back to the Main Thibos Page
back to the Main Thibos Page
The central theme of my research program is the role of the eye's optical system and neural retina in determining the quality of our visual experience and in setting the limits to visual performance. I have approached this broad theme from several different directions, using a variety of experimental paradigms, techniques, and models. As a graduate student at Berkeley, and subsequently as a postdoctoral fellow and eventually as Research Fellow at the Australian National University (JCSMR), I concentrated on neurophysiological investigations of retinal information processing in animal models. After moving to Indiana University in 1983 I expanded into the domain of human experimentation and broadened my interests to include the optical system of the eye. Throughout these changes, the ultimate goal remained the same: to understand how the physiology of the eye determines the way we see, when we see clearly, and explains the failure to see, when we see poorly.
My graduate work began under R.D. Freeman 1,3,4 with a study of amblyopia, one of the most intriguing visual dysfunctions known to clinical optometry. Amblyopia is characterized by a loss of visual acuity for no apparent reason and our experiments were aimed at establishing where in the visual pathways lies the defect responsible for this condition. Using a combination of psychophysical and electrodiagnostic techniques, we examined a particular kind of amblyopia, for which vision is acute when patterns are at one orientation but not at another. Since this condition occurs only in persons with strong ocular astigmatism during childhood, the working hypothesis was that uncorrected astigmatism during the critical period of development produces a deprived visual environment which eventually leads to a permanent visual disability. Our experimental results provided the first direct evidence from humans to support this hypothesis and pointed to the initial stages of the visual system, possibly the retina, as the locus of meridional amblyopia.
Some years later, I was awarded an NIH postdoctoral fellowship to join W.R. Levick in Australia to test this idea in the cat animal model. We found that the simplest hypothesis, that the receptive fields of retinal neurons are modified by astigmatic visual deprivation, is not correct 12. Since we also found no optical defect from the deprived rearing conditions, we concluded that meridional amblyopia is an abnormality of the brain, not the eye. Only recently have A. Bradley and I arrived, by quite different paths, at yet another retinal hypothesis for amblyopia based on the idea of retinal undersampling which we are currently exploring.
Levick and I also studied the problem of pigment-related visual dysfunction. Albinism is a genetic abnormality in which a lack of the pigment melanin is associated with a misrouting of optic nerve fibers: axons of retinal ganglion cells project to the wrong side of the brain. This curious association between pigment production and disorganization of the visual pathway led us to study the blue-eyed white-cat, another variety which lacks pigment but is not albino. Surprisingly, our neurophysiological experiments indicated that the white cat has normal visual pathways 8. To solve this puzzle we undertook further histological experiments to determine the basis for hypopigmentation in the two types of cat. We found that the white cat lacks the melanocytes necessary for pigment production, whereas the albino has melanocytes but these cells fail to synthesize melanin, evidently because they lack the enzyme tyrosinase 9. Since the white cat has normal tyrosinase (as indicated by normal pigmentation of the retinal epithelial cells), we suggested that abnormal development of the visual pathways may depend fundamentally on a defective enzyme, and that reduced pigmentation is merely one reflection of that defect.
A third class of retinal disorders are the acquired and disease processes. Cats are particularly susceptible to dietary insufficiency of the amino acid taurine which leads to photoreceptor death and blindness in the central region of the retina. While I was in Australia, Levick and I began a long-term neurophysiological and histological study of this disease with the goal of understanding the consequences of photoreceptor death and the loss of photic input to retinal ganglion cells. Results so far indicate that ganglion cells outside the photoreceptor lesion have normal response properties, cells inside the lesion are blind, and cells on the border of the lesion have abnormal sensitivity and unusual receptive fields.5B Surprisingly, conduction of nerve action potentials by optic nerve fibers was normal despite the fact that some cells had probably not generated an action potential spontaneously for several years prior to our experiments.
