A Survey of Methods and Architectures for Early Detection and Segmentation of Diabetic Foot Ulcer Risk Zones Using a Cycle-Consistent Adversarial Adaptation Network from Multimodal Images

Abstract

Diabetic Foot Ulcers (DFUs) are among the most severe complications of diabetes mellitus, often leading to infection, amputation, and increased healthcare costs. Early detection and accurate segmentation of DFU risk zones are crucial for timely clinical intervention and improved patient outcomes. In recent years, Artificial Intelligence (AI), particularly deep learning, has emerged as a powerful tool for automated DFU analysis. This survey provides a comprehensive review of methods and architectures for early DFU detection and segmentation, focusing on Cycle-Consistent Adversarial Adaptation Networks and multimodal imaging techniques. Deep learning models such as Convolutional Neural Networks (CNNs), U-Net variants, and Vision Transformers have demonstrated strong performance in DFU classification and segmentation tasks. Multimodal imaging, including RGB, thermal, and hyperspectral data, enhances detection accuracy by capturing both structural and physiological features. However, challenges such as domain variability, data scarcity, and lack of paired datasets limit model performance. CycleGAN-based approaches address these issues by enabling unpaired image-to-image translation and cross-domain learning. This survey highlights recent advancements (2020–2023), compares state-of-the-art architectures, and identifies key research trends and challenges. The findings suggest that hybrid and multimodal models, combined with optimization techniques, achieve superior performance compared to traditional methods. Future research should focus on lightweight models, explainable AI, and real-time deployment systems for clinical applications.