Lateral interaction component and local luminance nonlinearities in the human pattern reversal ERG

A Novel Neural Model With Lateral Interaction for Learning Tasks

We propose a novel neural model with lateral interaction for learning tasks. The model consists of two functional fields: an elementary field to extract features and a high-level field to store and recognize patterns. Each field is composed of some neurons with lateral interaction, and the neurons in different fields are connected by the rules of synaptic plasticity. The model is established on the current research of cognition and neuroscience, making it more transparent and biologically explainable. Our proposed model is applied to data classification and clustering. The corresponding algorithms share similar processes without requiring any parameter tuning and optimization processes. Numerical experiments validate that the proposed model is feasible in different learning tasks and superior to some state-of-the-art methods, especially in small sample learning, one-shot learning, and clustering.

Contributions from lateral interaction mechanisms to the human ERG can be studied with a two-frequency method

We present a two-frequency method to investigate lateral interaction components (liERGs) in the human electroretinogram. Adjacent half cycles of sinusoidal gratings were modulated sinusoidally with different temporal frequencies f1 and f2. The liERGs were defined by the Fourier components at the intermodulation frequencies /f1 - f2/ and f1 + f2 which indicate nonlinear interactions between the half cycles. Significant liERGs were found in all subjects with a monotonic increase of the liERG magnitude in the spatial frequency range from f(s)=0.07 to 2.4 cpd. When /f1 - f2/ was below 5 Hz, liERGs were masked by noise intrusions. In a control experiment we demonstrated that the liERGs were not evoked by stray light artifacts. The liERGs may help to further differentiate the responses that are evoked by patterned stimuli within the retina.

Scanning laser ophthalmoscope-evoked multifocal ERG (SLO-mfERG) in patients with macular holes and normal individuals

AIMS

A scanning laser ophthalmoscope (SLO) has been used for multifocal electroretinography (mf ERG) measurements under simultaneous fundus monitoring. The aim of this study was to prove if the SLO-mfERG measurement reflects reliably the clinically registered underlying disease, and to demonstrate the importance of its main advantage, fixation monitoring.

METHODS

In all, 10 patients with macular hole stage II/III were included in the study, and 19 normal individuals served as the control group. The mf ERG device was combined with an SLO, which was used both as a stimulus and trigger unit as well as a fundus-monitoring system. Monitoring of the fundus was guaranteed by an infrared laser (780 nm). The stimulus matrix consisted of 61 hexagonal elements, covering 24 degrees of the posterior pole. We examined both, patients with macular holes and healthy individuals.

RESULTS

Compared to normal controls, patients with a macular hole (Gass stage III) showed a significant decrease in response density in the centre of the stimulus array, which correlated well with the morphological alteration observed by clinical examination. However, variation of response density of the central hexagonal area has been proved to be high.

CONCLUSIONS

SLO-mfERG is a feasible and reliable new technique to investigate macular function under simultaneous fundus control. The main advantage is that control of fixation can be used in order to obtain more reliable results that correlate well with visible fundus abnormalities such as in patients with macular holes. However, further investigations have to be performed in order to overcome sufficiently the problem of fixation instability.

Transient and steady state focal and pattern electroretinogram nerve section losses in cats with unilateral optic

This study shows the ketamine/xylazine anaesthetised cat is a useful model for the effect of unilateral optic nerve section on pattern electroretinograms (PERGs), especially if stimuli extending to previously untested low spatial frequencies and preferably down to the focal ERG (FERG) are included. The transient reversal rate, seldom used in animals,has advantages over steady state recording. Transient PERGs had signs of true spatial tuning, a higher amplitude and signal noise ratio and showed the effect of optic atrophy at low spatial frequencies more rapidly.

Pilot study of the multifocal electroretinogram in ocular hypertension

AIMS

To investigate the variation of retinal function in ocular hypertension (OHT) using the multifocal electroretinogram (MERG), that can objectively assess regional retinal responses using kernel analysis.

METHODS

Patients with OHT were recruited for 61 flash MERG recording. The first and second order kernel responses were analysed and compared with responses from a normal group. All the patients with OHT had full eye examinations and visual field analysis to ensure they had no ocular pathology, apart from high IOP (>/= 22 mm Hg).

