguing, that IPS [5] is involved, but ITG [20] is not. viewer a potential motor action, to localize the cogni- Significant differences also exist between the tive processing of visual images having 3D properties. monocular and the binocular stereopsis, as listed in The areas activated by stereopsis were shown in Fig. 3 Table 3. V3, V4, and areas 44 and 45 in the left hemi- and summarized in Tables 1 and 2. Although slight sphere are not involved in monocular stereopsis [8]. right hemisphere dominance was suggested, laterality could not be statistically confirmed in this study. DISCUSSION The activation of areas 18, 19, 40, 44, and 45 with right hemispheric dominance were observed in most M-pathway is involved in the stereoscopic subjects (8–13 of 13), as shown in Table 1. Using recognition. The main objective of the current PET, Ptito et al. [4] reported that areas 17, 18, and 45 study was to localize the cortical area involved in the in the right hemisphere were activated during stereop- visual perception in 3D stereopsis. The alternating 3D sis. Thus the results of fMRI were almost consistent versus 2D visual stimuli of random-dot stereogram with those of PET [4]. Because of the high spatial res- used in this study were selected on the basis of their olution and sensitivity to the transient phenomena in purely perceptual contents, without eliciting in the fMRI, we further addressed the relationships between

196 Japanese Journal of Physiology Vol. 52, No. 2, 2002 Functional MRI of Stereopsy the activated brain regions and the several distinct vi- gyrus (area 4). They also observed that a higher mem- sual pathways [14]. ory load increased the activity of the relevant brain re- It is known that the retina has two major classes of gions. ganglion cells, P␣ and P␤ cells, which project sepa- Our results suggest that the activation of areas 44 rately to the cortex through the lateral geniculate nu- and 45 is not relevant to so-called working memory in cleus, i.e., P␣ cells through its magnocellular (M) lay- the utilized task sequence. This observation suggests ers and P␤ cells through its parvocellular (P) layers. that these areas are possibly involved in some kind of Figure 6 shows the simplified diagrammatic represen- information processing, such as depth calculation em- tation of the projection of the M and P pathways to the ploying information on the discrimination of binocu- specialized areas of the striate and prestriate visual lar disparity, though the detailed mechanism is un- cortex [5, 14, 22, 23]. Activated brain regions during known. the stereoscopic recognition based on the binocular As shown in Fig. 2a and b, tasks A and B are con- disparity were colored gray, and nonactivated regions ducted simultaneously but with two different on–off were colored white. It is interesting that IPS [5] is frequencies. These tasks are essentially independent involved but ITG [20] is not, indicating that the M in the task frequency domain. The statistical tests are pathway is predominantly involved in the stereoscopic done on the acquired images by using the three kinds recognition. of correlations shown in Fig. 2c–e. Correlation I and Neuroimaging studies implicate several anatomic correlation II are independent; correlation III is not sites in this function [4–8]. Gulyás et al. [5] reported completely independent of either of them, but reflects that the discrimination of stereoscopic disparity infor- the common activation area caused by both tasks A mation takes place in the striate cortex, the peristriate and B. The effect of intention is considered to be par- cortex, the parietal lobe, the prefrontal cortex, and the allel to correlation III, since the intention may not be cerebellum. Orban et al. [6] reported that the human specifically connected to task A or B, but related non- MT/V5 is involved in extracting depth from motion. specifically to both tasks. As demonstrated here, this Kwee et al. [7] reported that IPS is involved in the FLT scheme is quite useful to separate task-specific stereopsis of MR angiography. Inui et al. [8] reported brain responses and nonspecific ones, especially for the symmetrical activation of premotor and parietal an analysis of complex brain functions, such as stereo- areas during Necker cube perception. Our findings are scopic recognition. basically consistent with the obtained results in this literature, but we particularly emphasize the involve- This study was supported in part by Grants-in-Aid for Sci- ment of neurons along the M-pathway with a slight entific Research from the Ministry of Education, Culture, right hemisphere dominance. Sports, Science and Technology of Japan (07680714, 10670039). Inclusion of the frequency axis in the task sequence (FLT). Significant differences exist between the monocular and the binocular stereopsis, as REFERENCES listed in Table 3. It is reported that V3, V4, and areas 1. Marr D and Poggio T: Cooperative computation of 44 and 45 in the left hemisphere were not involved in stereo disparity. Science 194: 283–287, 1976 monocular stereopsis [8]. To confirm further that 2. Poggio GF and Poggio T: The analysis of stereopsis. areas 44 and 45 are involved in stereoscopic recogni- Annu Rev Neurosci 7: 379–412, 1984 3. Archie KA and Mel BW: A model for intradendritic com- tion based on the binocular disparity, we applied the putation of binocular disparity. Nat Neurosci 3: 54–63, FLT method. 2000 In the FLT experiments, we observed the significant 4. Ptito A, Zatorre RJ, Petrides M, Frey S, Alivisatos B, activation of the areas 44 and 45 when correlation I and Evans AC: Localization and lateralization of stereo- was used (Fig. 5a), but the activation of area 44 and scopic processing in the human brain. Neuroreport 4: 1155–1158, 1993 45 could not be clearly seen when correlation II or III 5. Gulyás B and Roland PE: Binocular disparity discrimi- was used (Fig. 5c). This result indicated that areas 44 nation in human cerebral cortex: functional anatomy by and 45 are specifically associated with the stereo- positron emission tomography. Proc Natl Acad Sci USA scopic recognition, but not related to the finger-tap- 91: 1239–1243, 1994 ping motion or nonspecific intention artifacts. 6. Orban GA, Sunaert S, Todd JT, Hecke PV, and Marchal Recently, De Fockert et al. [16] reported that the G: Human cortical regions involved in extracting depth from motion. Neuron 24: 929–940, 1999 higher memory load is associated with the increased 7. Kwee IL, Fujii Y, Matsuzawa H, and Nakada T: Percep- prefrontal activity, i.e., inferior frontal gyrus (area 44), tual processing of stereopsis in humans: high-field the middle frontal gyrus (area 6), and the prefrontal (3.0-tesla) functional MRI study. Neurology 53:

