AI models for the early detection of diabetic feet

• Main Author: Lim, Qian Hui
• Other Authors: Xia Kelin
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Mathematical Sciences
• Online Access: https://hdl.handle.net/10356/175611
• Institution: Nanyang Technological University

Diabetic foot ulcers (DFUs), characterised by wounds or open sores on the feet, are common implications of diabetes, resulting from impaired wound healing and increased susceptibility to infection. This can lead to amputations in severe cases. With the advancement of artificial intelligence and machine learning frameworks, computer-aided diagnosis has allowed for more accurate and efficient diagnosis of DFUs. Consequently, this has led to the development of newer computer vision models such as Vision Transformers (ViT) and Vision Graph Neural Networks (ViG). In this study, we explore the newly developed ViT and ViG models in addition to the classic convolutional neural network models (CNNs) for the classification of healthy and ulcerated diabetic foot images. The proposed approach compares three pre-trained models: DenseNet121, ViTB/16, and ViG-Ti to extract features and classify the foot sample images. Experimental results demonstrate the effectiveness of such newly developed models in DFU image classification, with ViT-B/16 achieving the best results of 99.05% accuracy, 99.02% precision, 99.02% recall and 99.02% F1-score. The findings of this study hold meaningful implications for clinical practice, offering great potential in the use of ViT and ViG frameworks for automated early diagnosis of DFUs.