Diabetic Foot Ulcer Diagnosis Through Deep Learning Model

Diabetes mellitus is a huge detriment to world health as it is associated with some serious complications like diabetic foot ulcers (DFUs) which frequently lead to amputation. However, in order to manage DFUs effectively, one needs to understand factors such as infection and ischemia which will determine the treatment decisions. In 2019, there were 463 million people suffering from diabetes and this number is expected to rise up to 700 million by 2045. DFUs also pose great risks such as amputation particularly in poorer countries where treatments are quite expensive. Hence medicines must be cheap, people should visit hospitals often and observe good hygiene so as to avoid them. Visual inspections of DFUs by traditional means also have limitations especially in case of obesity or impaired sightedness. The newer advancements in deep learning like the application of convolutional neural networks (CNNs) gives prospects for enhancement of DFU detection and classification. On example is the increased use of CNNs in classifying DFUs showing an F1 score as high as 94.5%, among other methods that are capable of achieving high accuracy rates. Various CNN architectures associated with DFU diagnosis are explored in this paper through pre-trained models, data augmentation plus fine-tuning among others making them more effective. ResNet50 outperformed previous models with 99.51% accuracy for ischemia and 84.78<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> accuracy for infection, showing superior performance overall. It aims at identifying the most effective CNN model for DFU classification thus having more accurate and efficient diagnostic processes, and contributes towards betterising accuracy and specificity of diagnostic procedures largely relying on CNNs.