Representation of the Visual Field in the Human Occipital Cortex: A Magnetic Resonance Imaging and Perimetric Correlation

Objectives  To evaluate the retinotopic map of the human occipital cortex by correlating magnetic resonance imaging (MRI) findings with visual field defects in patients with occipital lobe infarcts and to assess the compatibility between our cliniconeuroimaging findings and the location of lesions predicted by the classic Holmes map and a revised map.

Methods  Magnetic resonance images were obtained in 14 patients with occipital lobe infarcts. Visual field analysis was performed with tangent screen, the Goldmann perimeter, and the Humphrey Field Analyzer. Based on the pattern of visual field deficit, the location of the lesion in the mesial occipital lobe in each patient was predicted using the Holmes map and other retinotopic maps of the occipital cortex. The predicted location of the lesion was then compared with its actual location shown on MRI to assess the compatibility between our data and the other maps. These maps determine retinotopic correlates of the medial occipital lobe, but they cannot establish correlates of the striate cortex (V1). The medial occipital representation of central vision was evaluated by regression analysis.

Results  The MRI correlations in this study confirmed gross estimates of the retinotopic organization of the occipital cortex. However, our findings did not correlate exactly with the Holmes map. We determined that the central 15^o of vision occupies 37% of the total surface area of the human medial occipital lobe. Based on our data, a refined retinotopic map is presented.

Conclusions  The resolution of conventional MRI testifies to its considerable value in localizing occipital lobe lesions. Our findings support, and refine, the Holmes map of the human occipital cortex.

MUCH OF our knowledge of the representation of the visual field in the human occipital cortex was derived from studies of patients with penetrating injuries to the head during wartime. Inouye,¹ Holmes and Lister,² Holmes,^3,4 and, later, Spalding⁵ correlated visual field deficits with the location of missile wounds in the occiput and produced retinotopic maps of the striate cortex.

The map presented by Holmes⁴ is widely adopted as a detailed representation of the visual field in the human striate cortex. It depicts a point-to-point localization of the contralateral hemifield of vision in the striate cortex. The upper field is represented in the inferior calcarine cortex, and the lower field is represented in the superior cortex. The central field occupies the posterior pole, whereas the peripheral field is mapped anteriorly.⁴ The macula, which subserves central vision, has a disproportionately large cortical representation⁴—up to 25% of the surface area of the striate cortex has been assigned to the central 15° of vision.⁶

The accuracy of the Holmes map was later supported by activation study of the visual cortex using positron emission tomography scanning⁷ and by cliniconeuroimaging correlation using computed tomography (CT).^8-11 Spector et al,¹¹ using CT to study patients with occipital lobe infarction, reported a good match between cliniconeuroimaging findings and the location of lesions predicted by the Holmes map. However, these studies were limited by the resolution of early CT scans. With the introduction of magnetic resonance imaging (MRI), Horton and Hoyt⁶ reexamined the Holmes map. By correlating MRI findings with homonymous field deficits in 3 patients with occipital lobe lesions, they found that the Holmes map did not correlate well with their findings.

Based on their findings, together with electrophysiologic data on Old World primates—which showed that in macaque monkeys as much as 70% of the total surface area of the striate cortex is occupied by the central 15° of vision^12,13—Horton and Hoyt⁶ proposed that the cortical magnification of the human striate cortex is scaled to that in the macaque striate cortex. They concluded that the Holmes map underestimated the cortical magnification of central vision, and they proposed a revised map.⁶ In that revised map,⁶ the human striate cortex was approximately an ellipse measuring about 80×40 mm. The area subserving central vision was expanded such that approximately 70% of the total surface area of the human striate cortex was assigned to the central 15° of vision.

