An Improved Vision Transformer for Early Detection of Diabetic Foot Using Thermogram

Diabetic foot is a complication of diabetes mellitus caused by prolonged hyperglycemia. It can lead to serious consequences such as ulceration, infection, and even amputation. Early detection and treatment are essential to reduce the risks of these complications and improve the quality of life of patients. Several studies have demonstrated the potential of combining foot thermogram with deep learning for early detection of diabetic foot. Based on these insights and studies, our study focused on applying state-of-the-art networks in the field of image classification, vision transformers, to this task. In this study, A learnable block was added to the original vision transformer, effectively improving the performance of the original vision transformer. Experimental results showed that in this task, compared to the vision transformers with a maximum accuracy of 90% and convolutional neural networks with a maximum accuracy of 100%, the proposed vision transformer can achieve an accuracy of 99%, demonstrating the potential of vision transformers for application on small-size medical image datasets.