الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 109 |  |  |
| | | from | 109 | | |
Abstract
Diabetes mellitus is a common health problem where the body struggles to control the level of sugar in the blood. This is mainly because the body doesn’t make enough insulin or can’t use it properly. When blood sugar levels are not managed well, it can lead to different health issues. One serious problem is Diabetic Foot Ulcer (DFU), which is not always given enough attention. DFUs happen because of changes in the nerves and blood vessels in the feet and can easily get infected. If these ulcers are not treated quickly, they might lead to the need to amputate the limb. DFUs are a big concern because they put a lot of pressure on the healthcare system, especially in places where medical care is limited or very expensive. This affects not just the people with the ulcers but also the overall health system.
In the field of Artificial Intelligence (AI), which includes areas like Machine Learn- ing (ML) and Deep Learning (DL), there’s a growing interest in making smart systems that can think and learn like humans. These kinds of technology are used in many different areas, including healthcare. In healthcare, ML and DL are really important for working with and understanding medical images. They help in sorting and sep- arating different parts of these images. Deep Learning is a special kind of Machine Learning that uses complex systems called neural networks. These networks learn from a lot of data and can figure out important details in the data by themselves. This means people don’t have to do as much of the work in picking out these details.
In our study, we tackled the tricky task of spotting diabetic foot ulcers (DFUs) by diving into the world of medical image analysis. Our main goal was to get better at diagnosing all sorts of foot problems to help patients get the care they need faster and make the whole healthcare process more efficient. We put two high-tech tools to the test to see which one was the best at finding these ulcers from pictures of feet. First up were Support Vector Machines (SVMs), a kind of smart computer program that’s been trained to pick out important details in images using a method called
Wavelet Transform. We also used something called Principal Component Analysis to boil down all the complicated data to just the essentials. We tried three different SVM setups—Linear, Gaussian, and Polynomial—to match the challenge of the data. The Gaussian one really stood out, hitting an accuracy score of 88%, which means it was pretty good at its job. But then we tried a different approach, using a fancy method called deep learning with a model named ResNet18. We played around with how many times ResNet18 would look at the images (epochs) and how many images it would look at each time (batch sizes). After finding the sweet spot, ResNet18 blew us away with a 97.9% accuracy rate, proving it’s really good at making sense of the complex patterns in the images and spotting those DFUs.
Advancing our exploration, we applied a multi-class classification lens using a hy- brid model that integrated the prowess of the ResNet50 architecture with the creative potential of Generative Adversarial Networks (GANs). This innovative approach not only distinguished between DFUs and normal conditions but also accurately identified other prevalent foot diseases. The hybrid model, in its best configuration, demon- strated exceptional accuracy and precision, with an average accuracy of 84% across multiple classes. These results from our experiments provide compelling evidence of the viability of machine learning and deep learning models in enhancing diagnostic accuracy for DFUs. The nuanced comparison between the binary and multi-class classification approaches reveals a significant leap in performance when employing hybrid models that leverage both the analytical strengths of ResNet50 and the gen- erative capabilities of GANs. This study carves a path forward for future research and the potential integration of these models into clinical practice, heralding a new era of tech-driven healthcare solutions.