**Background:** Type 2 diabetes mellitus (T2DM) is a common metabolic disease that leads to long-term hyperglycemia, damaging multiple organs including the eyes. Ocular complications (OCs) induced by T2DM are a major cause of blindness worldwide and include both retinopathy and non-retinopathy conditions such as corneal disease, dry eye disease, cataract, glaucoma, fundus disease, and optic neuropathy. Tears and blepharons (meibomian glands) are essential for maintaining ocular surface health through lubrication, cleaning, nutrition, and defense. Deep learning (DL) has shown promise in medical image analysis and clinical decision support. This narrative review aims to summarize the pathogenesis and treatment of T2DM-induced OCs, analyze the correlation between OCs and tear/blepharon function, and explore the application of DL in this field.
**Methods:** This is a narrative review that synthesizes existing literature on T2DM-related ocular complications, tear and blepharon function, and DL applications. The review covers diabetic retinopathy (DR), keratopathy, xerophthalmia, glaucoma, and cataract, discussing their pathogenesis and treatment. It also reviews studies on tear secretion changes in diabetes, meibomian gland dysfunction (MGD), and DL-based diagnostic approaches.
**Key Results:** The review identifies several key findings: (1) DR pathogenesis involves hyperglycemia-induced endothelial damage, increased vascular permeability, oxidative stress, and release of vascular proliferation factors such as VEGF. Risk factors include poor blood glucose control, hypertension, diabetes duration, dyslipidemia, and microalbuminuria. (2) Keratopathy results from hyperglycemia causing metabolic disorders of corneal epithelial cells, accumulation of advanced glycation end products (AGEs), hypoxia, and immune-inflammatory responses. The KCNQ1OT1/miR-214/caspase-1 signaling pathway is a newly identified mechanism. (3) Xerophthalmia is linked to reduced tear secretion and tear film instability, with studies reporting inconsistent findings: Andersen et al. found no significant difference in basic tear secretion between diabetes patients and controls, while Yeung and Dwarakanathan reported that 47% of diabetes patients had decreased tear secretion. Patnaik et al. found no obvious abnormality in tear film function. (4) Glaucoma pathogenesis involves increased intraocular pressure due to aqueous humor imbalance, leading to optic nerve damage. Diabetes may contribute through increased intraocular pressure, vascular lesions, or direct optic nerve damage. (5) Cataract is associated with oxidative stress, accumulation of glycation end products in the lens, and apoptosis of lens cells. Age and blood sugar control are consistently correlated with T2DM-induced cataract. (6) Tear function changes in diabetes are related to peripheral neuropathy, abnormal glucose metabolism, neurotransmitters, and extracellular matrix components. Herber et al. found differences in tear protein composition, and Kalló et al. reported differences in tear protein content and saturation. Nokhoijav et al. found significantly increased Apo(a-I) in tears of T2DM patients, closely related to DR severity. Kuo et al. observed uneven tear film lipid layer distribution and decreased tear film rupture time and corneal sensation. (7) Meibomian gland dysfunction (MGD) is a major cause of evaporative dry eye, with up to 86% of dry eye syndrome patients showing signs of MGD. Diabetes is a risk factor for asymptomatic MGD. Fan et al. reported significantly worse meibomian gland changes in diabetes patients, including higher meibomian gland abscission rate, lower number of expressible glands, and higher meibomian margin abnormality score. The loss rate of upper and lower meibomian glands in T2DM patients is significantly higher than in non-diabetes patients. (8) DL algorithms, such as convolutional neural networks (CNNs) and U-Net, have been applied to automatically diagnose and classify DR, predict glaucoma onset and progression, segment retinal microvessels, and assess cataract surgical risk. Tan et al. found DL effective for risk stratification of myopic macular degeneration. Li et al. developed a DL system for predicting glaucoma onset and progression with high sensitivity and specificity. Zekavat et al. used CNN to measure retinal vascular fractal dimensions and density, finding low values significantly associated with higher risk of mortality, hypertension, congestive heart failure, renal failure, T2DM, sleep apnea, anemia, and multiple ocular diseases.
**Clinical Implications:** This review underscores the importance of comprehensive management of T2DM to prevent or delay ocular complications. Blood sugar control remains the cornerstone of treatment for all OCs. For DR, treatments include anti-VEGF drugs, laser therapy, and surgery. Keratopathy management involves blood sugar control, artificial tears, repairing additives, anti-inflammatory drugs, and optical treatments. Xerophthalmia treatment includes artificial tears, environmental modifications, medications, and surgical options such as lacrimal gland ostomy or transplantation. Glaucoma is managed with intraocular pressure-lowering drugs, laser therapy, and surgery. Cataract treatment primarily involves surgery with artificial lens replacement. The review highlights the critical role of tear and meibomian gland function in ocular health, suggesting that monitoring tear film and meibomian gland changes could aid early detection and management of OCs in T2DM patients. DL-based tools offer potential for automated screening, diagnosis, and risk prediction, improving efficiency and accuracy in clinical practice. However, challenges such as data demand, opacity, generalization ability, data bias, privacy issues, and sample imbalance need to be addressed for reliable clinical implementation.