Perhaps because of my engineering background, I became especially attracted as a graduate student to those tractable problems of retinal research which appear amenable to quantitative analysis. For my dissertation research I took up the challenge of quantifying F.S. Werblin's well-known, qualitative model of neural adaptation in the vertebrate retina based on studies of the mudpuppy. The basic idea was that if sustained adaptation of bipolar and ganglion cells is mediated by lateral antagonism occurring at the outer plexiform layer from the network of horizontal cells, then the graded antagonism measured in bipolar and ganglion cells should be closely correlated with the graded responses of horizontal cells. Similarly, the transient adaptation of ganglion cells by the network of amacrine cells at the inner plexiform layer should be correlated with the response properties of amacrine cells. By independently measuring the adaptation effects and the response properties of the antagonistic neurons over a broad range of stimulus parameters and configurations, I was able to give quantitative support to Werblin's model 5,6 Furthermore, analysis of the data indicated that the apparently nonlinear (divisive) nature of adaptation in bipolar cells could be explained by a linear (subtractive) feed-back synaptic-mechanism at the outer plexiform layer. On the other hand, transient adaptation of ganglion cells appeared to be due to a linear, feed forward mechanism at the inner plexiform layer
Following my graduate work on amphibian retina I moved to Australia to learn the mammalian preparation from one of the world's experts, W.R. Levick. Together we explored for eight years the rich field of spatial information processing by receptive fields of retinal ganglion cells. At that time a definitive scheme for classifying retinal ganglion cells had just been proposed by Cleland and Levick for the cat and we aimed to see if the same scheme could be applied also to the rabbit, an animal thought by some to have quite a different retinal organization. Although obvious quantitative differences emerged, we found that the qualitative behavior of ganglion cells in rabbit and cat were quite similar 11 , thus supporting the emerging generalization that the mammalian nervous system contains not one visual pathway but several distinct, parallel visual projections from retina to brain.1B
In order to assess the functionality of the various parallel pathways of the cat retina, we set out to characterize the response properties of the major classes of retinal ganglion cells from within the conceptual framework of spatial frequency analysis. In this context, the long-term goal is to describe neural responses to the sinusoidal grating stimulus as a function of spatial frequency, contrast, orientation and phase, for a variety of cell locations across the visual field. As a first step we showed that different cell types identified by Cleland and Levick, based on the qualitative assessment of response to simple geometrical targets like spots, bars and disks of light, can also be distinguished with sinusoidal gratings.14 Within a given functional class, response properties varied systematically with eccentricity of the cell from the central area. At any given eccentricity, each class gives a distinctly different weighting to the spatial frequency spectrum and each has a characteristic limit to the finest detectable grating. Thus we developed the viewpoint that each functional class of retinal ganglion cell filters the retinal image differently and so passes on its unique view of the visual world to the brain.
Several of our experiments produced results which challenged established views on the functional architecture of the retina. One of the more interesting examples concerned the basic structure of the concentric type of receptive field. Since its first discovery over 30 years ago, the concentric field has been described as the union of two distinct components, the center and antagonistic surround, both of which are roughly circular in shape and concentrically positioned. However, we found that a number of cells had double-centers which were characterized by unusual behavior for high-frequency gratings that was reminiscent of spurious resolution in optical systems. 16
Perhaps the most interesting feature of retinal organization we uncovered in the cat is radial tuning. Previously it had generally been assumed that receptive fields of retinal ganglion cells are circular, but we found that most fields are actually elliptical. Furthermore, these elongated fields are systematically organized across the retina with the long axis of the ellipse oriented radially like the spokes of a wheel with its hub centered on the region of highest density of retinal ganglion cells, where the cat has it's greatest visual acuity. 10,13,17 This discovery led to subsequent work by others indicating that elongation of dendritic fields is the anatomical basis for radial tuning in ganglion cells and that similar tuning occurs also in primate retina and cat visual cortex. Those results have led in turn to several interesting hypotheses about the developmental mechanisms responsible for radial tuning and the role of radial tuning in the establishment of orientation columns in visual cortex. In my own work, the possibility of radial tuning in humans led to new experiments on peripheral vision at Indiana University described further on.
