The medical and healthcare domains require automatic diagnosis systems (ADS) for the identification of health problems with technological advancements. Biomedical imaging is one of the techniques used in computer-aided diagnosis systems. Ophthalmologists examine fundus images (FI) to detect and classify stages of diabetic retinopathy (DR). DR is a chronic disease that appears in patients with long-term diabetes. Unattained patients can lead to severe conditions of DR, such as retinal eye detachments. Therefore, early detection and classification of DR are crucial to ward off advanced stages of DR and preserve the vision. Data diversity in an ensemble model refers to the use of multiple models trained on different subsets of data to improve the ensemble’s overall performance. In the context of an ensemble model based on a convolutional neural network (CNN) for diabetic retinopathy, this could involve training multiple CNNs on various subsets of retinal images, including images from different patients or those captured using distinct imaging techniques. By combining the predictions of these multiple models, the ensemble model can potentially make more accurate predictions than a single prediction. In this paper, an ensemble model (EM) of three CNN models is proposed for limited and imbalanced DR data using data diversity. Detecting the Class 1 stage of DR is important to control this fatal disease in time. CNN-based EM is incorporated to classify the five classes of DR while giving attention to the early stage, i.e., Class 1. Furthermore, data diversity is created by applying various augmentation and generation techniques with affine transformation. Compared to the single model and other existing work, the proposed EM has achieved better multi-class classification accuracy, precision, sensitivity, and specificity of 91.06%, 91.00%, 95.01%, and 98.38%, respectively.

Automated diabetic retinopathy (DR) grading from colour fundus photographs can achieve strong predictive performance, but clinical interpretation requires more than an image-level label. It requires understanding how lesion evidence is distributed around retinal vessels and how this evidence relates to quantitative vascular biomarkers. We present a dual-edge spatial-Jacobian image graph for interpretable DR grading. Each fundus image is represented as a graph node with four aligned evidence streams: AutoMorph vessel information ($X_1$), DR-XAI-style lesion evidence maps ($X_2$), a 128-dimensional lesion-based contrastive image embedding ($X_3$), and AutoMorph morphometric biomarkers ($X_4$). The spatial edge branch ($X_{12}$) encodes vessel-lesion geometry, while the Jacobian branch ($X_{34}$) models embedding-biomarker sensitivity. Lightweight two-token attention fuses both edge families into a final image graph. On 2,910 matched non-augmented APTOS images, the full graph achieves 0.8076 accuracy, 0.8312 quadratic weighted kappa, 0.5915 macro-F1, and 0.9330 adjacent-grade accuracy; referable DR reaches 0.9055 accuracy and 0.9711 AUROC. The framework is positioned as an explainable representation-learning tool for lesion-biomarker hypothesis generation, rather than as a deployment-ready clinical classifier. The code is available at https://github.com/Inamullah-Colab/dual-edge-dr-graph-xai.

Learning causal models from fragmented biomedical data is challenging because clinical, molecular, and imaging variables are often incomplete or not jointly observed. We propose RetiSEM, a domain-constrained structural equation modelling (SEM) framework for causal graph recovery and mediation analysis under limited multimodal resources. This proposed work organises variables into biologically informed blocks, applies forbidden-edge constraints, and decomposes pathway-level effects into TE, NDE, and NIE components. We evaluate RetiSEM across ten synthetic benchmark scenarios that vary in dimensionality, nonlinearity, causal depth, and pathway structure, together with a fragmented real-world setting that combines NHANES clinical variables with externally derived retinal representations. This approach achieves lower structural error and higher causal accuracy than unconstrained baselines across the synthetic benchmarks. In the real-data analysis, retinal variables behave mainly as downstream biomarker-like indicators, with smaller but detectable indirect effects. These findings support our strategy as an interpretable framework for testing structured causal hypotheses in limited-resource biomedical AI. The code and resources for this work are publicly available at: https://github.com/Inamullah-Colab/ReitSEM.

Current retinal computational pipelines are largely predictive rather than mechanistic. To address this, we introduce RetiSEM, a domain-constrained structural equation modelling framework that enforces biological directionality via forbidden-edge masking and estimates mediator-centric pathways through TE/NDE/NIE decomposition. Validated on synthetic benchmarks and real-world NHANES datasets, Re-tiSEM achieves superior structural recovery, demonstrating low SHD and high causal accuracy. In NHANES, our analysis reveals robust direct cardiometabolic effects but small mediated signals, confirming that these specific retinal traits function primarily as downstream biomarkers.