This research is centered on the development of an AI-based diagnostic model for central-involved diabetic macular edema (ci-DME) using ultra-widefield color fundus photography (UWF-CFP). The model’s training and validation datasets were constructed from data collected under the EviRed project framework at 14 hospitals in France, supplemented with independent data from the LAZOUNI Ophthalmic Clinic in Tlemcen, Algeria. The development process entailed the construction of a comprehensive and robust dataset, coupled with the implementation of advanced optimization techniques. The results of this study indicate that both EfficientNet-B7 and ResNet152, with specially designed training strategies are effective models for the diagnosis of ci-DME. EfficientNet-B7 demonstrated superior overall discrimination ability, while ResNet152 showcased strong classification performance and reliable confidence predictions. These findings highlight the significant potential of AI in advancing the diagnosis and treatment of ci-DME, potentially facilitating earlier interventions and resulting in improved patient outcomes.