Figure 4.

Deep learning attention maps overlaid on fundus autofluorescence (FAF) images and color fundus photography (CFP) images. For each of the 3 representative eyes, the FAF images (left column) and CFP images (right column) are shown in the top row, with the corresponding attention maps overlaid in the bottom row. The heatmap scale for the attention maps also is shown: signal range from −1.00 (purple) to +1.00 (brown). A, (Top left) Original FAF image. Reticular pseudodrusen (RPD) are observed as ribbon-like patterns of round and oval hypoautofluorescent lesions with intervening areas of normal and increased autofluorescence in the following locations: (1) in an arcuate band across the superior macula (black arrows), extending across the vascular arcade to the area superior to the optic disc, and less prominently and affecting a smaller area, and (2) in the inferior macula (broken black arrows), extending across the vascular arcade to the area inferior to the optic disc. (Bottom left) Fundus autofluorescence image with deep learning attention map overlaid. The areas of highest signal correspond very well with the retinal locations observed to contain RPD in the original FAF image. The superior arcuate band, extending to the area superior to the optic disc, is demonstrated clearly (black arrows), as are the 2 inferior locations (broken black arrows). The predominance of superior over inferior macular involvement also is captured in the attention map. (Top right) Original CFP image. Reticular pseudodrusen are not observed in most retinal locations known to be affected from the corresponding FAF image. However, an area of possible involvement may be present on the CFP image at the superonasal macula (broken gray arrow). (Bottom right) Color fundus photography image with deep learning attention map overlaid. The areas of highest signal correspond very well with both (1) the retinal locations observed to contain RPD in the original FAF image (although the deep learning model never received the FAF image as input) and (2) the areas of highest signal in the FAF image attention map. Hence, the CFP attention map has areas of high signal in both (1) the area (superonasal macula) with RPD possibly visible to humans on the CFP image (broken gray arrow) and (2) other areas where RPD appear invisible on the CFP image, but visible on the corresponding FAF image. B, (Top left) Original FAF image. Reticular pseudodrusen are observed clearly as widespread ribbon-like patterns of round and oval hypoautofluorescent lesions with intervening areas of normal and increased autofluorescence. The area affected is large, affecting almost the entire macula, but sparing the central macula, that is, in a doughnut configuration. (Bottom left) Fundus autofluorescence image with deep learning attention map overlaid. The doughnut configuration is captured well on the attention map, that is, the areas of highest signal correspond very well with the retinal locations observed to contain RPD in the FAF image. (Top right) Original CFP image. Reticular pseudodrusen are observed in some, but not all, of the retinal locations known from the corresponding FAF image to contain RPD: the superior peripheral macula seems to contain RPD (broken gray arrow), but they are not visible clearly in the inferior peripheral macula. (Bottom right) Color fundus photography image with deep learning attention map overlaid. The CFP attention map has areas of high signal in both (1) the area (superior macula) with RPD visible to humans on the CFP image (broken gray arrow), and (2) other areas (inferior macula) where RPD seem invisible on the CFP image but visible on the corresponding FAF image (although the deep learning model never received the FAF image as input). C, (Top left) Original FAF image. Reticular pseudodrusen are observed as ribbon-like patterns of round and oval hypoautofluorescent lesions with intervening areas of normal and increased autofluorescence. These are relatively widespread across the macula (black arrows; corresponding almost to a doughnut configuration), as well as the area superior to the optic disc (black arrow), but are less apparent in the inferior and inferotemporal macula (broken black arrows). (Bottom left) Fundus autofluorescence image with deep learning attention map overlaid. The areas of highest signal correspond very well with the retinal locations observed to contain RPD in the original FAF image (black arrows). The doughnut configuration, with partial sparing of the inferotemporal macula (broken black arrows), complete sparing of the central macula (affected by geographic atrophy), and additional involvement of the area superior to the optic disc, is captured well. (Top right) Original CFP image. Reticular pseudodrusen are not observed in most retinal locations known from the corresponding FAF image to be affected. However, an area of possible involvement may be present on the CFP image at the superotemporal macula (broken gray arrow). (Bottom right) Color fundus photography image with deep learning attention map overlaid. The CFP attention map shows areas of high signal in both (1) the area (superotemporal macula) with RPD potentially visible to humans on the CFP image (broken gray arrow), and (2) another area (inferior to the optic disc) that potentially may contain additional signatures of RPD presence.