.jpg)
Glaucoma represents one of the leading causes of irreversible blindness worldwide, and its early diagnosis remains a clinical challenge. Automated systems based on artificial intelligence have demonstrated high performance, but they often operate as black boxes that offer no transparency to specialists. In this context, GlaKG emerges as a biomarker-centered knowledge graph that integrates structural findings, clinically validated rules, and features from fundus images, enabling traceable reasoning for glaucoma risk classification and stratification. Each prediction is accompanied by an explicit chain of evidence linking biomarkers to activated rules, facilitating clinical audit and trust in the system.
GlaKG's architecture combines a ResNet50 model to extract image representations with a rule-based reasoning module, all orchestrated through a post-processing fusion framework that prevents information leakage. Results on a public dataset labeled via artificial intelligence achieve an F1 of 0.9953 in binary classification and 93% accuracy in four-level risk stratification. However, the authors warn that these metrics represent an upper bound, as they are highly correlated with biomarker annotations, underscoring the importance of clean and well-structured data.
This approach is particularly relevant for developing custom applications in the healthcare domain, where explainability is not a luxury but a regulatory and ethical requirement. Companies like Q2BSTUDIO, specialized in artificial intelligence for businesses, offer capabilities to design and implement similar knowledge graphs, integrating AWS and Azure cloud services to scale image processing and ensure clinical data cybersecurity. Furthermore, combining AI agents with rule-based reasoning systems enables the generation of auditable reports that facilitate decision-making.
Beyond the medical field, the principles of GlaKG can be applied to other sectors where predictive model transparency is critical, such as financial fraud detection or industrial predictive maintenance. Business intelligence tools, like Power BI, can consume the outputs of these graphs to visualize reasoning chains and provide interactive dashboards for analysis teams. Ultimately, the combination of custom software with explainable artificial intelligence techniques opens new avenues for building robust, reliable systems aligned with regulatory standards.
