A 3D Spatial Color Space Transformation of Automated Glaucoma Detection

Glaucoma is a leading cause of irreversible blindness worldwide, and early detection is crucial for preventing progressive vision loss. Addressing this problem requires feature representations that can capture both the chromatic and structural alterations in the retina associated with glaucomatous damage. This study presents an automated glaucoma detection framework based on three-dimensional (3D) spatial color space transformation using the FIVES public dataset. The proposed system integrates image preprocessing, 3D reconstruction, and statistical analysis of multiple color spaces, including red, green, blue (RGB), hue, saturation, value (HSV), luma, blue chrominance, red chrominance (YCbCr), and luminance, a*, b* (Lab), to enhance the discriminative representation of the retinal structures. The intensity, texture and morphology-based features were computed and combined to capture both spatial and chromatic variations relevant to glaucomatous changes. Analysis of Variance (ANOVA) was applied prior to classification to reduce feature redundancy. Several machine-learning classifiers were evaluated to determine the most effective diagnostic performance. Experiments show that the combination of Lab and YCbCr with a nonlinear classifier provides the best performance, achieving an average accuracy of 0.87, sensitivity of 0.98, and specificity 0.92. Random Forest and k-nearest neighbor (k = 5) models consistently outperformed the parametric linear model across all four tested color spaces. These findings indicate that separating luminance and chrominance components and modeling nonlinear relationships between color and morphological features are crucial for improving the reliability of glaucoma detection based on fundus images, making the combination of RF and kNN a promising baseline configuration for automated glaucoma screening systems.