Deep Learning and Optimization Approaches in Early Detection and segmentation of Diabetic Foot Ulcer Risk Zones Using a Cycle-Consistent Adversarial Adaptation Network from multimodal images: A Review

Abstract

Diabetic Foot Ulcers (DFU) are a major global health concern and a leading cause of non-traumatic lower-limb amputations, often due to delayed detection of tissue damage. Traditional diagnostic methods, including visual inspection and sensory testing, frequently fail to identify early physiological changes such as localized hyperthermia and deep-tissue ischemia. To overcome these limitations, recent research has focused on integrating advanced deep learning techniques such as Cycle-Consistent Adversarial Networks (CycleGANs) and multimodal image fusion for early risk detection. CycleGANs address the challenge of limited infrared thermography data by enabling the generation of pseudo-thermal images from widely available RGB data without requiring paired datasets. Additionally, transformer-based architectures and cross-attention mechanisms enhance feature fusion by combining structural and thermal information. Optimization techniques, including meta-heuristic algorithms, further improve model performance and computational efficiency, enabling real-time clinical deployment. These advanced frameworks demonstrate significantly higher accuracy and sensitivity in detecting at-risk regions compared to conventional models. Overall, such approaches shift DFU management from reactive treatment to proactive risk prediction, offering improved patient outcomes and reduced amputation rates.