Technical Issues in Retinopathy Screening

The article by Gómez-Ulla et al. (1) in the August issue of Diabetes Care presents a strong case for a simplified method of detecting people at risk for vision loss from diabetic retinopathy. Their work reports that the use of a two-field imaging protocol for each eye, telemedicine and centralized analysis, achieve diagnostic capabilities comparable to physician-provided ophthalmoscopy. This increases cost-effectiveness and allows retinopathy screening to meet the key criteria of a population-based screening program (2).

However, this report raises an important question regarding image quality. The authors suggest that JPEG compression permits rapid data transmission, requires less server space, and preserves “adequate definition of the lesions.” However, they admit underestimation of retinal neovascularization from image analysis, since “high-grade DR [diabetic retinopathy] has image features that may be lost when inspection of the digital images is made.” Yet, proliferative disease is precisely the most critical clinical situation in which patients may benefit from immediate laser therapy, thus preventing a predictable pattern of events that results in permanent vision loss.

We believe that the inability to detect crucial pathology is caused by both the original image acquisition (low-resolution video camera used along with a frame grabber, which further reduces the available resolution) and by the JPEG compression algorithm.

Image-based screening systems rely on visible surrogate markers, lesions that denote alterations of the retinal microcirculation. When these markers are centrally located and unobscured by overlying media opacities (and assuming they are large enough to be rendered by several overlying pixels to provide morphologic clues), observers should be able to detect them. Problems arise when they are located toward the image periphery (in which case optical aberrations may obscure their true shape), when lenticular changes or vitreous turbidity mask them, or when not enough pixels are used to render them. Clearly, if the original image was barely adequate to portray unambiguous lesions, then those markers that are more difficult to visualize would be under-reported.

The proposed screening programs in England (3) and Scotland (4) specify that the original image exhibit a minimum pixel matrix of 1,300 × 1,000. This enhances the detection of retinopathy when the image is displayed on a large CRT monitor at full resolution. In the U.S., Canon Medical Products and Digital Healthcare offer a high-resolution pure digital system, based on the Canon EOS D30 or D60 digital camera and CR-6 45NM fundus camera, that exceeds the U.K. standards. We have been using this system for more than a year and find that store-and-forward data transmission using digital subscriber lines is a practical means of connecting the camera with the reading center.

We hope that publication of the work of Gomez-Ulla et al. signals that the American Diabetes Association is willing to revisit their position statement on diabetic retinopathy (5), first published 4 years ago. With recent advances in digital technology allowing even greater sensitivity, high-resolution digital screening of high-risk patients, using a simplified protocol, offers a practical method of reducing the risk of vision loss in diabetes.