Free
Research Article  |   December 2002
The fine structure of multifocal ERG topographies
Author Affiliations
Journal of Vision December 2002, Vol.2, 5. doi:10.1167/2.8.5
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Charlotte M. Poloschek, Erich E. Sutter; The fine structure of multifocal ERG topographies. Journal of Vision 2002;2(8):5. doi: 10.1167/2.8.5.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

The multifocal electroretinogram (mfERG) allows for functional field mapping by concurrently deriving responses from a large number of retinal locations. The stimulus resolution most commonly used consists of 103 hexagonal elements. Here, we stimulated with an array of 509 elements. To determine the extent to which the multifocal ERG shows anatomical and physiological details, such as shadows cast by the retinal vasculature, we obtained mfERGs from two subjects using two different stimulus luminance levels and three light spectra. Good correspondence of some depressions with major blood vessels suggests relative angioscotomata. However, some reproducible local depressions cannot be attributed to blood vessel shadows cast on the retina, but more likely reflect local inhomogeneities in the physiological response characteristics.

Introduction
The conventional full-field electroretinogram became a valuable tool to objectively assess retinal function (Riggs, 1986; Plant, 1995). However, since it reflects a summed response to which the entire retina contributes, it does not provide insights on localized retinal function. Topographic information on the retinal response can be derived from the multifocal electroretinogram (mfERG) which concurrently stimulates a large number of retinal locations (Sutter & Tran, 1992; Sutter 1991). Numerous studies have reported the effects of various retinal diseases on the local responsiveness of the retina (i.e., Bearse & Sutter, 1996; Hood, Holopigian, Greenstein, Seiple, Sutter, & Carr, 1998; Palmowski, Sutter, Bearse, & Fung, 1997) confirming the ability of the mfERG to detect and map small dysfunctional regions. 
To properly interpret mfERGs in the clinic, it is important to know the response distribution in the normal retina as well as its intersubject variability. The protocol most commonly used for clinical recordings employs 103 test areas within the central 50°. At photopic luminance levels, the topography of the first order multifocal response derived from a normal subject resembles a smooth surface with a prominent central peak (Hood, Seiple, Holopigian, & Greenstein, 1997; Sutter & Tran, 1992; Verdon & Haegerstrom-Portnoy, 1999). The only anomaly appears to be a depression in the area of the blind spot. 
Two different groups (Meigen & Friedrich, 2000; Parks, Keating, Evans, Williamson, Jay, & Elliott, 1997) have studied intersubject variability and reproducibility of the mfERG. When data are collected under identical conditions, the factors affecting reproducibility are (a) the ability of the subject to fixate and (b) differences in noise contamination due to blinks or eye movements. Some of the observed variations, however, are clearly a consequence of inhomogeneities in retinal anatomy and physiology. At higher resolutions, the topographies acquire subtle, reproducible structures. The highest resolution commonly used places 241 hexagonal patches within the central 50°. Previous studies suggest that at this resolution recordings may reveal even more intricate spatial structure of normal and pathological response topographies (Heinemann-Vernaleken, Palmowski, Allgayer, & Ruprecht, 2001; Sutter, Tran, 1992). However, at recording times of 8 to 16 minutes, these records tend to be somewhat noisy. Since repeat records are rarely collected, it is generally not known how much of the fine structure is due to noise contamination and how much reflects inhomogeneities in local response properties. 
Some areas of local depressions in response density might be attributable to inner retinal blood supply. The largest such features are the upper and lower branches of the central retinal artery and vein that provide the largest retinal coverage and light absorption during stimulation. The resulting relative angioscotomata have been mapped by fundus-oriented perimetry (Schiefer, Benda, Dietrich, Selig, Hofmann, & Schiller, 1999) and, in principle, should also be seen in mfERG topographies. 
In the present study, we increased the resolution to 509 hexagonal stimulus patches. Achieving such high resolution requires recording times of approximately one hour, good signal-to-noise ratios and excellent fixation stability of the observer. While such recordings are not feasible in the clinic, this high resolution study contributes to a better understanding of normal local variabilities, their magnitude and origin. 
Methods
Stimulus Geometry
The stimulus consisted of 509 hexagonal elements and was displayed on a 21″ high luminance monochrome monitor with a P45 phosphor (Philips FIMI). The stimulus elements were scaled with eccentricity to generate responses of similar signal-to-noise ratio (SNR) across the stimulated retinal areas. The subjects’ viewing distance measured 40 cm resulting in a stimulus diameter of approximately 44° visual angle. The diameter of the stimulus patches increased from 0.8° at the center to 2.8° at the periphery of the array. Figure 1 shows the stimulus array superimposed on a fundus photograph taken from Subject 1 and relates various fundus structures such as the optic disc and retinal vasculature to individual stimulus elements. 
Figure 1
 
The stimulus matrix of 509 hexagonal elements overlaid on the fundus photograph of Subject 1. The diameter of the stimulus elements increased from 0.8° in the center to 2.8° in the periphery. The array is scaled with eccentricity to generate responses of similar signal-to-noise ratio across the stimulated retinal area. Some of the stimulus elements project on one or more blood vessels. The rings indicate increments of 5° eccentricity.
Figure 1
 
