XAI-Driven Consumer OCT Analytics for Vascular–Immune–Neural Retinal Health Monitoring

Consumer-level optical coherence tomography (OCT) devices have the potential to revolutionize the way people monitor their retinal health through democratization; however, the current analytical systems do not provide sufficient interpretability and computational efficiency for home use. The presented paper proposes an explainable artificial intelligence (XAI) analytics platform, VINExplainNet (Vascular–Immune– Neural Explainable Network), that provides a comprehensive disease classification with attention-based visual explanations of OCT images. The VINExplainNet platform combines a lightweight multitask deep network and hierarchical attention mechanisms to produce clinically relevant visual explanations with diagnostic predictions. The architecture contains an innovative Hierarchical Attention Fusion Module (HAFM) that effectively captures multi-scale features of the retina while remaining computationally efficient using depthwise separable convolution and channel-wise pruning approaches. Evaluation on three publicly available benchmark datasets with 8,417 images (Kermany 1,000-test, Duke 3,231 images, and OCTID 572 images) performed under consumer imaging conditions demonstrates that this system achieves 98.47% accuracy on Kermany, 97.63% on Duke and 94.27% on OCTID, while maintaining an inference latency of less than 200ms on edge devices. Attention-based explanation maps are scored with a 0.912 faithfulness score and a 4.23/5.0 clinical alignment rating from 23 board-certified ophthalmologists. Ablation testing confirms that the HAFM module provides a 1.92% improvement in accuracy compared to non-HAFM baseline architectures. This work provides a solid scientific basis for using artificial intelligence in the area of consumer wellness to help people better manage retinal health. The framework bridges the gap between clinical quality diagnostics and everyday health-monitoring tools1.