<![CDATA[Diabetic Foot Ulcer (DFU) is a serious complication of diabetes that may lead to amputation if not diagnosed early. Manual visual identification of DFU is challenging due to the heterogeneous characteristics of ulcer shape and texture. This study proposes an automatic DFU classification approach based on transfer learning using four Convolutional Neural Network (CNN) architectures: ConvNeXt-Tiny, DenseNet201, MobileNetV2, and ResNet50. The dataset consists of 1,055 original images (512 ulcer and 543 normal) used as testing data, while the training data were generated through augmentation techniques including rotation, horizontal flip, vertical flip, contrast adjustment, and brightness enhancement, resulting in 5,275 training images. All models were trained with 224×224×3 image input, using the SGD optimizer with a learning rate of 0.0001 for 30 epochs and a batch size of 2. Model performance was evaluated using accuracy, precision, and recall metrics. Experimental results indicate that ResNet50 achieved the best performance with 99.60% accuracy, 99.33% precision, and 99.88% recall. ConvNeXt-Tiny and DenseNet201 also demonstrated competitive performance with accuracy above 98%, while MobileNetV2 achieved 92.85% accuracy. Comparative analysis with previous studies shows that the proposed approach achieves competitive results. The findings demonstrate that transfer learning combined with appropriate data augmentation can produce an accurate and reliable DFU identification system to support computer-aided diagnosis.]]>
Copyrights © 2026