Another line of investigation that Levick and I pursued was to establish the physical and biological factors which limit the ultimate sensitivity of retinal neurons for detecting dim flashes of light. We approached this problem from within the conceptual framework of signal detection theory, as developed for the context of neural signals by T.E. Cohn. As a graduate student, I had developed some appreciation for the power of this approach in a psychophysical study with Cohn in which we showed that, for a simple detection task, the human observer acts like an ideal photodetector with imperfect memory for stimulus parameters. 2 To set the stage for the corresponding experiment on single neurons, it was necessary for us to first advance the theory of ideal detectors of Poisson signals. 7 We were then in a position to test the hypothesis that detection of brief increments and decrements of light by ganglion cells is limited only by the inherent, quantal fluctuations in the light itself. The results showed that although quantal fluctuations are an important limiting factor, quantitative agreement between theory and experiment was possible only when a second, biological source of noise was postulated. 15 As a result of these experiments, it became necessary to reassess the broader issue of the meaning of the term "quantum efficiency" and to develop a new method of approach which incorporates both the physical and biological limits to detection. 2B
I arrived at Indiana in 1983 with a rich experience in retinal physiology and a desire to better understand the human visual system. Initially, I set out to determine whether radial tuning exists in humans as it does in the cat. The logical approach seemed to be to test for changes in visual acuity with stimulus orientation in the peripheral field. However, in the course of these experiments I discovered unexpectedly that patterns much too fine to be resolved were nevertheless quite visible. The very idea that vision might exist beyond the resolution limit was very exciting because it was totally outside the conventional framework for thinking about peripheral vision that had existed, unchallenged, for over a century of vision research. The obvious potential for revolutionizing our understanding of peripheral vision compelled me to pursue these observations experimentally and, at the same time, to completely re-think the question of how retinal factors might limit visual performance. This Indiana phase of my work has now grown to include the areas of pattern resolution, pattern detection and contrast sensitivity of the periphery plus the related issue of retinal image quality for off-axis peripheral objects.
Retinal factors have long been suspected of limiting human visual performance for fundamental tasks like pattern resolution and pattern detection. Sampling theory predicts that, if the filtering effects of the eye's optical system are avoided, then visual resolution should be limited by the spacing of retinal sampling units, the so-called Nyquist limit. Since ganglion cells form a sparser array than do the photoreceptors, a stronger prediction is that visual resolution should vary with stimulus eccentricity from the fovea in accordance with the spacing between retinal ganglion cells. We have tested this theory experimentally by measuring resolution for sinusoidal grating stimuli formed directly on the retina as interference fringes, a technique which can produce gratings of nearly unit contrast independent of spatial frequency and optical focus. Quantitative comparison of the results with published data on retinal anatomy and physiology of the primate retina indicates that pattern resolution is limited by the spacing of primate beta ganglion cells, a class of neurons which provides the major retinal input to visual cortex, thus confirming the theoretical expectation. 19
A further prediction of sampling theory is that if retinal images contain spatial frequencies beyond the Nyquist limit, then these images will be undersampled and therefore will not be perceived veridically, a phenomenon called aliasing. By using the interferometric stimulus, we were able to apply such stimuli and so test the prediction experimentally. The results confirm that aliased patterns are detectable for the full range of spatial frequencies extending from the resolution (Nyquist) limit up to the finest pattern detectable by individual cone photoreceptors.18 Surprisingly, spatial integration over the receptive fields of ganglion cells does not appear to limit detection acuity.
The optical system of the eye presents a low-pass filter at the front end of the visual system and thus represents a potential limiting factor for any visual task. Optical limitations are especially likely in the peripheral field because of the presence of off-axis aberrations. Having identified the retinal limits to performance when optical factors are avoided with the interferometer stimulator, the next logical step was to re-introduce the optical apparatus of the eye by using ordinary test stimuli (e.g. CRT display). Results indicate that resolution acuity outside the fovea is unaffected by the introduction of the eye's optical system.2B This proves that, for natural viewing conditions, peripheral resolution is determined by neural, not optical, factors. The results further showed that perceptual aliasing exists for natural viewing when the stimulus frequency is higher than the Nyquist limit of the retinal network. Thus we have established conclusively that vision beyond the resolution limit occurs under natural viewing conditions in peripheral vision.
The highest detectable spatial frequency for natural viewing was found to be less than that found for interferometric viewing. We conclude from this result that optical filtering becomes increasingly important as spatial frequency increases and, in the limit, it is optical quality which becomes the dominant factor that determines detection acuity. Because of the anisotropic nature of off-axis optical aberrations, we may further predict that detection acuity should vary with stimulus orientation and be maximal when grating stimulus is radially oriented. This prediction was confirmed experimentally and found to be of the same order of magnitude expected on the basis of optical theory.3B, 4B
As a result of our finding that detection acuity in peripheral vision is limited by retinal image quality, it has become apparent that to fully appreciate the limits to visual performance requires an understanding of off-axis optical performance of the human ocular system. Although several of the chief optical aberrations which affect peripheral vision have been studied by others, no thorough assessment of ocular chromatic aberration has previoulsy been attempted. To fill this gap, we began by developing the theory of ocular chromatic aberration through analysis of a simple model of the eye's optical system. The goal was to devise a realistic model capable of making explicit, quantitative predictions suitable for experimental testing.
