Hybrid GhostNasNet Model with Advanced Feature Extraction for Diabetic Foot Ulcer Classification

This work aims to develop a hybrid DL approach for DFU classification. The process starts with an input DFU image, which is subjected to preprocessing. In the preprocessing phase, the DFU images are processed through image resizing and CLAHE enhancement. The preprocessed inputs are subsequently subjected to a data augmentation stage. Next, feature extraction process is carried out using edge-detection methods, such as Canny, Sobel, and the proposed approach incorporates Botox Optimization based ResNet (BO-ResNet) to extract deeper and more discriminative features from the augmented images. By integrating these methods, the system benefits from both low-level edge information and high-level learned representations, resulting in a more comprehensive feature set. BO-ResNet is developed by applying the Botox Optimization Algorithm (BOA) to train and optimize the ResNet model, enabling improved performance and more accurate feature extraction outcomes. Lastly, a novel approach, named Hybrid GhostNasNet is proposed by integrating GhostNet with NasNet for DFU classification. This approach strengthens the performance and reliability of DFU detection, supporting better diagnostic outcomes in clinical workflows.