Failing vision that doctors cannot correct with glasses or surgery is a frequent occurrence in aging, often caused by a condition called age-related macular degeneration (AMD).
Professor Norberto Grzywacz, the chair of the USC Viterbi School of Engineering Department of Biomedical Engineering leads the effort. He is working with other faculty from USC as well as scientists from Harvard and UC Berkeley.
AMD results from the progressive death of photoreceptor cells in the central area of the light sensitive retinal portion of the eye. However, AMD does not usually cause total blindness, with patients normally perceiving light and even being able to recognize objects at which they are not looking directly. These patients see with their peripheral vision.
However, using retinal periphery has drastic limitations for useful perception. The study that Dr. Grzywacz and his colleagues have been working on since 2007 is an effort to use deeper understandings of these limitations to find ways around them. This work is funded by the National Institutes of Health
Two teams worked separately, but in coordination with each other, on two separate and quite distinct peripheral vision limitations. Both teams had considerable success when their products were shown to human subjects.
One of these limitations is that AMD drastically reduces the contrast range of vision. Photographic processing has long included methods to increase contrast. However, most of these methods are insensitive to details of portions of the image. For example, a bright sky may make other areas of the image dark. Other methods can increase contrast in isolated areas of the image, but end up boosting simultaneously small noisy variations that are normally imperceptible. Thus, optimizing contrast for the specific problems of macular degenerative eyes is not a trivial process, particularly achieving this goal in real time. USC Viterbi Professor of Computer Science Gerard Medioni has been investigating these kinds of problems for years, and found effective optimizations.
Computerized enhancement will help AMD patients peripheral vision recognize edges and contrast differences
Contrast enhancement is not enough, because the peripheral retina has difficulty separating shapes that are close together due to a phenomenon called crowding. Therefore, the next critical area of enhancement is image segmentation through contour detection. These shape-defining contours mark the boundaries of image objects, helping both to combat crowding and to eliminate noise inside them. The contour-detection problem was the area attacked by Viterbi School Professor of Biomedical Engineering Bartlett Mel.
These new enhancement technologies were followed by a series of experimental tests. A third group of researchers, mostly psychologists, ran experiments testing the effectiveness of the solutions. The experiments ran first on people with normal sight with their macular areas blocked off and then on AMD patients.
Careful experimental protocols scored the results. They showed steady progression and improvement in tests administered by Bosco Tjan and Irving Biederman of the USC Dornsife College, Susana Chung from UC Berkeley, and Eli Peli from Harvard.
The next step is to develop technology and perform experiments to solve three more problems: First, the team must find ways to generalize the technology for movies. Second, algorithms must be developed to individualize the parameters of the system to optimize it for specific patients. Third, the team must begin importing the technology to augmented-reality systems. The effort continues and it may take a while, says Grzywacz, Director of the Center for Vision Science and Technology. However, the goal is clear: intelligent, image-enhancing spectacles that will allow AMD patients to explore the visible world.