EnDFUD: Enhanced Diabetic Foot Ulcer Detection with DETR and YOLOv5

Patients with diabetes mellitus are at increased risk of complications such as diabetic foot ulcers. Early-stage ulcers, especially those with dark skin, may be difficult to see visually. Ulcers can be chronic and require detailed documentation to track the healing process. Object recognition algorithms are a promising strategy for early detection and reporting of diabetic foot ulcers, but they may struggle to detect other associated symptoms, including deformed toes, hyperkeratosis, and rhododendron. This study compared DETR and You Only Look Once (YOLOv5), two state-of-the-art object detection frameworks for diagnosing diabetic foot ulcers. We evaluate two algorithms using a newly accessible dataset of images created specifically for this purpose. Additionally, we look at ways of self-training to improve detection performance. In our tests, DETR and YOLOv5 outperformed other state-of-the-art diabetic foot ulcer detection systems. We examine the pros and cons of each framework and recommend ways to improve it in the future.