Multi Modal Knowledge Graph Augmented Retrieval for Explainable Mycetoma Diagnosis

Mycetoma, a neglected tropical disease endemic to arid regions, requires accurate differentiation between bacterial (Actinomycetoma) and fungal (Eumycetoma) forms for appropriate treatment. While recent deep learning approaches achieve moderate classification accuracy from histopathological images, they lack explainability, which is a critical requirement for clinical adoption in resource limited settings. We present the first multi modal knowledge graph integrating visual, clinical, laboratory, geographic, and literature data for Mycetoma diagnosis. Our retrieval augmented generation framework combines InceptionV3 based deep learning predictions with knowledge graph based contextual reasoning to produce clinically grounded explanations. The system achieves 94.8% accuracy, a 6.3% improvement over CNN only approaches, while providing transparent, multi evidence diagnoses rated 4.7/5 by expert pathologists. By bridging the explainability gap in medical AI for neglected tropical diseases affecting vulnerable populations in resource constrained regions, our approach demonstrates how knowledge graphs can enhance both performance and clinical trust in automated diagnostic systems.