Our theoretical development has provided for the first time a unified treatment of both the transverse and longitudinal components of ocular chromatic aberration.22 This was achieved through analysis of a model eye consisting of a volume of water with a single, spherical refracting surface and an internal aperture. An important conceptual result of the analysis was the identification of a new reference axis, called the achromatic axis, which is of crucial importance in specifying the magnitude of transverse chromatic aberration. We call the angle between the achromatic and visual axes Y in honor of Indiana University. We have shown that angle Y is the primary determinant of how much transverse chromatic aberration is present for foveal vision. When present, chromatic aberration of the eye reduces image contrast for two reasons. First, the longitudinal aberration defocusses the image by an amount which varies with wavelength. Second, the transverse aberration induces a phase shift that varies with wavelength which also results in a loss of contrast. On the basis of our model we have computed the expected loss of image contrast for peripheral stimuli located anywhere in the visual field. The results are presented as a family of optical transfer functions parameterized by stimulus eccentricity and orientation.20
We have also analyzed the performance of the interferometric stimulator we used in the peripheral vision experiments described above. This is one example of how the model eye can be used to solve applied problems. Although the results for natural viewing are more relevant to every-day visual performance, the interferometer results are especially convincing because this stimulator is insensitive to aberrations of focus (astigmatism, refractive error) and so isolates the chromatic aberration effect for experimental study.
The model of ocular chromatic aberration developed above makes quantitative predictions about the magnitude of both the transverse and longitudinal chromatic aberration of the eye. Although measurements of the longitudinal component are available in the literature, no useful measurements have been reported for the transverse component. We therefore devised an experimental method for measuring transverse chromatic aberration based on a two-color, vernier-alignment task. The results closely followed theoretical predictions thus supporting the postulated model. Unexpectedly, our measurements indicate that angle Y is nearly zero in the majority of subjects. The implication of this result is that, although the eye has substantial chromatic aberration, the pupil is positioned so as to minimize the transverse component of the aberration for central vision, thereby optimizing image quality for foveal viewing of polychromatic objects.22
Additional experiments on a population of aging individuals has indicated that, contrary to previous reports, the magnitude of chromatic aberration does not vary significantly with age.21
If the contrast of the retinal image is reduced because of chromatic aberration, as we have argued above, then visual performance should suffer. We have tested this idea four ways. First, we measured the loss of resolution acuity of foveal vision for white gratings viewed normally and second when the grating is produced interferometrically. In both cases acuity dropped about three-fold when the entrance pupil was displaced from its normal location near the achromatic axis to a point near the margin of the iris. In the third experiment we tested for orientational anisotropy with natural viewing and again in a fourth experiment using the interferometer. In each case the qualitative predictions were verified and found to be of about the same magnitude as expected from the quantitative predictions of the water eye model.
One possible way to avoid the deleterious effects of ocular chromatic aberration is to correct the aberration with external optical lenses. The idea is to introduce an equal but opposite aberration which cancels the eye's aberration. Although such achromatizing lenses do an acceptable job of correcting the longitudinal component of ocular chromatic aberration, they may introduce extra transverse chromatic aberration. The net result can be an overall loss of image quality. We have calculated that one well-known lens design will magnify the transverse aberration five-fold. Experimental measurements confirmed the prediction. We conclude that in many applications the penalty of increased transverse aberration introduced by the achromatizing lens will greatly outweigh any benefit from correction of the longitudinal aberration.
2B. Thibos, L.N., Still, D.L. Optical and neural factors limiting resolution acuity and detection acuity across the visual field. (in preparation)
3B. Walsh, D. J. and Thibos, L.N. Orientational anisotropy for resolution of gratings in peripheral vision. (in preparation)
4B. Cheney, F.E. and Thibos, L.N. Orientational anisotropy for detection of gratings in peripheral vision. (in preparation)
5B. Levick, W.R. and Thibos, L.N. Neurophysiology of central retinal degeneration in cat. (in preparation)