Fig. 3From: Automated detection of myopic maculopathy from color fundus photographs using deep convolutional neural networksTypical examples of true positive, false negative and false positive images and the corresponding CAM heatmaps. a Is a true positive of TF, the heatmap predominantly visualizes TF region. b, c Are true positives of PM respectively, the corresponding heatmaps highlight the atrophy lesions. In the row below, d is a false negative of PM which is recognized as TF by the DCNN-DS model, e, f are false positives of TF and PM respectively caused by other macular pathologies. The corresponding heatmaps also visualize the major interested regions of the DCNN-DS model. CAM, class activation map; TF, tessellated fundus; PM, pathologic myopia; DCNN-DS, dual-stream deep convolutional neural networkBack to article page