Therapeutic modalities targeting vascular endothelial growth factor (VEGF) have been used to treat neovascularization and macular edema. However, anti-VEGF treatment alone may cause up-regulation of connective tissue growth factor (CTGF) in the retina, increasing the risk of fibrosis and tractional retinal detachment. Therefore, in this study, we employ a novel dual-target intervention that involves intravitreal injection of the VEGF inhibitor ranibizumab and a transfection reagent-treated non-viral vector carrying anti-CTGF short hairpin RNA (shRNA) driven by human RNA polymerase III promoter U6. The effects of the dual-target intervention on the expression of VEGF and CTGF and on microvessel ultrastructure were examined in retina of streptozocin-induced diabetic rats. CTGF was significantly up-regulated at week 8 after diabetic induction, whereas VEGF was not up-regulated until week 10. The high expression of both genes was maintained at week 12. Transmission electron microscopy also revealed progressive exacerbation of microvessel ultrastructure during the same period. In addition, ranibizumab significantly lowered VEGF but elevated CTGF mRNA, whereas CTGF shRNA significantly reduced the mRNA levels of both CTGF and VEGF in diabetic retinas. Importantly, dual-target intervention normalized the transcript levels of both target genes and ameliorated retinal microvessel ultrastructural damage better than either single-target intervention. These results suggest the advantages of dual-target over single-target interventions in diabetic retina and reveal a novel therapeutic modality for diabetic retinopathy.

Microglia cells (MGCs) play a key role in scavenging pathogens and phagocytosing cellular debris in the central nervous system and retina. Their activation, however, contributes to the progression of multiple degenerative diseases. Given the potential damage created by MGCs, it is important to better understand their mechanism of activation. Here, we explored the role of MGCs in the context of retinitis pigmentosa (RP) by using four independent preclinical mouse models. For therapeutic modeling, tamoxifen-inducible CreER was introduced to explore changes in MGCs when RP progression halted. The phenotypes of the MGCs were observed using live optical coherence tomography, live autofluorescence, and immunohistochemistry. We found that, regardless of genetic background, MGCs were activated in neurodegenerative conditions and migrated beyond the layers where they are typically found to the inner and outer segments, where degeneration was ongoing. Genetic rescue not only halted degeneration but also deactivated MGCs, regardless of whether the intervention occurred at the early, middle, or late stage of the disease. These findings suggest that halting long-term disease progression may be more successful by downregulating MGC activity while co-administering the therapeutic intervention.

Hypoxia associated with the high metabolic demand of rods has been implicated in the pathology of age-related macular degeneration (AMD), the most common cause of adult blindness in the developed world. The majority of AMD-associated severe vision loss cases are due to exudative AMD, characterized by neovascularization. To further investigate the causes and histopathology of exudative AMD, we conditionally induced hypoxia in a novel preclinical AMD model (Pde6g^creERT2/+;Vhl^−/−) by targeting Vhl and used multimodal imaging and immunohistochemistry to track the development of hypoxia-induced neovascularization. In addition to developing a preclinical model that phenocopies exudative AMD, our studies revealed that the photoreceptor hypoxic response initiates and drives type 3 neovascularization, mainly in the outer retina. Activation of the VHL-HIF1a-VEGF-EPO pathway in the adult retina led to long-term neovascularization, retinal hemorrhages and compromised retinal layers. Our novel preclinical model would accelerate the testing of therapies that use metabolomic approaches to ameliorate AMD.

Objective: This study constructs a deep learning-based combined algorithm named WaveAttention ResNet (WARN) to investigate the classification accuracy for seven common retinal diseases and the feasibility of AI-assisted diagnosis in this field.Methods: First, a deep learning-based classification network is constructed. The network is built upon ResNet18, integrated with the Convolutional Block Attention Module (CBAM) and wavelet convolution modules, forming the WARN method for retinal disease classification. Second, the public OCTDL dataset is used to train WARN, which contains classification data for seven retinal disease types: age-related macular degeneration (AMD), diabetic macular edema (DME), epiretinal membrane (ERM), normal (NO), retinal artery occlusion (RAO), retinal vein occlusion (RVO), and vitreomacular interface disease (VID). During this process, ablation experiments and significance tests are conducted on WARN, and comprehensive analyses of various indicators for WARN, ResNet-18, ResNet-50, Swin Transformer v2, EfficientNet, and Vision Transformer (ViT) are performed in retinal disease classification tasks. Finally, data provided by Shanxi Eye Hospital are used for testing, and classification results are analyzed.Results: WARN demonstrates excellent performance on the public OCTDL dataset. Ablation experiments and significance tests confirm the effectiveness of WARN, achieving an accuracy of 90.68%, F1-score of 91.29%, AUC of 97.50%, precision of 93.31%, and recall of 90.68% with relatively short training time. In the dataset from Shanxi Eye Hospital, WARN also performs well, with a recall of 90.85%, precision of 79.94%, and accuracy of 89.18%.Conclusion: This study fully confirms that the constructed WARN is efficient and feasible for classifying seven common retinal diseases. It further highlights the enormous potential and broad application prospects of AI technology in the field of auxiliary medical diagnosis.

Objective: Currently, publicly available Optical Coherence Tomography (OCT) datasets are commonly plagued by limited coverage of disease categories, scarce samples and severe class imbalance, which leads to insufficient generalization ability of deep learning models in real-world clinical settings. This study aims to construct a high-quality OCT dataset encompassing six key types of fundus lesions and normal controls, and to systematically evaluate the improvement effect of training strategies for multi-source data fusion on the performance of multi-class classification. Methods: We integrated local clinical data from Shanxi Eye Hospital with the latest public dataset OCTDL to establish a combined dataset. This dataset consists of 6,165 images, covering seven categories: age-related macular degeneration (AMD), diabetic macular edema (DME), retinal artery occlusion (RAO), retinal vein occlusion (RVO), epiretinal membrane (ERM), vitreomacular interface disease (VID), and normal controls (NO). On this basis, six representative deep learning architectures were selected, and two training paradigms were compared under unified experimental settings: (1) Training exclusively on open-source OCTDL data (S1); (2) Joint training using both local data and OCTDL data (S2). All models were evaluated on the identical OCTDL test set. A comprehensive analysis was conducted using multi-dimensional metrics including accuracy, weighted F1-score, class-specific recall, and area under the curve (AUC), with a particular focus on the misdiagnosis rate. Results: The S1 strategy exhibited significantly limited model recognition capability due to the extremely small sample sizes of certain categories. In contrast, the S2 strategy markedly improved the overall performance of the models. Confusion matrix analysis demonstrated that ViT-Base achieved the optimal performance under the S2 strategy: the accuracy reached 93.61%, the misdiagnosis rate of RAO was reduced to 0%, the misdiagnosis rate of AMD was controlled at 1.34%, and the misdiagnosis rate of RVO decreased from 14.89 to 8.51%. Conclusion: Multi-source data fusion serves as an effective approach to enhance the robustness of OCT multi-category classification models, and it can notably strengthen the recognition capability for certain diseases in particular. This study not only verifies the universal benefits of this strategy but also reveals the critical impact of model selection on the transfer learning effect.