Detection of Hypergranulation Tissue in Chronic Wound Images Using Artificial Intelligence Algorithms

Abstract

Hypergranulation in chronic wounds reflects impaired healing, leading to delayed recovery, increased risk of infection and higher treatment costs for healthcare systems. Despite its impact, hypergranulation is often misidentified in the early stages, hindering timely intervention. This study presents a deep learning-based method to distinguish between hypergranulated and non-hypergranulated tissue. The dataset comprised 6235 wound images from Hospital de la Santa Creu de Vic (Catalonia, Spain) and DFUC 2022, with balanced classes. Five architectures were evaluated using transfer learning: ViT, VGG16, RepVGG, MobileViT and RepGhost. RepGhost achieved the best results (81.4% accuracy, 89.4% area under the receiver operating characteristic curve [AUC]), outperforming both CNNs and transformers. Lightweight models (RepGhost, MobileViT) also demonstrated high performance, making them suitable for implementation on mobile devices. This tool may assist clinicians in the early detection of hypergranulation and improve wound treatment outcomes. This appears to be the first deep learning study on this condition to utilise a large, balanced dataset.