RESULTS

In OHT, both the first and the second order kernel responses showed a reduction in magnitude compared with control values. The second order kernel responses showed larger relative reduction of amplitude than the first order responses. In addition, the macula showed a greater reduction in response than the periphery.

CONCLUSIONS

First order and second order kernel analyses are useful for detecting changes of retinal response in OHT. The second order kernel analysis is important in measuring the inner retinal activity and is an important factor in detecting the early glaucoma case. Diminished macular response may be a useful sign in early glaucomatous changes.

The multifocal ERG in open angle glaucoma--a comparison of high and low contrast recordings in high- and low-tension open angle glaucoma

High and low contrast multifocal ERG (MF-ERG) recordings were obtained from the right eyes of 24 patients with OAG (high-tension OAG: n=16, low-tension OAG: n=8) and compaired to those recorded from 18 healthy volunteers. High contrast MF-ERG recordings were obtained at a mean luminance of 100 cd/m2 with a contrast of 99%, while low contrast MF-ERGs were obtained at a mean luminance of 100 cd/m2 with a contrast of 50%. During MF-ERG recordings the central 50 degrees of the retina were stimulated by 103 hexagons. A MF-ERG recording lasted eight minutes, a M-sequence of 2(15) was used. The first order response component (KI, mean focal flash response) and the first and second slice of the second order response component (mean focal two flash interaction of flashes one, KII. 1, or two, KII.2, base intervals apart) were analyzed for group differences. Group differences were found mainly in latency measures. These included a delay in the central response average of the first positive peak, P1, in KII.2 (p < or = 0.05) in OAG high contrast recordings. Low contrast recordings showed a significant delay in the central response average of the first negative peak, Nl, in KII.2 as well as in the peripheral response average of N1 in KI and of P1 in KII.2 (p<0.05) in OAG. Amplitudes were only affected significantly in KI of the low contrast recordings. Here the amplitude N1P1 was significantly higher in high tension (n=16) than in low tension (n=8) OAG patients. However, an overlap in all of the response parameters tested allowed only group differences to be characterized. Under these stimulus conditions, neither high contrast recordings nor low contrast recordings seem sensitive enough to reliably recognize early glaucomatous retinal dysfunction.

The optic nerve head component of the human ERG

The local responses of the multifocal ERG reveal continuous changes in the second order waveforms from the nasal to the temporal retina. Scrutiny of these changes suggests the presence of an additive component whose latency increases with the distance of the stimulus from the optic nerve head. This observation led to the hypothesis of a contributing source in the vicinity of the optic nerve head whose signal is delayed in proportion to the fiber length from the stimulated retinal patch to the nerve head. The hypothesis was tested with two independent methods. In Method 1, a set of different local response waveforms was approximated by two fixed components whose relative latency was allowed to vary and the fit of this two component model was evaluated. In Method 2, two signals were derived simultaneously using different placements for the reference electrode. The placements were selected to produce a different ratio of the signal contributions from the retina and the nerve head in the two recording channels. The signals were then combined at a ratio that canceled the retinal component. Method 1 yielded an excellent fit of the two component model. Waveforms and latencies of the hypothetical optic nerve head component derived from the two methods agree well with each other. The local latencies also agree with the propagation delays measured in the nerve fiber layer of the monkey retina. In combination, these findings provide strong evidence for a signal source near the optic nerve head.

Temporal analysis of the topographic ERG: chromatic versus achromatic stimulation

The topographic electroretinogram evoked by multi-focal exchange of black and white or red and green stimuli was analysed into linear and non-linear Wiener kernels. The first-order (temporally linear) response showed a biphasic waveform which inverted as the luminance ratio of the exchanged colours passed through unity (established both psychophysically and photometrically). A short latency non-linearity which was dependant on luminance contrast was observed in both chromatic and achromatic ERG. However, in the chromatic second-order response, a long-latency non-linearity, foveally prominent, with a distinct skew in power towards the nasal retina, appeared around the isoluminant point, between the points of silent substitution for the L and M-cone types. Modelling of the second-order responses showed that over a wide range of luminance ratios, the chromatic ERG is well described by a linear combination of the achromatic (contrast-dependent) component and the response at isoluminance. The difference in second-order response between coloured and black and white stimulation, at the same luminance contrast, showed that the long-latency non-linearity is recorded when the red and green cone types are operating out of phase and peaks in amplitude at a green/red luminance ratio of 0.8. This interpretation was confirmed by the lack of the long-latency non-linearity in colour-anomalous subjects (whether deficient in the L or the M-cone type). A marked similarity exists between the properties of the long-latency non-linearity and the frequency-doubled response generated in the ganglion cells of the magnocellular pathway.