The stimulus matrix of 509 hexagonal elements overlaid on the fundus photograph of Subject 1. The diameter of the stimulus elements increased from 0.8° in the center to 2.8° in the periphery. The array is scaled with eccentricity to generate responses of similar signal-to-noise ratio across the stimulated retinal area. Some of the stimulus elements project on one or more blood vessels. The rings indicate increments of 5° eccentricity.
Estimation of Record Length
We estimated the required record length using the following considerations. Even at the higher spatial resolution used in this study, lateral interactions between the stimulated retinal patches contribute relatively little to the focal responses (Sutter, 2001) and, in subjects with clear optical media, contrast attenuation at the stimulus borders is small. We thus expect that the focal response amplitudes are approximately proportional to the area of the stimulus patches. With 509 patches covering the same visual field as the 103 patches used for a standard recording, the SNR should thus be about 1/5. Since the SNR increases with the square root of the record length, the recording time needed to achieve similar record quality would have to be increased by a factor of 25. We know that in a good experimental subject, the recording time required to obtain adequate SNRs at the resolution of 103 is about 2 minutes. Based on these considerations we estimated the minimum recording times required for our study at approximately one hour. 
Stimulus Protocol
In order to optimize the quality of the high resolution response topographies, we chose a special protocol known to increase the SNR of the first order kernel at the expense of higher order kernels. Each step in the binary stimulation sequence consisted of two video frames whereby every m-sequence controlled video frame is followed by its complement. If a stimulus patch was flashed in the first frame of the pair, it remained dark in the second, and conversely. The stimulus was thus updated on every video frame, that is, at a rate of 75/s. In the computation of the focal first order kernels, the subsets of all the first frames and all the second frames contribute responses of opposite polarities with a relative lag of one frame interval. Since the frame interval is almost precisely equal to the interval between the first dominant positive peak and the following trough, the two contributions add “in phase” (Figure 2). 
Figure 2
 
Derivation of the first order kernel in the case of our special stimulus protocol. The computation is equivalent to a weighted average of the response waveforms following all stimulus frames in the record. All focal stimulus events of the type shown in the top line are added while all those shown in the bottom line are subtracted. Stimulus elements are either flashed (white), not flashed (black) or have a 50%-chance of being either flashed or not flashed (shaded elements). The two consecutive counter modulated frames contribute response waveforms of opposite polarity shifted relative to each other by one frame interval (13.3 ms). As the shift approximately equals the peak-to-trough time of the two response waveforms, they enhance each other.
Figure 2
 
Derivation of the first order kernel in the case of our special stimulus protocol. The computation is equivalent to a weighted average of the response waveforms following all stimulus frames in the record. All focal stimulus events of the type shown in the top line are added while all those shown in the bottom line are subtracted. Stimulus elements are either flashed (white), not flashed (black) or have a 50%-chance of being either flashed or not flashed (shaded elements). The two consecutive counter modulated frames contribute response waveforms of opposite polarity shifted relative to each other by one frame interval (13.3 ms). As the shift approximately equals the peak-to-trough time of the two response waveforms, they enhance each other.
To test the performance of this stimulation mode, we compared the results with records obtained from the same subject and in the same recording time with the standard protocol under otherwise identical conditions: For each stimulus patch we calculated the ratio of the first order amplitude densities derived with the two methods of stimulation. This ratio averaged over all 509 stimulus patches was 1.88 in favor of the new stimulation protocol. Since each stimulus interval now consisted of two frames, the number of stimulus events presented in the same recording time was reduced by half. This resulted in an increase in the noise level by a factor of Image not available. Thus, we effectively achieved a net improvement in SNR by a factor of 1.33. It is important to keep in mind that the first order kernel waveforms presented in this paper consist basically of the sum of two responses relatively inverted and shifted by 1.33 ms. They should not be directly compared to those obtained with the standard multifocal stimulation protocol. 
Effect of Stimulus Chromaticity
Part of our study aimed at testing whether there is a correspondence between local depressions in the retinal response topography and shadows cast by retinal blood vessels. The hemoglobin molecule shows two absorption maxima for short and middle-long wavelengths (Figure 3). Both its oxygenated (HbO2) and deoxygenated (Hb) states show higher absorption coefficients for short wavelengths. Based on the absorption characteristics we initially assumed that the outline of angioscotomata should be more visible for recordings in blue rather than green or white light. To test this we modified the spectrum of the stimulating light with additive blue and green dichroic color filters (L52-530 and L52-533 respectively, Edmund Industrial Optics, Barrington, NJ) (Figure 4). 
Figure 3
 
The molar extinction coefficient [cm-1/M] plotted as a function of wavelength. The hemoglobin molecule shows two absorption maxima for its oxygenated (HbO2) and deoxygenated (Hb) state. HbO2 has maximum light absorption at 412 and 576 nm, and Hb has maximal absorption at 432 and 556 nm, both states having higher absorption coefficients for blue light.
Figure 3
 
The molar extinction coefficient [cm-1/M] plotted as a function of wavelength. The hemoglobin molecule shows two absorption maxima for its oxygenated (HbO2) and deoxygenated (Hb) state. HbO2 has maximum light absorption at 412 and 576 nm, and Hb has maximal absorption at 432 and 556 nm, both states having higher absorption coefficients for blue light.
Figure 4
 
Typical transmittance curves for the additive green and blue dichroic filter (green and blue lines, respectively) shown in transmission-% as a function of wavelength. The black line indicates the emission spectrum of the phosphor P45 (emission-% as a function of wavelength).
Figure 4
 
Typical transmittance curves for the additive green and blue dichroic filter (green and blue lines, respectively) shown in transmission-% as a function of wavelength. The black line indicates the emission spectrum of the phosphor P45 (emission-% as a function of wavelength).
Luminances of green, white and blue stimuli were adjusted to the desired values with the VERIS software after calibration of software with the Vλsensitive VERIS autocalibrator. For all three stimulus chromaticities the resulting luminance levels were in good agreement with those measured with a Pritchard photometer. 
We estimated factors for the reduction of retinal stimulation by hemoglobin. The spectrum of the stimulating light is defined as the product of the phosphor P45 emission spectrum and the transmission spectra of the filters. This stimulus spectrum multiplied by the photopic luminous efficiency (Vλ) provides a good estimate for the resulting retinal stimulation. 
To estimate the effect of the blood vessels, retinal stimulation was corrected for the molar extinction coefficient (e) of hemoglobin (e (HbO2) and e (Hb) were averaged). We found that hemoglobin reduced photopic retinal stimulation in white light by a factor of 12, in green light by a factor of 11 and in blue light by a factor of 9. It follows that light absorption should have the strongest effect in white light. Nevertheless, our results showed that despite the absorption characteristics angioscotomata were most apparent in blue light. This was confirmed in maps of the weighted difference between the topographies derived with blue and white stimuli (see Figure 9). 
Figure 9
 
