Rendimiento de un algoritmo para el diagnóstico de glaucoma basado en la relación excavación/disco del nervio óptico

Abstract

Glaucoma screening algorithms determine the presence or absence of disease based on the cup/disc ratio (Armaly’s classification). However, findings suggestive of optic nerve damage may be missed and misclassified as healthy in affected eyes. Spaeth’s Disc Damage Likelihood Scale helps to identify early changes of glaucomatous optic neuropathy. Objective To determine the performance of an algorithm for glaucoma screening based on Armaly’s classification and Spaeth’s Disc Damage Likelihood Scale by analyzing optic nerve images. Methods Cross-sectional diagnostic test study. An algorithm was designed to identify findings suggestive of optic nerve damage and not only the presence or absence of glaucoma. Optic nerve photos were classified using Armaly’s classification and Spaeth’s scale. The algorithm segments the optic nerve and cup by analyzing hue, saturation, and lighting values, then extracts contours using Otsu’s method with multiple thresholds. Each contour is represented using Fourier series. All the information feeds a two-layer neural network. Results The agreement of the algorithm with the specialist's criteria was 80% (p less than 0.05) for determining the cup/disc ratio, 91% (p less than 0.05) for Spaeth’s Disc Damage Likelihood Scale (DDLS), 92% (p less than 0.05) Spaeth’s modified DDLS and 99% (p less than 0.05) for the glaucoma damage classification. The agreement between the two specialists was 94% (p less than 0.05) for the glaucoma damage classification, 40% for the excavation/disc ratio, 12,4% for Spaeth’s DDLS and 12.5% for Spaeth’s modified DDLS. Conclusions We propose an algorithm based on a 2-layer neural network that in preliminary results achieves high accuracy to estimate glaucoma risk based on Armaly’s classification, Spaeth’ original and modified DDLS scales and glaucoma damage classification.