<![CDATA[Diabetes mellitus is a chronic non-communicable disease with a continuously increasing global prevalence. The number of adults living with diabetes worldwide has reached approximately 589 million, with 252 million remaining undiagnosed. One of the most serious complications is Diabetic Foot Ulcer (DFU). This study developed SmartScan-DFU, an early detection system for diabetic foot ulcers based on deep learning, by comparing four model architectures: Custom CNN, EfficientNet-B3, ResNet-18, and ResNet-50. The dataset consists of 4,446 images obtained from the Roboflow Universe platform. Evaluation results show that ResNet-50 achieved the best performance with an accuracy of 87.19%, precision of 0.87, recall of 0.87, and F1-score of 0.87. This model outperformed ResNet-18 (81.22%), EfficientNet-B3 (72.44%), and Custom CNN (61.00%). The comparison indicates that more advanced CNN architectures, particularly ResNet-50, demonstrate superior spatial feature extraction and generalization capabilities for DFU image variations. The best-performing model was then integrated into a Flask-based web interface, enabling automatic, fast, and accurate classification of DFU images. This system is expected to assist in the early digital diagnosis of diabetic foot ulcers, accelerate clinical decision-making, and contribute to achieving the Sustainable Development Goals (SDG) point 3 on good health and well-being.]]>
Copyrights © 2026
