A Hybrid Attention-Guided U-Net++ Architecture for Diabetic Foot Ulcer Segmentation

Diabetic Foot Ulcers (DFUs) represent a serious complication associated with diabetes and may progress to infections or limb amputation if early monitoring is not ensured. Automated analysis of clinical images plays a crucial role in enabling accurate wound assessment and timely clinical intervention. However, obtaining such segmentation results manually is time-consuming and prone to inconsistency, highlighting the need for an efficient and accurate automated approach. Correct segmentation of Diabetic Foot Ulcers (DFUs) plays an important role in wound assessment, treatment planning, and monitoring healing progress. Existing deep learning-based segmentation models, including U-Net and U-Net++, often fail to precisely delineate ulcer boundaries and effectively eliminate back-ground noise when faced with variations in wound texture, shape, and illumination. To solve these challenges, this study presented a hybrid deep learning architecture known as AG-U-Net++, which combines the dense skip connections of U-Net++ with attention gate mechanisms. The results of our experiments demonstrated that the proposed model achieved a Dice score of 0.81, an IoU score of 0.73, and an accuracy of 99.91 % which results in better performance than other architectures such as VGG16, MobileNetV2, EfficientNetB0, and ResNet34. The results ensure that the AG-U-Net++ model effectively and efficiently improves feature extraction, segmentation quality, and computational efficiency, making it suitable for real-time DFU analysis.