We correlated MRI findings with the location and extent of visual field deficits in a series of patients with occipital lobe infarcts to determine the retinotopic representation of the human occipital cortex. To assess the accuracy of the Holmes map⁴ and a revised map,⁶ we compared our cliniconeuroimaging findings with the locations of lesions predicted by the 2 maps. Based on the findings in our patients, the percentage of the total surface area of the medial occipital cortex devoted to central vision was estimated, and a refined retinotopic map is presented.

Consecutive patients with visual field defects and occipital lobe lesions seen in the neuro-ophthalmology clinic at The Toronto Hospital, Toronto, Ontario, were screened using the central 30-2 threshold program of the Humphrey Field Analyzer (Allergan-Humphrey Instruments, San Leandro, Calif). Patients with incomplete homonymous hemianopia were included in the study. The central 30-2 program only tests the retinal threshold at 76 predetermined points within the central 30° of vision, with 19 points tested in each quadrant. For example, adjacent to the horizontal meridian, the tested points lie at approximately 2°, 8°, 14°, 20°, and 26° of eccentricity. Thus, the program does not provide information on the retinal threshold at approximately 3° to 7°, 9° to 13°, 15° to 19°, 21° to 25°, and 27° to 30° of eccentricity, although the gray-scale printout gives the impression that the threshold at every point is tested. In this study, all recruited patients were also tested with tangent screen and Goldmann perimetry to delineate more precisely the extent of visual field loss. If there was discrepancy among the 3 perimetric measurements, the mean value of the 3 measurements was used (in all study patients, the discrepancy among measurements was ≤5°).

Serial axial and sagittal T₁-weighted (repetition time=516-517 milliseconds and echo time=8-11 milliseconds) and T₂-weighted (repetition time=2200-4383 milliseconds and echo time=80-95 milliseconds, 2 separate acquisitions) images with a slice thickness of 5 mm were obtained in all recruited patients using the 1.5-T Signa imaging system, version 5.4.2 (General Electric Medical Systems, Milwaukee, Wis). The MR images were then processed on a workstation (SUN/SPARC 10 "Advantage Windows," General Electric Medical Systems). Patients with well-defined infarcts in the occipital cortex with a minimum duration of 3 months were included in the study. Patients with lesions that may have functional neurons remaining within the neuroimaged boundaries, such as edema or hemorrhage secondary to tumors, arteriovenous malformations, or primary intracerebral hemorrhages, were excluded. Using a software program (Advantage Window v.1.2, General Electric Medical Systems), the anterior and posterior extent of the infarct (as measured by its linear distance from the occipital pole) was independently determined by assessing T₁- and T₂-weighted images in axial (along the calcarine fissure) and midsagittal orientations by us. If there were discrepancies in measurements between T₁ vs T₂ images, orientations, or between us, the mean distance from the occipital pole was used (in all study patients, the discrepancy between measurements was ≤3 mm). Using the same software program, the infarcted and normal portions of the occipital cortex were traced out using a pointing device independently by us, and the corresponding surface areas were calculated by the planimetric program available in the software. If there was a discrepancy in measurements between us, the mean surface area was used (in all study patients, the discrepancy between measurements was ≤30 mm²).

To assess the accuracy of the Holmes map⁴ and a revised map,⁶ the location of the lesion in each patient was predicted using the 2 maps based on the patient's visual field defect. We then compared the predicted location of the lesion with its actual location on MRI to assess the compatibility between our data and the 2 maps.

To evaluate the cortical representation of central vision, the percentage of the surface area of the infarcted or noninfarcted (see below) occipital cortex was plotted against the degree of eccentricity from fixation.

In patients with posterior lesions that began at the occipital pole, the surface area of the lesions was used in the analysis. For example, patient 1 (Figure 1), who had an inferior scotoma extending from 2° to 10°, was shown to have an infarct extending from the occipital pole to 12 mm anteriorly above the calcarine fissure by MRI, which measured 300 mm². Assuming the total surface area of the average human striate cortex to be about 2500 mm² (based on postmortem specimens and after allowing for shrinkage),^6,14 the percentage of surface area of the lesion above the calcarine fissure was calculated to be 12%, which corresponded to an eccentricity of 10°. Because the area above and below the calcarine fissure is represented at any eccentricity, 24% (12%×2) of the total surface area was assigned to an eccentricity of 10°.

