Wound Detection and Size Estimation From DFU Images

Diabetic foot ulcers (DFU) and leg ulcers are significant complications of diabetes, affecting millions of people and often requiring long-term hospitalization and presenting a high risk of limb amputation. Current clinical approaches for treating DFU and leg ulcers rely heavily on manual measurements, which are time consuming, costly, and prone to inaccuracies. This paper presents an innovative deep learning-based system for the automated discovery and size estimation of DFU and leg ulcers, aiming to address these limitations. We introduced a new dataset comprising of 2571 images of diabetic foot and leg ulcer and evaluated several neural network architectures, including FCN, U-Net, SegNet, and DeepLabV3 + (ResNet-18). Our experiment results demonstrate that DeepLabV3 + (ResNet-18) achieve the highest performance, with a mean accuracy of 98.71% and BF Score of 98.83% on the AZH dataset and 97.89% and 95.56% receptively on our YSJI_MUST dataset. Additionally, we developed a smart GUI application to facilitate easy and accurate detection and measurement of wounds, leveraging these neural networks. This system represents a significant advancement in the automated analysis of DFU and leg ulcers, offering potential improvements in clinical outcome and patient care.