Identifikasi Penyakit pada Kuku Manusia Menggunakan Arsitektur Dynamic Feature Fusion Berdasarkan Graph Convolutional Network

Abstract

Nail disease is an important bio-indicator for detecting serious systemic diseases, but its diagnosis is often hampered by limited access to specialists and the subjectivity of visual observation. Early detection and accurate diagnosis are crucial to prevent further complications. Image processing technologies like Convolutional Neural Networks (CNN) have been widely used, but they have limitations in capturing complex visual features at various scales. This research aims to design and implement a high-accuracy nail disease identification system to assist in medical diagnosis. This study uses the Dynamic Feature Pyramid Network (DG-FPN) architecture, an enhancement of the Feature Pyramid Network (FPN) architecture, which has proven effective in image processing. This architecture integrates FPN to extract multi-scale features and Graph Convolutional Network (GCN) to model the relationships between features, thereby improving model efficiency and accuracy. The model was trained using a dataset from Kaggle and Dermnet containing ten types of nail disorders. The data underwent a pre-processing stage consisting of normalization, augmentation, and segmentation. After pre-processing, the data was fed into the DG-FPN architecture for feature extraction and model training. The results of using the DG-FPN architecture show an accuracy of 95.7%. Based on these results, it can be concluded that the developed system is very effective in identifying nail diseases through images.