Weierstrass scale space representation and composite dilated U-net based convolution for early glaucoma diagnosis

Abstract

Glaucoma is one of the common causes of blindness in the current world. Glaucoma is a blinding optic neuropathy characterized by the degeneration of retinal ganglion cells (RGCs). Accurate diagnosis and monitoring of glaucoma are challenging task through eye examinations and additional tests. To achieve accurate diagnosis of glaucoma with higher sensitivity and specificity, novel method called Weierstrass scale space representation and composite dilated U-net based convolution (WSSR-CDC) is introduced. At first, the Weierstrass transform scale space representation is employed to enhance image structures at various scales with higher accuracy of region of interest (ROI) detection using Euler’s identity. Next, CDC model is utilized with several layers. In input layer, preprocessed input images are taken as input. Fragment derivative are formulated for every preprocessed input. Log cosh dice loss function and optic cup to disc ratio are computed for segmented glaucoma detected results. With this, the accurate diagnosis of glaucoma is made with minimal error. The WSSR-CDC method was evaluated using the glaucoma fundus imaging dataset with several factors. The results show that the WSSR-CDC method outperforms conventional techniques, improving accuracy by 24% and sensitivity by 18%. It demonstrates promising results in fast, accurate, diagnosis of glaucoma.