Abstract:
Diabetic Foot Ulcer (DFU) is a complication of diabetes that is very severe and common and may result in amputation and infection in case it is not diagnosed in early stages. Early diagnosis plays a vital role in the effective treatment, and the manual clinical diagnosis can be time-consuming and subjective. The proposed study will create an automated DFU detection system based on machine learning algorithms and deep learning to help in diagnosing it accurately and quickly. The implemented and assessed CNN models include VGG16, ResNet50 and EfficientNetB0, which are used in DFU image classification. Normalization, resizing and data augmentation were used to pre-process a dataset of labeled foot images. Each model was fine-tuned using transfer learning and the performance of each model measured based on accuracy, precision, recall, and F1-score. The findings indicated that the three models all had good performance in the detection of the ulcerated and non-ulcerated foot images. Nevertheless, EfficientNetB0 demonstrated the best overall accuracy as compared to VGG16 and ResNet50 because of its smaller trade-off in network depth, width, and resolution. The suggested system proves to be highly promising to assist healthcare workers in the early DFU detection and elimination of diagnostic errors and better patient care outcomes. To sum up, the current study demonstrates that deep learning-based models are useful to automatize the detection of diabetic foot ulcers, which is a reliable, efficient, and scalable tool that can be incorporated into clinical and telemedicine to enhance the management of diabetes.