Functional perimetry combined with topographical VEP analysis

The processing of visual input depends on the position of the visual stimuli in the visual field. Based on the anatomical structure of the retina and the cortex, the function and perception vary with the location in the visual field. Due to the low signal-to-noise ratio, electrophysiological recordings in human subjects commonly have to use large stimuli and, therefore, yield poor spatial resolution. The combination of the method of quasi-simultaneous stimulation of many small (1.5 degrees x 1.5 degrees squares) visual field elements by binary m-sequences and topographical recordings allowed us to reconstruct the potential maps elicited at each of 54 visual field locations independently. Twenty-two normal subjects participated in the experiments and observed monocularly a stimulation field of 13.5 degrees x 9 degrees filled with the 54 squares. Mean luminance was 6.5 cd/m2 and contrast was 95%. The EEG was recorded in 30 channels with a dense array of electrodes over the occipital brain areas. Individual noise levels of the subjects were estimated and significant signals were analyzed quantitatively. We determined three components between 90 ms and 220 ms latency. Both global field power (GFP) and topography of the components were affected by retinal stimulus location, showing a significant decline of GFP with retinal eccentricity. Our data demonstrate that even small retinal targets may evoke brain activity which can be recorded simultaneously. Scalp field topography depends critically on the exact stimulus location within the foveal and parafoveal retinal areas while response strength mainly depends on eccentricity.

Local luminance and pattern reversal stimuli yield different visual evoked potential topography

We studied how the stimulation of quadrants of the visual field affect brain potential topography, and we compared activity elicited by conventional pattern reversal or by local luminance stimuli. The method of quasi-simultaneous stimulation of many small visual field elements by binary m-sequences allowed us to reconstruct the potentials evoked at each of 54 visual field locations independently. Data from all field elements within each quadrant and in the whole stimulation field were summed and compared to those elicited by checkerboard reversal stimuli presented in the four quadrants or as full field stimuli. In twenty-two healthy adults evoked brain activity was recorded in 30 channels with an electrode array densely spaced over the occipital brain areas. With local flash stimuli as well as with checkerboard reversal the topographical distributions of cortical activation changed significantly with retinal stimulus location. Analysis of three components occurring between 50 and 240 ms revealed significant differences between pattern reversal and local luminance evoked brain activity. Reversal stimuli yielded not only larger amplitudes but also a completely different component structure and topography. Our results illustrate that different neuronal generators are activated by pattern reversal and local luminance stimuli although visual field location of the stimuli was identical indicating that the same retinal and cortical structures respond in a different way depending on stimulation mode.

The use of m-sequences in the analysis of visual neurons: linear receptive field properties

We have used Sutter's (1987) spatiotemporal m-sequence method to map the receptive fields of neurons in the visual system of the cat. The stimulus consisted of a grid of 16 x 16 square regions, each of which was modulated in time by a pseudorandom binary signal, known as an m-sequence. Several strategies for displaying the m-sequence stimulus are presented. The results of the method are illustrated with two examples. For both geniculate neurons and cortical simple cells, the measurement of first-order response properties with the m-sequence method provided a detailed characterization of classical receptive-field structures. First, we measured a spatiotemporal map of both the center and surround of a Y-cell in the lateral geniculate nucleus (LGN). The time courses of the center responses was biphasic: OFF at short latencies, ON at longer latencies. The surround was also biphasic--ON then OFF--but somewhat slower. Second, we mapped the response properties of an area 17 directional simple cell. The response dynamics of the ON and OFF subregions varied considerably; the time to peak ranged over more than a factor of two. This spatiotemporal inseparability is related to the cell's directional selectivity (Reid et al., 1987, 1991; McLean & Palmer, 1989; McLean et al., 1994). The detail with which the time course of response can be measured at many different positions is one of the strengths of the m-sequence method.