In patients with anterior lesions that began at some distance from the occipital pole, the surface area of the normal area of the occipital cortex posterior to the lesions was used. For example, patient 4 (Figure 2), who had an inferior quadrantanopia sparing the central 11°, was shown to have a lesion that began at 16 mm and extended anteriorly to 25 mm from the occipital pole above the calcarine fissure. The surface area of the normal cortex posterior to the lesion above the calcarine fissure was 375 mm², representing 15% of the total surface area and corresponding to an eccentricity of 11°. Therefore, 30% (15%×2) of the total surface area was assigned to an eccentricity of 11°.

Regression analysis of the relationship between eccentricity from fixation and the corresponding percentage surface area of the medial occipital cortex was performed. The percentage of the area of the occipital cortex subserving the central 15° of vision was then determined. Based on the cliniconeuroimaging findings in study patients, we constructed a new retinotopic map of the human occipital cortex.

Fourteen patients had well-defined infarcts in the occipital cortex on MRI scans and were included in the analysis (Table 1). There were 9 males and 5 females, with ages ranging from 17 to 74 years (mean, 50 years). The mean duration of lesions was 14 months (range, 3-84 months). All 14 patients had incomplete homonymous visual field defects by Humphrey perimetry (Figure 3), tangent screen, and Goldmann perimetry. Figure 4 shows the visual field defects on tangent screen, and a schematic representation of the location of the corresponding lesions, using the axial brain templates by Damasio and Damasio.¹⁵

The visual field defect, the locations of the lesions predicted by the Holmes⁴ map and a revised map⁶ based on visual field loss, and the actual locations of the lesions on MRI scans are shown in Table 2. For example, for patient 1, who has a right homonymous inferior quadrant scotoma from 2° to 10°, the Holmes map predicts a lesion to extend from the occipital pole to 9 mm anteriorly, whereas a revised map⁶ predicts it to extend from 8 mm from the occipital pole to 32 mm anteriorly. However, on MRI, the actual location of the lesion began at the occipital pole and extended to 12 mm anteriorly, which was in close agreement with the predicted location according to the Holmes⁴ map. Similarly, for patient 14, who has a left homonymous inferior quadrant scotoma sparing the central 6° (Figure 5, A and B), the Holmes map predicts a lesion to begin at 5 mm from the occipital pole and to extend rostrally to 25 mm from the pole; on the other hand, a revised map⁶ predicts it to begin at 23 mm from the occipital pole and to extend beyond 50 mm from the pole. However, on MRI (Figure 5, C), the actual lesion began at 9 mm from the occipital pole and extended to 35 mm from the pole, a finding that was in close agreement with the predicted location according to the Holmes⁴ map. For the remaining 12 patients (Table 2), the locations of lesions predicted by the Holmes map were in closer agreement with their actual locations on MR images than were those predicted by a revised map.⁶

A plot of the percentage of the surface area of the infarcted or noninfarcted occipital cortex vs eccentricity from fixation is shown in Figure 6. The best-fit linear regression is represented by the solid line; 37% of the total surface area of the medial occipital cortex subserves the central 15° of vision.

Our correlations of MRI findings with visual field defects in patients with occipital lobe lesions confirmed gross estimates of the retinotopic organization of the occipital cortex. However, the visual field deficits in our patients did not correspond exactly to the Holmes map,⁴ and they did not correlate with a revised map,⁶ which was based on findings in 3 other reported patients. In addition, we found that 37% of the total surface area of the medial occipital cortex subserves the central 15° of vision.

