Tailored Deep Learning Approaches for Binary Classification and Evaluation of Diabetic Foot Ulcer Images

The frequent issue of blood glucose level that causes Diabetic Foot Ulcer (DFU) which is not properly controlled and diagnosed that leads to gangrene and amputation. To identify the early detection of diabetic foot ulcer promptly, we used deep learning approaches such as DenseNet121, CNN, VGG16, VGG19 and ResNet50 for the classification of normal and abnormal foot images. Evaluating various algorithms functionality and choosing the best structure to obtain precise DFU recognition. The dataset was preprocessed with pixel values being normalized and further rotation, shifts, zoom and various DL models were applied. To evaluate, the data set was classified into validation and training phases. Multiple convolutional, pooling layers, flatten, dense, dropout and ReLU architecture are used for binary classification of DFU pictures. Adam optimizer, Binary cross entropy and Loss Function were employed by the model. The experimental results demonstrated that Densenet121 achieved an outstanding accuracy of 99%. The model exhibits an impressive ability to differentiate between normal and abnormal DFU cases. It is evidenced by sensitivity-specificity calculations and the analysis of the confusion matrix. This study demonstrates the way deep-learning algorithms that categorize the DFU images and provide support models for healthcare professionals to accelerate the diagnosis of DFUs and to reduce the leg amputation and mortality rate.