Response density distribution for Subject 1 (left) and 2 (right) after subtracting weighted responses to the white light from the blue light stimulus spectrum. The depressions in the areas of the upper and lower arcades indicate that the blood vessel shadows are more effective in blue light. Subject 1 shows a depression in the central 5°, Subject 2 displays a strong nasal-temporal asymmetry.
Figure 9
 
Response density distribution for Subject 1 (left) and 2 (right) after subtracting weighted responses to the white light from the blue light stimulus spectrum. The depressions in the areas of the upper and lower arcades indicate that the blood vessel shadows are more effective in blue light. Subject 1 shows a depression in the central 5°, Subject 2 displays a strong nasal-temporal asymmetry.
To obtain good SNRs, recordings in green and white light were set to 8 cd*s/m2. The maximum luminance level achievable with the blue filter was 1 cd*s/m2. To compare responses to the different stimulus chromaticities at the same luminance level, recordings in green and white light were repeated at 1 cd*s/m2. Contrast of luminance modulation was always ∼100%. 
Recording Procedure
The eye was kept light-adapted at room illumination prior to recording. The cornea was anesthetized (proparacaine hydrochloride 0.5%) and the pupil was dilated (tropicamide 1%). In both subjects the responses were derived monocularly from the cornea of the left eye by means of a bipolar Burian-Allen contact-lens electrode. The signal was amplified (gain 50000), band-pass filtered (10–300 Hz) and sampled 16 times per video frame, that is, every 0.83 ms. VERIS 4.3 software (Electro-Diagnostic Imaging, Inc., San Mateo, CA) was used for stimulation, data collection and analysis. 
Monitoring Fixation Stability
Both subjects were experienced scientific observers. However, when the diameter of the stimulus patches becomes as small as 0.8°, one approaches the limits in the ability of a human observer to sustain accurate fixation throughout the recording. We therefore continuously checked the raw signal for eye movement artifacts. Moreover, the eye was monitored with an infrared camera. Even small changes in the direction of gaze could be detected by the relative motion between the reflection of an illuminating IR diode on the contact lens electrode and the subject’s pupil. We verified that, with this method, eye movements as small as 1° visual angle could be detected. The subjects were instructed to ask the operator to restart the recording of a segment as soon as they felt they had deviated from the fixation target. We are confident that the quality of fixation was adequate for the chosen resolution. Higher resolution recording might require tracking the fundus with the stimulus. 
Response Amplitude Estimation
We used the scalar product method to estimate response amplitudes (Sutter & Tran, 1992). In this calculation, a template for the response waveform is normalized to a RMS amplitude of 1. The amplitude of a local response waveform is estimated by multiplying it, data point by corresponding data point, with the matching template and summing all the products. The templates are derived as the average of traces known to contain approximately the same response waveform. As the patches of high resolution stimulus arrays are very small, individual waveforms are noisy and a substantial amount of averaging is needed to obtain good quality templates. The derivation of normalized template waveforms used for the analysis in each instant is described at appropriate places in the Results section. 
Data Presentation
Temporal responses to individual hexagons are shown as 509 focal waveforms. Amplitude densities calculated from the temporal responses, however, are presented in a refined resolution derived through an interpolation procedure (Sutter & Tran, 1992). To extract the fine structure in the response topographies, special methods of filtering were applied. These techniques are described in related paragraphs of the Results section. 
Results
Retinal Stimulation and Amplitude Size
The response densities and waveforms, shown in Figure 5, exemplify our first order high resolution data for stimulation with the white light at a flash intensity of 8 cd*s/m2 and blue light at 1 cd*s/m2. The local response amplitudes were estimated as the scalar product of each waveform with a normalized template (Sutter & Tran, 1992). Since waveforms change predominantly with retinal eccentricity, normalized ring averages were used as templates. To avoid using the central waveform as its own template that would lead to overestimation of the central amplitude, the template chosen for estimation of the central amplitude was the average of the central seven traces. The resulting amplitude estimates were locally averaged over the four records and divided by the area of the stimulus patches to yield the response density distributions shown in Figure 5
Figure 5
 
First order kernel response densities and focal trace arrays derived from the average of four 15 minute records of Subject 1. The responses to white light at 8 cd*s/m2 are shown on the left, those to blue light at 1 cd*s/m2 on the right. Both recordings show a sharp foveal peak and a well defined minimum in response density in the area of the blind spot. With the low intensity blue stimulus, the center/periphery ratio of the response densities is lower than with the high intensity white stimulus. Response densities and focal trace arrays are presented in retinal view.
Figure 5
 
First order kernel response densities and focal trace arrays derived from the average of four 15 minute records of Subject 1. The responses to white light at 8 cd*s/m2 are shown on the left, those to blue light at 1 cd*s/m2 on the right. Both recordings show a sharp foveal peak and a well defined minimum in response density in the area of the blind spot. With the low intensity blue stimulus, the center/periphery ratio of the response densities is lower than with the high intensity white stimulus. Response densities and focal trace arrays are presented in retinal view.
Considering the large difference in stimulus luminance in the two experiments, it is not surprising that the amplitudes of the responses to the white stimulus are larger than those to the blue stimulus. However, considering the factor 8 difference in stimulus intensity, the responses to the blue stimulus are unexpectedly large. In fact, the ratio of the corresponding response amplitudes (white to blue) was only 1.33:1 in Subject 1 and 1.14:1 in Subject 2. Approximately the same ratios were obtained for the responses to the green light and blue stimuli. The ratios were derived from the spatial average of all corresponding local response amplitudes. At equal luminance levels, responses to blue light were higher by a factor of 1.8 for Subject 1 and 1.3 for Subject 2. These data are summarized in Table 1
Table 1
 
