A Hybrid Deep Learning Architecture for Automatic Identification of Diabetic Foot Ulcers (DFUs) from Clinical Thermal Images

Diabetic foot ulcer (DFU) leads in causation of most common and severe complication of diabetes. DFUs are sometimes diagnosed at later stages, which cause infections and amputations. In the world, 20% to 25% of people with diabetes get foot complications. The main cause is due to poor awareness and late treatment. To solve the problem diagnostic systems based on technology can be very helpful in early detection as well as preventive care. The study developed AI system for the early detection and classification of diabetic foot ulcers utilizing deep learning. The study examines the performance of DenseNet121, EfficientNetB0, ResNet50, VGG16, and a proposed hybrid model DenseNet+SVM which is the combination of Densenet121 with SVM. The model was trained and evaluated using a dataset of 1866 pair of Thermal foot images of healthy person and DFU case. According to the results of the experiments conducted, the hybrid model DenseNet121+SVM is capable of attaining an accuracy of 99.8% which outperforms DenseNet121 (98.2%), EfficientNetB0 (96.5%), ResNet50 (89.7%), and VGG16 (84.1%). The results reveal that hybrid deep learning systems have the promise to enhance the accuracy and reliability of DFU detection systems based on Wagner-Meggitt chart for DFU.