From the visual field and imaging findings in our patients, a refined retinotopic map is presented (Figure 7). Figure 7 illustrates that the human occipital cortex is situated along the superior and inferior lips of the calcarine fissure. The foveal representation is located posteriorly at the occipital pole, whereas the peripheral visual field is represented anteriorly at the juncture of the parieto-occipital and calcarine fissures. Because the extent of striate cortex (V1) is not determined by imaging, the retinotopic correlation is with MRI of the medial occipital cortex, not with the retinotopic organization of area V1.

Several limitations confound conventional MRI in correlating lesion location with visual field defects. These limitations also apply to CT imaging of occipital lesions.^8-11 First, because visual processing in the striate cortex is organized in ocular dominance columns that have a cross-sectional width of 0.4 mm,¹⁶ 5-mm cuts that are widely used in major MRI centers may not provide sufficient resolution to determine the exact extent and location of the lesion identified by perimetry. However, because the rostral-caudal extent of lesions in our 14 patients differed more than 5 mm from a revised map,⁶ which was based on MR images of 3 patients whose slice thickness was not specified,⁶ limitation of slice thickness does not explain the disparate result. Second, the determination of the extent of the lesion is complicated by areas of edema surrounding the lesion, which may be difficult to differentiate from the actual area of infarction. However, all our patients were imaged many months or years (mean duration of lesions, 14 months) after the onset of visual field loss, when edema should have completely resolved. Third, tissue damage may disrupt function but be undetected by imaging. For example, patients 3 through 7 and 10 had homonymous field defects without sparing of the temporal crescent, indicating that the most anterior portion of the occipital pole should be involved. However, on MRI, no signal change was detected in the anterior medial occipital cortex, suggesting that tissue damage may extend beyond visible MRI change; alternatively, the monocular temporal crescent of field (representing the nasal retina of the contralateral eye) may not extend to the extreme anterior part of the occipital cortex, although this seems unlikely. Fourth, areas of signal change in MRI do not specify destruction of all neural elements, which may be spared in an area of visible damage. This pertains particularly to neoplasms and hemorrhages, which we therefore excluded from our study. Fifth, because we used a 2-dimensional image to estimate the size of a 3-dimensional lesion, the accuracy of measurements depends on the orientation of the cuts. However, there is no ideal head orientation for studying the calcarine fissure because the incline of the calcarine fissure relative to other brain and skull reference coordinates is variably oriented among individuals,¹⁷ and it also varies with minor changes in head position. Finally, and most important, the exact boundaries of the striate cortex cannot be determined by MR images, and there is a natural and substantial variation across individuals in the exact dimension and location of the striate cortex.¹⁴ Moreover, no compensation for cortical infolding or curvature of the calcarine fissure can be made using MR images, which allows only estimation of the extent of the lesion.

The discrepancy between the results of the present study and those leading to a revised map⁶ may be due in part to the small number of patients on whom a revised map was based (3 patients). In addition, whereas we excluded lesions that may have functional neurons remaining within the neuroimaged boundary, a revised map⁶ included 2 patients with such lesions (1 patient had a tuberculoma and another had an arteriovenous malformation), which may lead to an overestimation of the actual area of damage. Although that revised map⁶ was later supported by McFadzean and coworkers¹⁸ using either CT or MRI in a larger series of patients, they also included 11 patients with such lesions (neoplasms, arteriovenous malformations, cerebromalacia, and hematomas). Moreover, because edema and an ischemic penumbra frequently surround the area of infarction, we only included MR images that were performed at least 3 months after patients' initial presentation, when they should have resolved. However, because previous MRI studies^6,18 did not specify the duration of lesions in their patients, bordering ischemia and edema may partly explain the disparate results.

