Revolutionizing Diabetic Foot Ulcer Detection with Hybrid CNN-SVM Models: An Accurate and Scalable Approach

A major complication of diabetes, Diabetic foot ulcers (DFUs) are recognized by significant health outcomes when detected early. In this work, we propose the novel hybrid CNN-SVM model for accurate and scalable DFU detection that combines feature extraction strengths of deep learning and reliable classification power of machine learning. We evaluate the proposed model on the publicly available DFUC2021 dataset of 5000 annotated DFU images and that are split into 60%, 20% and 20% training, validation and testing set respectively. Or, this hybrid approach demonstrated a stunning accuracy of 94.5% which surpassed both the standalone CNN model with 91.2% specificity, and the traditional SVM classifier with handcrafted features at 82.5%. In addition, it exhibits higher precision (93.2%), recall (93.8%), F1 score (93.5%) and on an AUC - ROC of 0.98, attesting to its ability to discriminate between DFU and non - DFU cases better. To implemented the CNN component, based on the ResNet-50 architecture, for extracting deep hierarchical features and the SVM classifier with RBF kernel for generating robust decision boundaries thereby reducing over fitting and improving generalization. Experimental results, along with confusion matrix analysis and ROC curve evaluation, show that the model has potential to be applied in clinic, achieving significant reduction of false positives and false negatives. Proposed hybrid approach to detect DFU proposes a scalable and efficient solution for DFU detection that can be deployed for real world healthcare and telemedicine applications. Further work will investigate merging explainable Al techniques to add interpretability to the model and extending the model for multi class DFU classification.