Response Ratios Between Responses to the Blue Light Stimulus and White or Green Light Stimulus Respectively
Table 1
 
Response Ratios Between Responses to the Blue Light Stimulus and White or Green Light Stimulus Respectively
Focal flash intensities (cd/m2) Subject 1 Subject 2
white/blue 8 : 1 1.33 1.14
white/blue 1 : 1 0.538 0.77
green/blue 8 : 1 1.28 1.18
green/ blue 1 : 1 0.588 0.704
Both recordings of Figure 5 show a steep drop-off in amplitude with eccentricity and a well defined foveal peak. The topographic distribution of the response density to blue light, however, shows relatively lower responses towards the center. This raises the question whether this difference was due to the difference in stimulus spectrum or flash intensity. This question was addressed by the experiments with a white stimulation at the same flash intensity of 1 cd*s/m2. As the ratio of central to peripheral responses was similar in these two records, we attribute the observed differences in the drop-off with eccentricity to the stimulus luminance. All recordings showed a very well defined blind spot. However, especially in white light, the focal trace arrays indicate some residual responses that were also present in the recording in white light at 1 cd*s/m2
Mapping Spatial Fine Structure in the Response Topography
Distinguishing the fine structure from the gross features of the response topography requires high-pass spatial filtering. Two main filtering methods were used to visualize and estimate the small-scale local variations. First, the scaling of the stimulus elements with eccentricity generated responses of similar amplitude across the stimulated field. Mapping the response amplitudes directly rather than first dividing them by the area of the stimulus patches effectively eliminated the most conspicuous gross features, namely the central peak with its steep drop-off with eccentricity. As the noise level is equal in the extracted focal responses, the noise contamination in the resulting plots is also uniform across the entire field. While this presentation of the data makes it easier to visualize local variability, we must remember that the plots no longer reflect the response density of the retina. 
Second, since the resulting maps still contained larger-scale local variations due, for example, to nasal-temporal and superior-inferior asymmetries in physiological and anatomical properties of the retina, proper estimation of the response fine structure required further high-pass spatial filtering. This was accomplished indirectly with the help of a low-pass spatial filter available in VERIS known as spatial averaging (Sutter & Tran, 1992). When spatial averaging is used, the amplitude of each element is averaged with a certain percentage of its neighbors. This spatial filtering is adjustable by changing the percentage value and by defining the number of times the averaging process is iterated. Computing the difference between the response topography smoothed by heavy low-pass filtering and the corresponding unfiltered response topography, we achieved appropriate high-pass spatial filtering. For the derivation of the smoothed topography, each focal amplitude was averaged with 100% of the amplitudes of its six surrounding neighbors and this procedure was repeated once. To reduce local noise contributions, we also found it necessary to apply a small amount of spatial averaging to the unsmoothed response topography. In this instant, each focal amplitude was averaged with only 30% of the amplitudes of each of its nearest neighbors. Both low-pass filters (1 × 30% and 2 × 100%) accurately preserve the remaining gross features such as nasal/temporal asymmetries. When computing the difference between the two resulting topographies these features cancel out. What remains is the desired high-pass spatially filtered fine structure. 
The local templates for estimation of the high-pass filtered response amplitudes were obtained in two stages. First, we averaged the response waveforms of the four single records to obtain good response estimates at each stimulus location. In the second stage, the strong spatial averaging used for the derivation of the smoothed response arrays produced substantial additional improvement in the SNR of the local templates. Response amplitudes were then computed as the scalar products between the local waveforms and the corresponding templates. 
Figure 6 shows the spatially filtered fine structure (first order kernel) of the blue stimulus recording from subjects 1 and 2. 
Figure 6
 
Responses of Subject 1 (left) and 2 (right) to the blue light stimulus spectrum after high-pass filtering are shown in retinal view to match the location of fundus structures. Some local depressions fall in the region of blood vessel shadows, especially along the upper and lower arcades. Other regions of depressed responses do not fall within the location of retinal vasculature.
Figure 6
 
Responses of Subject 1 (left) and 2 (right) to the blue light stimulus spectrum after high-pass filtering are shown in retinal view to match the location of fundus structures. Some local depressions fall in the region of blood vessel shadows, especially along the upper and lower arcades. Other regions of depressed responses do not fall within the location of retinal vasculature.
Applying the same filtering to a kernel slice that contains no signal, we determined the noise level in the fine structure plots of Figure 6. Averaging all 509 traces of a fifth order kernel slice, we verified that an array of traces did not contain any detectable signal and then chose this kernel slice for the noise estimation. In order to achieve identical filtering for noise and data, the same templates used for estimation of the first order kernel amplitudes were also used for estimation of the noise amplitudes. The noise topographies corresponding to the data sets of Figure 6 are shown in Figure 7. Please note that this figure depicts a representative noise distribution. It is not the precise noise topography contained in the plot of Figure 6 which, if accessible, could be subtracted. 
Figure 7
 
