Nail Disease Classification Using Graph Attention Network (GAT) and Resnet

Nail diseases can serve as critical indicators of systemic health issues, making their early detection and classification essential for effective diagnosis and treatment. This study integrates deep learning and graph-based learning to classify nail diseases, utilizing ResNet for feature extraction and Graph Neural Networks (GNNs) for relational learning. A dataset comprising multiple nail disease categories, including Acral Lentiginous Melanoma, Onychogryphosis, Clubbing, Pitting, Blue Finger, and Healthy Nails, was utilized. The ResNet model extracts meaningful feature representations, which are then structured as a graph to capture inter-class relationships using a Graph Attention Network (GAT). Cosine similarity is employed to construct the graph edges to improve connectivity between samples, ensuring that nodes with high feature similarity are more likely to be connected. This approach enhances learning by leveraging relationships between visually similar nail disease patterns. Experimental results demonstrate high classification performance, achieving an accuracy of 0.8791 (88%). This research highlights the effectiveness of combining deep learning with graph-based learning for automated nail disease classification, paving the way for more robust AI-assisted diagnostic tools in dermatology.