Diabetic Foot Ulcers (DFUs) Detection Using Deep Learning Techniques to Development of a Computer Vision System

Diabetic Foot Ulcers (DFUs) are a severe complication of diabetes, leading to infection, amputation, and increased mortality rates if not diagnosed early. This study presents an automated DFU detection system using CNN EfficientNet, integrated with Explainable AI (XAI) techniques for enhanced clinical interpretability. The proposed model is trained on the DFUC2021 dataset, achieving 98.79% classification accuracy, 96.87% F1-score, and an AUC-ROC score exceeding 0.98, significantly outperforming conventional CNN and machine learning models. The model's high specificity (99.5%) ensures precise non-ulcer classification, while 95.0% recall confirms strong ulcer sensitivity, minimizing false negatives. The segmentation accuracy, validated through Dice Score (0.890), MSE ($<0.002$), and RMSE (< 0.04), demonstrates superior feature extraction capabilities. Compared to Transformer-based ViT models (93.1% accuracy, 420 ms inference time), CNN EfficientNet maintains a lower computational cost with an inference time under 200 ms, making it suitable for real-time clinical applications. The incorporation of Grad-CAM and SHAP visualizations provides heatmaps for ulcer localization, increasing clinician trust in AI-based medical imaging. This research confirms that CNN EfficientNet-based DFU detection offers a scalable, high-accuracy AI solution for early ulcer diagnosis, supporting mobile health applications and risk assessment tools. Future work will focus on expanding datasets, refining ulcer severity classification, and optimizing deep learning models for lightweight deployment. The findings suggest that deep learning-powered DFU detection can significantly reduce amputation risks and improve diabetic patient outcomes in real-world healthcare environments.