Classification of Diabetic Foot Ulcer Classification using DenseNet201 Compared with ResNet152 to Improve Accuracy

Diabetes foot ulcers (DFUs) are a serious side effect of the disease that, if left untreated, can result in infections and even amputation. This research compares the performance of ResNet152 and DenseNet201 algorithms to ascertain which model is the most efficient for accurate DFU identification. A dataset from Kaggle containing 512 abnormal and 543 normal images was used for training and testing. Images were resized to $\mathbf{1 5 0} \times \mathbf{1 5 0}$ pixels and underwent data augmentation to improve model generalization. ResNet152 and DenseNet201 were trained with batch sizes of 32 and 50 epochs using TensorFlow and Keras. The models' evaluations were conducted utilizing specificity, F1-score, accuracy, sensitivity, and precision. Statistical analysis, including the test of Levene and the twosample T-test, was performed using IBM SPSS v27. DenseNet201 achieved a higher accuracy (99.06%) than ResNet152 $(92.45 \%)$. It also demonstrated superior sensitivity, specificity, and F1-score, with a lower misclassification rate. Statistical analysis showed there is no discernible difference between the models ($p>0.05$). DenseNet201 outperformed ResNet152 in identifying DFUs, indicating its suitability for automated DFU detection, which could aid early diagnosis and treatment.