Deep Learning Strategies for Diabetic Foot Ulcer Diagnosis: An Empirical Comparison of Transformer and CNN Architectures

¹ Department of ECE St. Joseph’s Institute of Technology Chennai, Tamil Nadu, India This email address is being protected from spambots. You need JavaScript enabled to view it.
² Thiriloksha S Department of ECE St. Joseph’s Institute of Technology Chennai, Tamil Nadu, India This email address is being protected from spambots. You need JavaScript enabled to view it.
³ Tephillah S Department of ECE St. Joseph’s Institute of Technology Chennai, Tamil Nadu, India This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Diabetic Foot Ulcers (DFU) are among the most serious complications faced by diabetic patients. What usually starts as a small injury can worsen and in many cases lead to lower limb amputation. However with modern technologies like deep learning early screening and identification of DFU has become more achievable. This helps doctors and make their job easy. This study presents a practical comparison between ResNet18 a CNN and DeiT-Small a Vision Transformer. Both these models were tuned using transfer learning on datasets containing images of both DFU affected feet and healthy feet. There was a clear performance gap between both these models. DeiT-Small reached an impressive test accuracy of 99.17% while ResNet18 achieved 85.40% accuracy. To understand their performance better visualization tools like Grad-CAM and transformer attention maps were used. For future work, the study aims to create more consistent data splits, perform deeper architectural ablation and integrate IOT based sensors to support real time DFU monitoring.