DFUGCN: Image Classifying Technique for Localizing of Diabetic Foot Ulcers

Diabetic foot ulcers (DFUs) represent a serious and common complication of diabetes, occurring in $\mathbf{1 5 - 2 5 \%}$ of the world's diabetic population and resulting in high morbidity, amputations, and healthcare expense through delayed diagnosis. The current study responds to the need for immediate automated diagnostic systems that are accurate and cost-effective to facilitate early treatment and alleviate healthcare burdens. We introduce DFUGCN (Diabetic Foot Ulcer Graph Convolutional Network), a deep learning approach that combines Convolutional Neural Networks (CNNs) with a three-layer Graph Convolutional Network (GCN) to encode local features and spatial interactions usually not captured by classical CNNs. DFUGCN utilizes super pixel segmentation and region adjacency graphs to improve interpretability and robustness across different imaging modalities and clinically relevant patterns, enhancing the accuracy and trustworthiness of DFU detection. Experimental tests show the model's excellent performance, such as test image predictions with evidence from visual images. Expansion of the dataset and clinical deployment in future work will continue to enhance diagnostic accuracy.