Representative noise distribution derived from the record with blue light stimulation for Subject 1 (left) and Subject 2 (right). The plots show the topography of a fifth order kernel slice known to contain no measurable response power. These noise topographies have undergone precisely the same type of processing and high-pass filtering used for the derivation of the response topographies of the first order kernel shown in Figure 6. As in Figure 6 responses are not divided by area but plotted directly and displayed in retinal view. These noise topographies are thus representative of the noise contamination in the plots of Figure 6. In both subjects, the noise is randomly distributed across the 509 stimulus patches.
Figure 7
 
Representative noise distribution derived from the record with blue light stimulation for Subject 1 (left) and Subject 2 (right). The plots show the topography of a fifth order kernel slice known to contain no measurable response power. These noise topographies have undergone precisely the same type of processing and high-pass filtering used for the derivation of the response topographies of the first order kernel shown in Figure 6. As in Figure 6 responses are not divided by area but plotted directly and displayed in retinal view. These noise topographies are thus representative of the noise contamination in the plots of Figure 6. In both subjects, the noise is randomly distributed across the 509 stimulus patches.
From a comparison of Figures 6 and 7, it is immediately obvious that the peaks and valleys in the topography of the first order kernel cannot all be attributed to noise contamination. The fact that the largest depressions are found along the paths of major blood vessels suggests that they represent relative angioscotomata. The superimposed outlines of the retinal vasculature were derived from fundus photographs and adjusted to the size, aspect ratio and tilt angle of the three dimensional response topographies. Arteries and veins of the upper and lower arcades appear to fit well into the largest valleys. Groups of smaller vessels also appear to reduce retinal responses. An example of this is the cilioretinal artery that is accompanied by another artery in Subject 1. 
In order to obtain a quantitative estimation of the size of the peaks and valleys in the spatially filtered response topographies we measured them in units of standard deviations of the noise (SDN). The 509 high-pass filtered noise amplitudes from which the noise topographies of Figure 7 were derived, provide a suitable noise sample for the derivation of SDN. We defined amplitude values that deviated from the mean of the noise by more than 1.5 SDN as significant response inhomogeneities. 
The hexagons of Figure 8 marked black, green or pink are stimulus patches whose high-pass filtered response amplitudes were either larger or smaller than the spatial mean of the noise by at least 1.5 SDN. Depressed hexagons are marked black if they are traversed by blood vessels, and green otherwise. Hexagons whose response amplitudes exceeded the mean of the noise by 1.5 SDN or more are marked pink. 
Figure 8
 
Local amplitude deviations above (pink hexagons) and below (green hexagons) 1.5 SDN from the average noise level. Black hexagons mark local depressions below 1.5 SDN for elements that project onto blood vessels. In both Subject 1 (left) and 2 (right), there is good correspondence between the depressed areas and the upper and lower arcade, as well as with some smaller vessels. However, elements that are not affected by blood vessel shadows also show significant depressions evenly spread in subject 1 and concentrated on the temporal retina in Subject 2. Significant local amplitude enhancements (pink hexagons) are sparsely distributed.
Figure 8
 
