Automated Classification of Diabetic Ulcers and Ordinary Wounds Using Feautre Extraction from DenseNet201 and YOLOv7 for Medical Diagnosis

Diabetic ulcers are a major global health issue. They cause over eighty percent of lower limb amputations worldwide. Nearly 1 in 4 diabetic patients develop these wounds. It leads to long hospital stays and high mortality. Early detection is important to stop complications. Current methods use visual inspection and manual checking. These are slow, subjective, and error-prone. Deep learning and computer based vision changed medical imaging. Automated tools for ulcer detection are now possible. Convolutional Neural Networks, Faster R-CNN, and EfficientNet have been used. But problems exist like high computation, false positives, and slow processing. YOLOv7 improved speed and accuracy in object detection. This study used YOLOv7 to classify ulcers and wounds. $\mathbf{1, 1 4 0}$ labeled images were used. The dataset had both ulcer and nonulcer wounds. Preprocessing included augmentation, contrast changes, and resizing. 80 % was for training, 20 % for validation. Accuracy, precision, recall, F1- score, and mAP were used for evaluation. The model got $\mathbf{9 2. 5 \%}$ accuracy, better than Faster R-CNN at $\mathbf{8 8. 7 \%}$ and EfficientNet at $\mathbf{8 9. 4 \%}$. It detected ulcer boundaries, severity, and features well. Misdiagnosis dropped by 30 %. This can improve early treatment. Future work will add thermal and infrared images. The model will be tested on mobile AI for healthcare. This study gave a strong step for AI-based ulcer detection. It improved speed and accuracy in diagnosis.