A new automated method for grading retinal tessellation in myopic eyes using fundus images

Abstract

The tessellation of the retina is of clinical importance, especially in the study of myopia. This paper presents an automated grading algorithm for tessellation based on the calculation of three Tessellated Fundus Indices (TFIs). Our new algorithm utilizes the red (R), green (G), and blue (B) color components of the region of interest (ROI) surrounding the fovea in fundus images to determine the degree of tessellation, categorized into grades 0, 1, and 2. Excessive brightness in fundus images can result in overexposure, which in turn can introduce inaccuracies when calculating TFI values in the region of interest (ROI) using the red, green, and blue (R, G, B) components. Prior to calculating the TFIs, the method applies luminosity and contrast variation correction to the fundus images automatically. This correction process is achieved through a series of steps: first, applying row-wise and column-wise 1-dimensional low-pass filtering (1DLF); then, computing the luminosity surface by subtracting the smoothed image from the original grayscale image. To maintain luminosity consistency, the original image channels are equalized using the luminosity surface, followed by histogram stretching for enhanced contrast. Finally, the algorithm computes B/R (Blue/Red) and G/R (Green/Red) ratios for each pixel in the original image and multiplies them by the red channel of the contrast-stretched image. The proposed algorithm was evaluated on a dataset of 60 fundus images from varying degrees of myopia, demonstrating its effectiveness in grading tessellation accurately. The automated approach streamlines the grading process, offering potential benefits in clinical settings and facilitating large-scale screenings for myopic eyes.