Local amplitude deviations above (pink hexagons) and below (green hexagons) 1.5 SDN from the average noise level. Black hexagons mark local depressions below 1.5 SDN for elements that project onto blood vessels. In both Subject 1 (left) and 2 (right), there is good correspondence between the depressed areas and the upper and lower arcade, as well as with some smaller vessels. However, elements that are not affected by blood vessel shadows also show significant depressions evenly spread in subject 1 and concentrated on the temporal retina in Subject 2. Significant local amplitude enhancements (pink hexagons) are sparsely distributed.
Of all stimulus elements that projected onto blood vessels, 63% in Subject 1 and 62% in Subject 2 had significantly reduced responses. Some significant reductions are also found in the area of some smaller vessels towards the center of the stimulus array. It is clear, however, that not all of the local depressions are attributable to light absorption by blood vessels. Of all stimulus patches with depressed responses, 25% in subject 1 and 29% in subject 2 were not covered by blood vessels. We attribute these response depressions as well to areas of significantly enhanced responses caused by physiological inhomogeneities. 
Dependence of the Response Fine Structure on the Stimulus Spectrum
A comparison of the high resolution response topographies obtained with white, green and blue light suggested that the local depressions along the major blood vessels were best visible in blue and least visible in white light. The local amplitude difference between recordings in blue and white light should therefore enhance the visibility of relative angioscotomata. Figure 9 shows the difference between responses to the blue and white light spectrum, both obtained with a stimulus flash intensity of 1 cd*s/m2. The local templates for response amplitude estimation were derived in the same way as described in the Results section Retinal stimulation and amplitude size. Since at equal luminance levels, the responses to blue light were higher than those to white light (see Table 1), the two records were appropriately weighted to equalize their mean response amplitudes of Figure 9. The difference plots, shown in Figure 9, indeed reveal depressions in the regions of the major vessels. The central depression, which is particularly prominent in Subject 1, may be attributed to selective attenuation of the blue light stimulus by the macular pigment. 
Dependence of the Global Response Topography on Stimulus Intensity and Chromaticity
Compared to the high luminance white recording (8 cd*s/m2), the center to periphery response density ratio was smaller at the lower stimulus intensity of 1 cd*s/m2 with the blue as well as the white stimulus. On the other hand, the central depression in the difference plot of Figure 9 between the low intensity blue and white recordings indicates that the center to periphery ratio was somewhat lower with the blue stimulus. In both low intensity recordings the region of the blind spot shows a marked depression, but not a complete null. The residual response is attributed to scattered light. The fact that the topography of the weighted difference (blue minus white) also shows a depression at the optic disc suggests that stray light contributes less to the responses in the blue light than in white light. 
Discussion
The main goal of this study was to assess the response variability across the human retina. We first identified local response variations that were not readily attributable to noise contamination. This was accomplished by means of two different methods of spatial filtering. The pattern of the most salient local depressions in the resulting response topography suggested that they may be due to light absorption in the retinal vasculature. A quantitative evaluation of the correspondence between local depressions and the location of the retinal vasculature suggested that some, but not all of the response fine structure can be attributed to relative angioscotomata. Other local peaks and valleys must be due to local differences in physiological and anatomical properties of the retina. 
Visibility of Retinal Vasculature
According to our calculations, light absorption by hemoglobin reduces retinal stimulation in white light by a factor of 12, but for the blue light spectrum only by a factor of 9. It thus follows that the effect of light absorption by blood vessels should be most pronounced with white light stimulation. However, we found that the blue light stimulus provided best evidence for angioscotomata. One possible explanation for the discrepancy between the calculated reduction factors and our results might be the effect of stray light. The size of the scattered light response can be estimated from the responses elicited by stimulus patches falling within the optic disc (Figure 1). Both, the white stimulus and to a lesser extent the blue stimulus, show some residual responses in these areas. We expect that the major source of light scatter in the eyes of the two young subjects is back scatter from the red fundus. As the amount of long wavelength light is less with the blue stimulus, we expect that the total amount of intra-ocular scatter is also less than with the white light. Back scatter of the long wavelength component of the white stimulus might not only be the reason for the relatively higher residual responses at the optic disc but also for contrast degradation of angioscotomata. 
Local response reductions were most salient along the course of the upper and lower arcades. In these areas, the responses were depressed relative to the surrounding areas by more than 1.5 SDN. However, the correspondence between smaller vessels and response depressions was predictably poor. Even with accumulated recording times of one hour, such small depressions remain below the noise level. While the low-pass spatial filtering applied to the unsmoothed data (averaging each trace with 30% of its neighbors) helped to reduce the noise level, it also may have helped to obscure the smallest local depressions. 
To what extent does the ratio of stimulus element area covered by blood vessels to stimulus element size allow for an outline of angioscotomata? At an eccentricity of 15°, a stimulus patch measured approximately 2.3° in diameter. Assuming a mean diameter of 130 μm for major vessels, the width of an artery or vein corresponds to approximately 0.45° (Drasdo & Fowler, 1974). As Figure 1 shows, one single hexagon may be traversed by two vessels of the upper and lower arcades, therefore doubling the width of the overall vessel area to 0.9º. We thus estimate that at most 40% of such a stimulated retinal patch was affected by the light absorption. 
Large Amplitude Sizes of Responses to the Blue Light Spectrum
At equal photopic luminance levels, retinal stimulation with blue light produced substantially larger responses than those to white light (see Table 1). We can safely exclude rod responses as a possible cause for this difference. With a scotopic luminance of 22 cd*s/m2, the blue light stimulus exceeds the rod saturation level by far (Wyszecki & Stiles, 1967). Neither do we find evidence for an isolated S-cone response evoked by the blue stimulus light. The blue light responses showed no signs of the slow, long latency waveforms that are characteristic for S-cone mediated responses. Furthermore, S-cone responses in the multifocal ERG as well as in the Ganzfeld-ERG are known to be very small (Baseler, Schneck, & Sutter, 1996; Gouras & MacKay, 1990). The reason for the high amplitudes in blue light must be related to the relative contribution from M- and L-cones inputs and nonlinear interactions between the L-, M- and S-cone inputs. 
Conclusions
Judging from mfERG records obtained with lower resolution, one might be tempted to consider the retinal response topography as a radially symmetric surface, smooth, with the exception of the blind spot and the central foveal peak. Our results reveal a fine structure in local responsiveness that is neither attributable to noise contamination nor related to pathological changes in the retina. We conclude that some of the small-scale local inhomogeneities in the response topography have an anatomical/optical, and others a physiological origin. They are due in part to shadows cast by the retinal vasculature and to local differences in response amplitude and dynamics. It is not known at this point whether areas of depressed local amplitude might indicate regions that are more susceptible to pathological changes within the retina or the underlying structures. 
While such high resolution recordings are generally not feasible in the clinic, the study provides us with some appreciation of the size of response inhomogeneities that can be expected in a normal human retina. 
Acknowledgements
We thank Min Wang, M.D., for his comments throughout the study and help in data recording. We also wish to thank the Pacific Eye Foundation at the California Pacific Medical Center for taking fundus photographs. This work was supported by NIH grant EY 06861. C.M.P. was supported by a Rachel C. Atkinson Fellowship from the Smith-Kettlewell Eye Research Institute. Commercial relationships: P (E.E.S.). 
References
Baseler, H. A. Schneck, M. E. Sutter, E. E. (1996). Contribution of different receptor populations to multifocal ERGs and VEPs [Abstract]. Investigative Ophthalmology and Visual Science, 37, 1061.
Bearse, M. A.Jr. Sutter, E. E. (1996). Imaging localized retinal dysfunction with the multifocal electroretinogram. Journal of the Optical Society of America A, 13, 634–640. [PubMed] [CrossRef]
Drasdo, N. Fowler, C. W. (1974). Non-linear projection of the retinal image in a wide-angle schematic eye. British Journal of Ophthalmology, 58, 709–714.[PubMed] [CrossRef] [PubMed]
Fry, G. A. (1954) A re-evaluation of the scatter theory of glare. Illuminating Engineering, 49, 98–102.
Gouras, P. MacKay, C. J. (1990). Electroretinographic responses of the short wavelength-sensitive cones. Investigative Ophthalmology and Visual Science, 7, 1203–1209. [PubMed]
Heinemann-Vernaleken, B. Palmowski, A. M. Allgayer, R. Ruprecht, K.W. (2001). Comparison of different high-resolution multifocal electroretinogram recordings in patients with age-related maculopathy. Graefe’s Archives for Clinical and Experimental Ophthalmology, 239, 556–561. [PubMed] [CrossRef]
Hood, D. C. Seiple, W. Holopigian, K. Greenstein, V. (1997). A comparison of the components of the multi-focal and full-field ERGs. Visual Neuroscience, 14, 533–544.[PubMed] [CrossRef] [PubMed]
Hood, D. C. Holopigian, K. Greenstein, V. Seiple, W. Li, J. Sutter, E. E. Carr, R. E. (1998). Assessment of local retinal function in patients with retinitis pigmentosa using the multi-focal ERG technique. Vision Research, 38, 163–179.[PubMed] [CrossRef] [PubMed]
Hood, D. C. Wladis, E. J. Shady, S. Holopigian, K. Li, J. Seiple, W. (1998). Multifocal rod electroretinograms. Investigative Ophthalmology and Visual Science, 39, 1152–1162.[PubMed] [PubMed]
Meigen, T. Friedrich, A. (2000). On the reproducibility of multifocal ERG recordings [Abstract]. Ophthalmologe, 97, 38. Retrieved September 17, 2002, from http://www.dog.org/2000/e-abstract_2000/151.html [CrossRef] [PubMed]
Palmowski, A. M. Sutter, E. E. Bearse, M. A.Jr. Fung, W. (1997). Mapping of retinal function in diabetic retinopathy using the multifocal electroretinogram. Investigative Ophthalmology and Visual Science, 38, 2586–2596.[PubMed] [PubMed]
Parks, S. Keating, D. Evans, A. L. Williamson, T. H. Jay, J. L. Elliott, A. T. (1997). Comparison of repeatability of the multifocal electroretinogram and Humphrey perimeter. Documenta Ophthalmologica, 92, 281–289.[PubMed] [CrossRef]
Plant, G. T. (1995). Recent advances in the electrophysiology of visual disorders. Current Opinion in Ophthalmology, 6, 54–59.[PubMed] [CrossRef] [PubMed]
Riggs, L. A. (1986). Electroretinography. Vision Research, 26, 1443–1459.[PubMed] [CrossRef] [PubMed]
Schiefer, U. Benda, N. Dietrich, T. J. Selig, B. Hofmann, C. Schiller, J. (1999). Angioscotoma detection with fundus-oriented perimetry. A study with dark and bright stimuli of different sizes. Vision Research, 39, 1897–1909.[PubMed] [CrossRef] [PubMed]
Sutter, E. E. (1991). The fast m-transform: A fast computation of cross-correlations with binary m-sequences. SIAM Journal on Computing, 20(4), 686–694. [CrossRef]
Sutter, E. E. Tran, D. (1992). The field topography of ERG components in man—I. The photopic luminance response. Vision Research, 32, 433–446.[PubMed] [CrossRef] [PubMed]
Sutter, E. E. (2001). Imaging visual function with the multifocal m-sequence technique. Vision Research, 41, 1241–1255.[PubMed] [CrossRef] [PubMed]
Verdon, W. A. Haegerstrom-Portnoy, G. (1999). Topography of the multifocal electroretinogram. Documenta Ophthalmologica, 95, 73–90.[PubMed] [CrossRef]
Wyszecki, G. Stiles, W. S. (1967). Color science: Concepts and methods, quantitative data and formulas. New York, NY: John Wiley & Sons.
Figure 1
 
