The Effect of Fundus Image Quality on Detecting Clinically Significant Features in Fundus Imaging

Abstract

Abstract: : Purpose: To demonstrate a means for improving image quality in standard fundus imaging systems through better utilization of illumination spectral characteristics. A cursory assessment of digital retinal images, or digitized 35mm color slides will show the serious imbalance of reflected light from the typical human retina. Because of the highly blood perfuse nature of the retina and the high absorption by the ocular media, a color retinal image is generally low in information content for the red and blue channels of a color (RGB) image. Methods: Subjects (N=20) were selected from images collected imaged with a 12–bit camera mounted on a Zeiss FF450 fundus imager. Analytical and hardware changes to a fundus imager were made to create a high entropy camera. The cohort was imaged using three levels of illumination in order to evaluate statistically the relative illumination required to impinge on the retina for optimum image quality (IQ). IQ was established by two means. In a blind test, the 20 three–image sets were compared and qualitatively evaluated based on the visual criteria, such as sharpness and detectability of lesions. Independent quantitative parameters were correlated to the human visual preferences. Results: It was determined that by reducing the illumination from the red part of the spectrum and increasing the blue, one can improve the image quality significantly, as judged by ophthalmologists. The qualitative finding is supported by statistics from the quantitative metrics that show an average increase in entropy, contrast, and spatial frequency. A model that is based on these quantitative metrics was developed that predicts visual image quality. Conclusions: This project demonstrated an off–the–shelf technology for making significant improvements to fundus image quality. Additionally, with algorithms to remove lighting artifacts, the image quality was shown to be further improved.

Keywords: imaging/image analysis: clinical • image processing