Shape Distortions seen by Amblyopic Children: Pereceptograms and Neural Models - PROJECT SUMMARY Children need to learn many shapes, especially letters and numbers for reading. Amblyopic children read 25% slower than normal controls even when using both eyes. Although problems with fixation and saccadic control have been identified, shape misperceptions have not been examined. Striking perceptual distortions have been reported when amblyopic adults are asked to describe or draw oriented sinusoidal gratings or letters at sizes well above their visual acuity threshold. Shapes are processed through orientation selective neurons in primary visual cortex (V1), and a person would be severely handicapped in using vision if orientations were encoded or decoded erroneously. The distortions reported by roughly two-thirds of adult amblyopes are important because they are not explainable by existing theories of neural scrambling or shifts in the neural map but are compatible with aliasing by reduced neural sampling of orientation in V1. We have built a theory about how early differences between eyes in visual resolution or alignment impair development of cortical orientation selectivity. We discovered that amblyopes have a greater loss of acuity and salience for luminance increments (ON system) than decrements (OFF system). We also showed that a mosaic of cell properties, including spatial resolution and orientation selectivity, develops in V1 depending on the spatial gradient of ON and OFF balance in inputs to cortex. Together these two studies imply that the ON/OFF imbalance created by amblyopia will reduce spatial resolution and orientation selectivity in populations of V1 neurons. We are currently testing aspects and implications of this model for adult amblyopes. However, a critical missing piece for understanding the development of shape perception is knowing whether amblyopic children see distortions, what form these distortions take, when and how they develop, and whether they respond to treatment. If we could picture what amblyopic children perceive, we could answer these and many other questions. The drawings that 4–7-year- old children with amblyopia make of their percepts are unlikely to be accurate renderings, so we propose to adapt Generative Adversarial Networks to generate perceptograms that match in the good eye what the children perceive when they are shown gratings to the amblyopic eye. The image generating architecture will be trained on 30,000 sums of sines, and the adversarial discriminator driven by repeated choices of the closest match among multiple alternatives. A normal V1 model using marginal and correlational statistics of oriented cells will be used to compare responses of the perceptograms and their evoking stimuli, then reduced in coverage and selectivity to generate perceptogram responses from stimulus gratings. The model will be used to predict amblyopic children’s perceptograms of printed and handwritten letters and numbers from the EMNIST database, and the critical ones will be measured to understand if perceived shape distortions hamper reading and learning. This project will thus introduce powerful new tools to amblyopia research that are also likely to be used for many other childhood visual disorders.
