Figure 4 | Scientific Reports

Figure 4

From: Prediction of treatment outcome in neovascular age-related macular degeneration using a novel convolutional neural network

Figure 4

Representative horizontal scans of SD-OCT and corresponding superimposed heatmaps. Presented are (A) an example of an OCT image with the HDF-Net correctly predicted as an improved case (85 M; baseline VA:0.2; 12th month VA: 0.5); (B) The network located attention on the margin of SRF, and on retinal pigment epithelial mottling. Also, some attention corresponded to islands of preserved ellipsoid zone (EZ) reflecting preserved visual potential. (C) An example of an OCT image with the HDF-Net correctly predicted as an unimproved case (94 M; baseline VA:0.05; 12th month VA: 0.1); (D) The attention is located at the subfoveal disciform scar and disruption of the EZ. There is also some attention on the chorioretinal atrophy with adjacent loss of outer retinal layers. (E) An example of an OCT image with the HDF-Net predicted to improve but the patient failed to achieve VA improvement >  = 2 lines (96 M; baseline VA:0.4; 12th month VA: 0.4); (F) The attention is located primarily on the preserved EZ and the subretinal hyperreflective material (SHRM) with fibrovascular components. A relatively normal fovea contour was also identified. (G) An example of an OCT image with the HDF-Net predicted not to improve but the patient demonstrated VA improvement >  = 2 lines (82 M; baseline VA:0.4; 12th month VA: 0.6); (H) The attention is located on the SHRM with fibrovascular/hemorrhagic components. Also some attention on the disruption of the EZ and the hyperreflective zone of the inner retina.

Back to article page