A hybrid CNN and SVM framework for AI-driven clinical diagnosis and wound severity assessment

Introduction Automated wound-severity assessment has the potential to improve the consistency of image-based wound evaluation. However, its clinical applicability requires rigorous validation. This study evaluates a hybrid deep learning framework for automated wound-severity classification. Methods A retrospective technical study was conducted using wound images obtained from the AZH Wound and Vascular Center. The proposed framework integrates three stages: YOLOv3 for wound region-of-interest (ROI) localization, AlexNet for deep feature extraction, and a support vector machine (SVM) for classification into three severity categories: green, yellow, and red. The proposed DCNN-SVM framework was evaluated against four transfer-learning baselines using the same held-out test partition. Results The proposed DCNN-SVM framework achieved an accuracy of 80.67%, recall of 80.80%, precision of 80.75%, and an F1-score of 80.90%, outperforming all four evaluated transfer-learning baselines on the held-out test set. Discussion The findings demonstrate the feasibility of integrating YOLOv3-based wound localization with AlexNet-derived deep features and SVM classification for automated wound-severity assessment. The primary contribution lies in evaluating this integrated framework rather than introducing novel individual algorithms. Nevertheless, the small, single-center dataset and the absence of external or prospective clinical validation limit the generalizability of the findings. Therefore, the results should be considered preliminary proof-of-concept evidence rather than confirmation of clinical readiness.