Japanese Journal of Physiology Vol. 52, No. 2, 2002 197 T. NEGAWA et al.

1599–1601, 1999 16. De Fockert JW, Rees G, Frith CD, and Lavie N: The role 8. Inui T, Tanaka S, Okada T, Nishizawa S, Katayama M, of working memory in visual selective attention. Sci- and Konishi J: Neural substrates for depth perception ence 291: 1803–1806, 2001 of the Necker cube; a functional magnetic resonance 17. Hyder F, Phelps EA, Wiggins CJ, Labar KS, Blamire imaging study in human subjects. Neurosci Lett 282: AM, and Shulman RG: “Willed action”: a functional MRI 145–148, 2000 study of the human prefrontal cortex during a sensori- 9. Benton AL and Hécaen H: Stereoscopic vision in pa- motor task. Proc Natl Acad Sci USA 94: 6989–6994, tients with unilateral cerebral disease. Neurology 20: 1997 1084–1088, 1970 18. Ernst RR, Bodenhausen G, and Wokaun A: Principles 10. Rothstein TB and Sacks JG: Defective stereopsis in le- of Nuclear Magnetic Resonance in One and Two Di- sions of the parietal lobe. Am J Ophthalmol 73: mensions. Clarendon Press, Oxford, 1987 281–284, 1972 19. Oldﬁeld RC: The assessment and analysis of handed- 11. Lehmann D and Wächli P: Depth perception and loca- ness: the Edingburgh inventory. Neuropsychologia 9: tion of brain lesions. J Neurol 209: 157–164, 1975 97–113, 1971 12. Ptito M, Lepore F, and Guillemot J-P: Loss of stereopsis 20. Talairach J and Tournoux P: Co-Planar Stereotaxic following lesions of cortical areas 17–18 in the cat. Exp Atlas of the Human Brain, Thieme, New York, 1988 Brain Res 89: 521–530, 1992 21. Holmes AP and Friston KJ: Generalizability, random ef- 13. Gonzalez F, Krause F, Perez R, Alonso JM, and Acuña fects and population inference. NeuroImage 7: S754, C: Binocular matching in monkey visual cortex: single 1998 cell responses to correlated and uncorrelated dynamic 22. Hasnain MK, Fox PT, and Woldorff MG: Intersubject random dot stereograms. Neuroscience 52: 933–939, variability of functional areas in the human visual cor- 1993 tex. Hum Brain Map 6: 301–315, 1998 14. Zeki S and Shipp S: The functional logic of cortical 23. Van-Essen DC and Drury HA: Structural and functional connections. Nature 335: 311–317, 1988 analysis of human cerebral cortex using a surface- 15. Julesz B: Foundations of Cyclopean Perception, Uni- based atlas. J Neurosci 17: 7079–7102, 1997 versity of Chicago Press, Chicago, 1971

198 Japanese Journal of Physiology Vol. 52, No. 2, 2002