Robust benchmarking of DeepLabv3 hybrid models for multiclass fundus screening and referable-disease triage on the FIVES dataset | Publicación