A Hybrid Model for Accurate Detection of Diabetic Foot Ulcer Using Augmented Clinical Imagery

ABSTRACT One of the most common complications of diabetes mellitus is diabetic foot ulcers (DFUs), which are marked by a high death rate and a possible danger of lower limb amputation. Since DFU causes health problems if left untreated, early detection and accurate classification of DFUs are essential for effective treatment and improved patient outcomes. The research proposed three feature fusion–based hybrid model architectures for detecting DFUs from the DFU image dataset. These hybrid models were applied to a dataset named the DFU dataset, which consists of 4431 foot images comprising both normal and DFU images. First, the features extracted by the NASNetMobile and VGG16 models from DFU images were fused into one feature vector, and the sigmoid function classified the normal and DFU images with 98.11% accuracy, 99.00% sensitivity, 97.30% specificity and 97.06% precision. Second, the VGG16 and DenseNet121 models extract features from DFU images individually and fuse features into one feature vector. The sigmoid function classifies the DFU images with 99.05% accuracy. During the same procedure, the fused (DenseNet121 and NASNetMobile) hybrid model achieved 99.52% accuracy, 99.09% specificity, 100% sensitivity and 99.02% precision for DFU classification purposes, whose performance is higher than the other two hybrid models. The primary novelty of this study is the development and validation of a fused DenseNet121 + NASNetMobile hybrid model, which demonstrates superior reliability and efficiency for DFU detection. Additionally, we apply two validation techniques to the DFUC2021 DFU dataset, which has demonstrated strong and consistent model performance in qualitative analysis. The proposed model shows strong potential for detecting DFUs but requires further validation through clinical trials and real‐world deployment before routine healthcare use.