The stimulus matrix of 509 hexagonal elements overlaid on the fundus photograph of Subject 1. The diameter of the stimulus elements increased from 0.8° in the center to 2.8° in the periphery. The array is scaled with eccentricity to generate responses of similar signal-to-noise ratio across the stimulated retinal area. Some of the stimulus elements project on one or more blood vessels. The rings indicate increments of 5° eccentricity.
Figure 1
 
The stimulus matrix of 509 hexagonal elements overlaid on the fundus photograph of Subject 1. The diameter of the stimulus elements increased from 0.8° in the center to 2.8° in the periphery. The array is scaled with eccentricity to generate responses of similar signal-to-noise ratio across the stimulated retinal area. Some of the stimulus elements project on one or more blood vessels. The rings indicate increments of 5° eccentricity.
Figure 2
 
Derivation of the first order kernel in the case of our special stimulus protocol. The computation is equivalent to a weighted average of the response waveforms following all stimulus frames in the record. All focal stimulus events of the type shown in the top line are added while all those shown in the bottom line are subtracted. Stimulus elements are either flashed (white), not flashed (black) or have a 50%-chance of being either flashed or not flashed (shaded elements). The two consecutive counter modulated frames contribute response waveforms of opposite polarity shifted relative to each other by one frame interval (13.3 ms). As the shift approximately equals the peak-to-trough time of the two response waveforms, they enhance each other.
Figure 2
 
Derivation of the first order kernel in the case of our special stimulus protocol. The computation is equivalent to a weighted average of the response waveforms following all stimulus frames in the record. All focal stimulus events of the type shown in the top line are added while all those shown in the bottom line are subtracted. Stimulus elements are either flashed (white), not flashed (black) or have a 50%-chance of being either flashed or not flashed (shaded elements). The two consecutive counter modulated frames contribute response waveforms of opposite polarity shifted relative to each other by one frame interval (13.3 ms). As the shift approximately equals the peak-to-trough time of the two response waveforms, they enhance each other.
Figure 3
 
The molar extinction coefficient [cm-1/M] plotted as a function of wavelength. The hemoglobin molecule shows two absorption maxima for its oxygenated (HbO2) and deoxygenated (Hb) state. HbO2 has maximum light absorption at 412 and 576 nm, and Hb has maximal absorption at 432 and 556 nm, both states having higher absorption coefficients for blue light.
Figure 3
 
