One way to judge the effectiveness of some ophthalmic medications is the redness of the eye. It sounds simple, until it’s a criterion in a nationwide research project using computers and electronic images. As anyone who’s ever ordered clothes online can tell you, one monitor’s red is another’s pink.
Researchers in the Perelman School of Medicine, including Carolyn F. Jones Professor of Ophthalmology Maureen Maguire, William C. Frayer Professor of Ophthalmology Richard Stone, Assistant Professor of Ophthalmology Vatinee Bunya, and Associate Professor of Clinical Ophthalmology Mina Massaro-Giordano, were trying to solve this problem. They approached Professor of Psychology David Brainard, director of Penn’s Vision Research Center and an authority on color vision, to ask for his help with the color of the images.
“They were thinking about collecting data from patients at multiple sites using digital images,” Brainard says. “That then raises a problem of standardizing the image acquisition and the display so that you don’t add noise to the grading process [when the images are analyzed]. And so I said, yeah, I know a little bit about that.”
He developed protocols for calibrating the cameras used and the monitors for the reading centers where images would be evaluated. The group even tested to see if clothing or background color affected the image. They still were left with the problems of flash idiosyncrasies and different lighting in the rooms where the pictures were taken.
The solution ultimately was “kind of a simple idea that Richard Stone came up with,” says Brainard: Each patient held a white card next to his or her eyes. The processing would then correct each photo using the card as a baseline, so each image was now color-corrected consistently relative to all the others. A similar approach let them balance color uniformly. The group also developed a detailed protocol for the graders to use as the displayed images were evaluated, to improve consistency there as well.
The group has written a paper on the protocol, and is now using the method in a research study to examine a medication which produces bloodshot eyes as a possible side effect. A researcher at Duke wants to use it to measure skin redness related to hormonal changes in humans; “I shipped her the software and instruction manual,” says Brainard.
“The best part of this is that the group has carried on, and we’re working on a couple of different things,” he says. Most interesting to him is the question of whether it’s possible, now that they have standardized images, to extract numbers that characterize the state of redness of some tissue. Even with the protocols and scaling, human readers sometimes grade the same image differently at different times. Some variances are hard to see, and images frequently also have several small areas of differing redness. Converting the colors to numbers and extracting numbers from every region would provide an objective measurement.
Researchers in the future, then, instead of looking at pictures of eyes, might be looking at the data those images reveal. Brainard says, “It’s speculation, but to the extent that we could attach objective measures that correlate with the diseases being studied, it would make this part of medical research more quantitative.”