Absence of ganglion cell subcomponents in multifocal luminance electroretinograms

Multifocal flash electroretinograms (ERG) were recorded binocularly (n = 18). Areas were equal or scaled with excentricity. The latter were expected to increase the total amplitude if ganglion cell subcomponents were involved. Amplitudes were intercorrelated and the factor structure was established. Scaling had no influence on amplitudes or on factors. Reliability and sensitivity were high. The second kernel first slice from the nine hexagons across the midline showed differences near the macula but no component increasing in latency with distance from the disc. Thus, multifocal flash ERG have linear spatial summation and no ganglion cell subcomponents.

Bispectral analysis of visual interactions in humans

Previous electrophysiological studies have demonstrated interactions between dichoptic visual stimuli presented to the same location in visual space. In this study, we used non-liner spectral analysis, in particular the bispectrum, to study interactions between the electrocerebral activity resulting from stimulation of the left and right visual fields. The stimulus consisted of two squares, one in each visual field, flickering at different frequencies. Bispectra, bichoherence and biphase were calculated for 8 subjects monocularly observing a visual stimulus. Both phase vs. frequency and biphase vs. frequency plots were made to determine weighted time delays from stimulus application to signal appearance in the EEG electrodes. Bispectral analysis reveals non-liner interactions between visual fields occurring with weighted delay times of 410 + / - 58 msec while non-interactive components propagated with weighted time delays of 202 + / - 39 msec. Evaluating these results in light of the predictions of various models, we were able to conclude that this interaction does not occur in the retina. These results illustrate how bispectral analysis can be a powerful tool in analyzing the connectivity of neural networks in complex systems. It allows different neuronal systems to be labeled with stimuli at specific frequencies, whose connections can be traced using frequency analysis of the scalp EEG.

Flash and pattern electroretinogram changes with optic atrophy and glaucoma

We investigated recent reports that, contrary to common belief, glaucoma can affect flash as well as pattern electroretinograms. An extensive flash and pattern electroretinogram test protocol was used in a large sample of glaucoma patients and age-matched controls who were either visually normal or had other optic nerve diseases. All electroretinogram parameters were reduced and delayed in normal people > 55 years of age. The effect did not increase in later decades. In patients aged < or = 55 years, flash electroretinograms showed mild reductions and delays from optic atrophy alone. Glaucomatous ERG changes were larger and increased with disease severity. Pattern electroretinograms and oscillatory potentials were almost equally reduced in optic atrophy and all degrees of glaucoma. Mildly affected patients > 55 years of age had similar electroretinogram change to age-matched normals in most conditions. Advanced glaucoma patients showed similar differences from normal irrespective of age. This suggests that direct diagnostic application of these results to older patients will be difficult, that the ERG changes in glaucoma cannot be attributed simply to optic atrophy and that additional widespread outer retinal damage occurs in glaucoma.

Effect of kainic acid and NMDA on the pattern electroretinogram, the scotopic threshold response, the oscillatory potentials and the electroretinogram in the urethane anaesthetized cat

Kainic acid (KA, 12.5-100 nmol) or N-methyl-D-aspartate (NMDA 25-250 nmol) was injected into the vitreous of one eye of urethane anaesthetized cats. Pattern electroretinograms (PERGs) were recorded to transient contrast reversing bars. Scotopic luminance electroretinograms (ERGs) were recorded to blue flashes. All doses of KA reduced the oscillatory potentials (OPs), PERG and focal ERG (FERG). At 50 nmol KA, the b-wave and scoptic threshold response (STR) were normal. At 100 nmol KA, the STR was absent and the b-wave reduced by over 50%. OPs and STRs were reduced in all NMDA injected eyes. NMDA at 25 nmol enhanced the FERG, PERG, and b-wave and high doses (above 150 nmol) reduced them. Light microscopic examination of retinas showed 25 nmol KA only damaged dendrites of ganglion cells. NMDA damage was slight with < 200 nmol. These data show that the cat PERG has a proximal component which is very sensitive to low doses of KA; the PERG and FERG are very similar; the STR and PERG are generated by different structures and that the OPs and the FERG and PERG are all generated close to the ganglion cell layer, proximal to the STR.