The molar extinction coefficient [cm-1/M] plotted as a function of wavelength. The hemoglobin molecule shows two absorption maxima for its oxygenated (HbO2) and deoxygenated (Hb) state. HbO2 has maximum light absorption at 412 and 576 nm, and Hb has maximal absorption at 432 and 556 nm, both states having higher absorption coefficients for blue light.
Figure 4
 
Typical transmittance curves for the additive green and blue dichroic filter (green and blue lines, respectively) shown in transmission-% as a function of wavelength. The black line indicates the emission spectrum of the phosphor P45 (emission-% as a function of wavelength).
Figure 4
 
Typical transmittance curves for the additive green and blue dichroic filter (green and blue lines, respectively) shown in transmission-% as a function of wavelength. The black line indicates the emission spectrum of the phosphor P45 (emission-% as a function of wavelength).
Figure 9
 
Response density distribution for Subject 1 (left) and 2 (right) after subtracting weighted responses to the white light from the blue light stimulus spectrum. The depressions in the areas of the upper and lower arcades indicate that the blood vessel shadows are more effective in blue light. Subject 1 shows a depression in the central 5°, Subject 2 displays a strong nasal-temporal asymmetry.
Figure 9
 
Response density distribution for Subject 1 (left) and 2 (right) after subtracting weighted responses to the white light from the blue light stimulus spectrum. The depressions in the areas of the upper and lower arcades indicate that the blood vessel shadows are more effective in blue light. Subject 1 shows a depression in the central 5°, Subject 2 displays a strong nasal-temporal asymmetry.
Figure 5
 
First order kernel response densities and focal trace arrays derived from the average of four 15 minute records of Subject 1. The responses to white light at 8 cd*s/m2 are shown on the left, those to blue light at 1 cd*s/m2 on the right. Both recordings show a sharp foveal peak and a well defined minimum in response density in the area of the blind spot. With the low intensity blue stimulus, the center/periphery ratio of the response densities is lower than with the high intensity white stimulus. Response densities and focal trace arrays are presented in retinal view.
Figure 5
 
First order kernel response densities and focal trace arrays derived from the average of four 15 minute records of Subject 1. The responses to white light at 8 cd*s/m2 are shown on the left, those to blue light at 1 cd*s/m2 on the right. Both recordings show a sharp foveal peak and a well defined minimum in response density in the area of the blind spot. With the low intensity blue stimulus, the center/periphery ratio of the response densities is lower than with the high intensity white stimulus. Response densities and focal trace arrays are presented in retinal view.
Figure 6
 
Responses of Subject 1 (left) and 2 (right) to the blue light stimulus spectrum after high-pass filtering are shown in retinal view to match the location of fundus structures. Some local depressions fall in the region of blood vessel shadows, especially along the upper and lower arcades. Other regions of depressed responses do not fall within the location of retinal vasculature.
Figure 6
 
Responses of Subject 1 (left) and 2 (right) to the blue light stimulus spectrum after high-pass filtering are shown in retinal view to match the location of fundus structures. Some local depressions fall in the region of blood vessel shadows, especially along the upper and lower arcades. Other regions of depressed responses do not fall within the location of retinal vasculature.
Figure 7
 
Representative noise distribution derived from the record with blue light stimulation for Subject 1 (left) and Subject 2 (right). The plots show the topography of a fifth order kernel slice known to contain no measurable response power. These noise topographies have undergone precisely the same type of processing and high-pass filtering used for the derivation of the response topographies of the first order kernel shown in Figure 6. As in Figure 6 responses are not divided by area but plotted directly and displayed in retinal view. These noise topographies are thus representative of the noise contamination in the plots of Figure 6. In both subjects, the noise is randomly distributed across the 509 stimulus patches.
Figure 7
 
Representative noise distribution derived from the record with blue light stimulation for Subject 1 (left) and Subject 2 (right). The plots show the topography of a fifth order kernel slice known to contain no measurable response power. These noise topographies have undergone precisely the same type of processing and high-pass filtering used for the derivation of the response topographies of the first order kernel shown in Figure 6. As in Figure 6 responses are not divided by area but plotted directly and displayed in retinal view. These noise topographies are thus representative of the noise contamination in the plots of Figure 6. In both subjects, the noise is randomly distributed across the 509 stimulus patches.
Figure 8
 
Local amplitude deviations above (pink hexagons) and below (green hexagons) 1.5 SDN from the average noise level. Black hexagons mark local depressions below 1.5 SDN for elements that project onto blood vessels. In both Subject 1 (left) and 2 (right), there is good correspondence between the depressed areas and the upper and lower arcade, as well as with some smaller vessels. However, elements that are not affected by blood vessel shadows also show significant depressions evenly spread in subject 1 and concentrated on the temporal retina in Subject 2. Significant local amplitude enhancements (pink hexagons) are sparsely distributed.
Figure 8
 
Local amplitude deviations above (pink hexagons) and below (green hexagons) 1.5 SDN from the average noise level. Black hexagons mark local depressions below 1.5 SDN for elements that project onto blood vessels. In both Subject 1 (left) and 2 (right), there is good correspondence between the depressed areas and the upper and lower arcade, as well as with some smaller vessels. However, elements that are not affected by blood vessel shadows also show significant depressions evenly spread in subject 1 and concentrated on the temporal retina in Subject 2. Significant local amplitude enhancements (pink hexagons) are sparsely distributed.
Table 1
 
Response Ratios Between Responses to the Blue Light Stimulus and White or Green Light Stimulus Respectively
Table 1
 
Response Ratios Between Responses to the Blue Light Stimulus and White or Green Light Stimulus Respectively
Focal flash intensities (cd/m2) Subject 1 Subject 2
white/blue 8 : 1 1.33 1.14
white/blue 1 : 1 0.538 0.77
green/blue 8 : 1 1.28 1.18
green/ blue 1 : 1 0.588 0.704
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×