Based on endogenous changes in magnetic susceptibility caused by localized activity-dependent variations in blood flow and oxygenation, functional MRI (fMRI) has been used to map human cortical activity during stimulation of selected areas of the visual field in normal subjects.^19-21 Sereno et al²¹ and Engel et al,²² using fMRI, found that there is emphasis on the center of gaze in the human primary visual cortex. However, their results are not consistent with our data and those from previous studies.^7-11 Potential sources of spatial spread, such as that caused by lateral neural connections within the cortex and experimental artifacts (eg, slight head movement and brain pulsatility),²² may limit the spatial resolution of fMRI to accurately determine cortical magnification. Moreover, these studies measured a traveling wave of activity in the cortex of normal subjects by slowly changing the position of the stimuli pattern. Although these subjects were instructed to fixate at the center of the stimuli display, no attempts were made to monitor eye position during testing. Even the smallest eye movement can affect the measurement, especially at the exact center of the fovea. In addition, fMRI detects changes in signal from venous blood, not the activity of neural tissue. Some investigators²³ suggested that much of the signal change seen in fMRI might be seen within brain areas having little or no neural tissue, being instead images of the venous vasculature.

Correlation of visual field defects with neuropathologic change at autopsy may further clarify the topographic representation of the visual field in the human striate cortex.^24,25 After flattening the complete striate cortex in 4 normal adult patients with 1 eye and reconstructing the ocular dominance columns mosaic, Horton and Hocking²⁶ identified the blind spot representation, which lacks ocular dominance columns and extends from 12° to 18° along the horizontal meridian. They reported that 42% to 62% (mean±SD, 52%±2%) of the human striate cortex corresponds to the central 12° of vision. Horton and Hoyt⁶ presumed that human striate cortex and that of macaque monkeys had about the same cortical magnification of the central visual field, being 70% for the central 15°.^12,13 However, Horton and Hocking²⁶ reported that the cortical magnification of the central 12° of vision was 60% in monkeys and 42% to 62% (mean, 52%) in humans, an area less than that estimated in a revised map⁶ and considerably smaller than that estimated by fMRI.²¹ Flattening procedures²⁶ are subject to distortions that could affect retinotopic data. Three-dimensional MRI reconstruction of slices in different orientations may better correlate anatomy with visual field loss. Although the boundaries of striate cortex are not defined by neuroimaging methods, imaging correlates of lesions with visual field defects provide in vivo information about the retinotopic representation of vision in the occipital lobe.

The primary aim of this study was to establish a direct link between the anatomy of the medial occipital cortex along the calcarine fissure and the visual field locus and the extent of functional deficit. However, there is a limited causal relationship between the individual vagaries of the calcarine anatomic features and the retinotopic characteristics of a visual field defect. A more accurate and direct relationship is more likely to be between the location of cortical neurons receiving input from restricted receptive field positions in the retina and the location of a lesion with respect to those neurons. The arrangement of cortical neurons with respect to the gross anatomy of the calcarine fissure can be quite variable. Fundamentally, then, anatomically based retinotopic maps proposed by Holmes,⁴ others,^1,5,6 and us are limited in their accuracy. However, the location of lesions within the medial occipital cortex on imaging studies has substantial value in determining whether the pattern of a patient's visual field defect is adequately accounted for by imaging. Establishing the validity of the existing retinotopic maps based on lesion data therefore becomes important to account for the symptoms of occipital cortex lesions.

Our MRI correlations confirmed estimates of the retinotopic organization of the medial occipital cortex and corresponded better with the Holmes⁴ map than with a revised map⁶ based on conventional MRI correlations of lesions in 3 other reported patients. Thirty-seven percent of the total surface area of the mesial surface of the human occipital cortex subserves the central 15° of vision. Although precise knowledge of the retinotopic representation of the human striate cortex (V1) is limited by imaging techniques, the resolution of MRI testifies to its considerable value in diagnosing and localizing occipital lobe lesions.

Accepted for publication September 22, 1998.

This study was supported by grant MT 5404 from the Medical Research Council of Canada.

We thank David Mikulis, MD, for his expert advice.

Reprints: James A. Sharpe, MD, Division of Neurology, The Toronto Hospital, EC 5-042, 399 Bathurst St, Toronto, Ontario, Canada M5T 2S8.