Weakly-supervised detection of AMD-related lesions in color fundus images using explainable deep learning

•A novel explainable approach for the joint identification of age-related macular degeneration and its associated lesions.•Custom architectural setting that directly links the lesions and the diagnosis and enables the weakly-supervised coarse segmentation of lesions.•The diagnosis can be explained from the segmentation maps of the lesions and their identification.•The proposed approach only requires image-level labels commonly easy to obtain.•The proposed approach provides meaningful information that explains and complements the diagnosis and is of particular interest to clinicians for the diagnostic process.