Diabetic ocular disorders are among the most important public health concerns, affecting millions of people with severe degrees of visual loss worldwide. Mild, non-blinding conditions and severe complications, such as diabetic retinopathy, diabetic macular edema, glaucoma, and cataracts, often appear due to uncontrolled diabetes. Presently, more than 145 million people worldwide are diagnosed with the raging global diabetes epidemic, and this number is estimated to increase further. Thus, early detection and prompt intervention play an important role in preventing vision damage. Earlier diagnostic systems faced a major drawback, as the framework was unable to handle different types of data. This restriction limited their capability in systematic multiple eye disease [MED] detection, particularly in clinical settings where diagnosis is a procedural task and where data is multisourced, such as clinical records, fundus-captured photos, and OCT exam information. This review discusses the development of a single AI framework that can handle structural and unstructured datasets to identify different stages of MED. It is termed as MED as it deals with multiple diseases such as diabetic retinopathy, diabetic macular edema, glaucoma, and cataracts. This survey also describes the developmental stages of segmentation techniques and also highlights the advanced techniques such as U-Net and SegFormer, which can be effectively used for anatomical segmentation, which is used in disease identification for both optical coherence tomography (OCT) and fundus datasets. Furthermore, it helps in understanding the workflow for developing an effective MED framework.