Ocular diseases are one of the main causes of irreversible disability in people in productive age. In 2020, approximately 18% of the worldwide population was estimated to suffer of diabetic retinopathy and diabetic macular edema, but, unfortunately, only half of these people were correctly diagnosed. On the other hand, in Colombia, the diabetic population (8% of the country’s total population) presents or has presented some ocular complication that has led to other associated costs and, in some cases, has caused vision limitation or blindness. Eye fundus images are the fastest and most economical source of ocular information that can provide a full clinical assessment of the retinal condition of patients. However, the number of ophthalmologists is insufficient and the clinical settings, as well as the attention of these experts, are limited to urban areas. Also, the analysis of said images by professionals requires extensive training, and even for experienced ones, it is a cumbersome and error-prone process. Deep learning methods have marked important breakthroughs in medical imaging due to outstanding performance in segmentation, detection, and disease classification tasks. This article presents SOPHIA, a deep learning-based system for ophthalmic image acquisition, transmission, intelligent analysis, and clinical decision support for the diagnosis of ocular diseases. The system is under active development in a project that brings together healthcare provider institutions, ophthalmology specialists, and computer scientists. Finally, the preliminary results in the automatic analysis of ocular images using deep learning are presented, as well as future work necessary for the implementation and validation of the system in Colombia.