The topography of visual evoked response properties across the visual field

Visual evoked potentials (VEPs) to luminance and pattern reversal stimulation were derived for a large number of small areas throughout the central visual field. In one study, the field was tested with a stimulus array consisting of 64 equal-area patches. Local response components were extracted by independent m-sequence modulation of the patches. Field topographies were compared between and within subjects using different electrode placements. The subject-dependent local variability observed in response characteristics is attributed to contributions from two or more cortical representations of the visual field and to inter-subject variations in gross cortical anatomy. The second study used luminance modulation of 56 patches across a 15 degrees field, scaled to activate approximately equal cortical areas in area V1. This produced many robust signals at all eccentricities. Bipolar and double differential ("1-dimensional Laplacian") signals were compared. The double differencing reduced contributions from distant or distributed sources, enhancing nearby current source activity, and greatly improved S/N for many stimulus locations. The high-resolution visual field maps demonstrated that clinical field testing using the VEP is not feasible because of effects of cortical convolutions on responses. However, the vast improvement in data quality and quantity make it a useful tool for VEP source localization and identification.

An analysis of the VEP to luminance modulation and of its nonlinearity

Identification of the VEP luminance modulation system was carried out using a relatively long pseudorandom binary m-sequence stimulus. The complete first order cross-correlation function provides a "fingerprint" of the nonlinearity in the time domain. Volterra kernel slices up to third order are recognized in the complete first order cross-correlation function. They all lie close to the major diagonal. Higher order kernel slices are not measurable. A cascade block model (sandwich model) analysis shows the nonlinearity to be an asymmetric rectification. Luminance increase causes a stronger response than does luminance decrease. The pre-filter gain function is low-pass. The post-filter gain function is band-pass tuned near 10 Hz. Coherency spectra are used to examine the channel structure of the system. Two channels operating near 8 and 15 Hz are easily identified. Another broad channel (or possibly multiple channels) is present above 30 Hz. In many subjects, the pseudorandom stimuli allow a complete system identification to be carried out in less than 82 sec.

The field topography of ERG components in man--I. The photopic luminance response

A technique of multi-input systems analysis is used to explore the field topography of ERG responses to local luminance modulation. Variations in amplitude and wave form are studied within the central 23 degrees. Outside the fovea, the amplitude appears to follow a simple power law rx as a function of eccentricity r where x is approximately -2/3. The largest inter-subject variability is found in the fovea. Some nasal-temporal asymmetry is observed in all subjects with higher response densities in the temporal field outside the blind spot. The topography of the luminance response shares all these properties with the density of retinal cones.

Evidence for two distinct nonlinear components in the human pattern ERG

We obtained the electroretinogram (PERG) in three observers to a 4.6 c/deg grating pattern with a sinusoidal luminance profile which was modulated in time simultaneously with two sinusoidal temporal frequencies (f1 and f2), or at a single frequency (either f1 or f2). Input temporal frequency ranged from 2.4 to 7.5 Hz. A pattern modulated in the counterphase mode with a single frequency produces a response containing only even harmonics of that frequency. However, when the pattern is counterphase modulated with both f1 and f2, the PERG contains second order response frequencies corresponding to intermodulation components (sums and differences) between the two fundamental frequencies. Such frequencies do not exist in the stimulus--they can only be generated by nonlinear neural interactions. Our results provide evidence for at least two nonsaturating nonlinear response mechanisms in the human retina.

Fundamental differences between the nonlinearities of pattern and focal electroretinograms

We directly compared nonlinear kernels of normal human pattern electroretinograms (PERGs) and corresponding localized flash ERGs (FERGs). The FERG was triphasic and resembled an adaptive process because it decayed slowly without changing shape over several kernel orders and interpulse intervals. The PERG was biphasic in the slice nearest the diagonal of the second-order kernel, similar to the FERG in slices farther from this diagonal, and without power in higher-order kernels. The unique PERG features were short-term effects that immediately followed a contrast transition. The appearance-disappearance PERG had a triphasic first-order kernel and a biphasic second-order kernel. The latter was similar to, but half the size of, that for the contrast-reversal PERG. When the first off-diagonal slices of the two PERG second-order kernels were analyzed in detail, we found in both that the first positive peak was larger than the FERG at intermediate spatial frequencies. Both PERG peaks in the slice had a low contrast threshold and were linear with contrast. The three FERG peaks of the corresponding FERG slice had a higher threshold and were saturated with increasing contrast. These observations show that the PERG contains substantial pattern specific nonlinear components and cannot be dismissed as merely the nonlinear subcomponents of the corresponding FERG.