Glaucoma Diagnosis with Machine Learning Based on Optical Coherence Tomography and Color Fundus Images

Figure 5

Visualization of the important areas in our VGG19 model. One healthy eye and two representative glaucoma eyes were randomly selected to show the area of interest in the input images. Results from a healthy subject (a, b, c) and glaucoma subject 1 (d, e, f) and 2 (g, h, i), showing class-discriminative regions in grayscale disc fundus images, disc RNFL thickness maps, and macula GCC thickness maps, respectively. Dark orange regions correspond to high scores for the diagnosis.